├── .gitattributes ├── .gitignore ├── KPN.py ├── README.md ├── data_generation ├── __init__.py ├── ahd_demosaicking.py ├── constants.py ├── data_utils.py ├── denoise_wavelet.py ├── generate_dataset.py ├── image_io.py ├── image_processing.py ├── kernel.py └── pipeline.py ├── data_provider.py ├── dataset_specs ├── data_configspec.conf └── full_dataset.conf ├── eval_images ├── 10_gt.png ├── 10_kpn_5x5_38.68dB_0.9798.png ├── 10_noisy_34.50dB_0.8800.png ├── 11_gt.png ├── 11_kpn_5x5_38.84dB_0.9909.png ├── 11_noisy_34.61dB_0.8585.png ├── 12_gt.png ├── 12_kpn_5x5_38.04dB_0.9860.png ├── 12_noisy_34.56dB_0.8722.png ├── 13_gt.png ├── 13_kpn_5x5_38.88dB_0.9890.png ├── 13_noisy_34.48dB_0.8867.png ├── 14_gt.png ├── 14_kpn_5x5_37.04dB_0.9497.png ├── 14_noisy_34.69dB_0.8965.png ├── 15_gt.png ├── 15_kpn_5x5_38.31dB_0.9893.png ├── 15_noisy_34.50dB_0.9219.png ├── 16_gt.png ├── 16_kpn_5x5_37.81dB_0.9653.png ├── 16_noisy_34.38dB_0.8607.png ├── 17_gt.png ├── 17_kpn_5x5_34.16dB_0.9625.png ├── 17_noisy_28.32dB_0.7489.png ├── 18_gt.png ├── 18_kpn_5x5_36.83dB_0.9698.png ├── 18_noisy_28.09dB_0.6567.png ├── 19_gt.png ├── 19_kpn_5x5_39.11dB_0.9738.png ├── 19_noisy_28.13dB_0.5510.png ├── 1_gt.png ├── 1_kpn.png ├── 1_kpn_5x5_39.35dB_0.9866.png ├── 1_noisy.png ├── 1_noisy_36.73dB_0.9366.png ├── 20_gt.png ├── 20_kpn_5x5_33.79dB_0.9656.png ├── 20_noisy_28.72dB_0.7307.png ├── 21_gt.png ├── 21_kpn.png ├── 21_kpn_5x5_37.36dB_0.9316.png ├── 21_noisy.png ├── 21_noisy_27.61dB_0.5016.png ├── 22_gt.png ├── 22_kpn.png ├── 22_kpn_5x5_33.98dB_0.9473.png ├── 22_noisy.png ├── 22_noisy_28.10dB_0.7991.png ├── 23_gt.png ├── 23_kpn.png ├── 23_kpn_5x5_35.85dB_0.9759.png ├── 23_noisy.png ├── 23_noisy_28.39dB_0.6966.png ├── 24_gt.png ├── 24_kpn.png ├── 24_kpn_5x5_34.84dB_0.9692.png ├── 24_noisy.png ├── 24_noisy_28.60dB_0.6402.png ├── 25_gt.png ├── 25_kpn_5x5_40.11dB_0.9853.png ├── 25_noisy_29.38dB_0.5669.png ├── 26_gt.png ├── 26_kpn_5x5_32.85dB_0.9452.png ├── 26_noisy_26.66dB_0.7940.png ├── 27_gt.png ├── 27_kpn_5x5_35.52dB_0.9546.png ├── 27_noisy_27.42dB_0.7241.png ├── 28_gt.png ├── 28_kpn_5x5_36.13dB_0.9708.png ├── 28_noisy_28.70dB_0.6872.png ├── 29_gt.png ├── 29_kpn_5x5_35.68dB_0.9565.png ├── 29_noisy_27.39dB_0.7316.png ├── 2_gt.png ├── 2_kpn.png ├── 2_kpn_5x5_39.48dB_0.9919.png ├── 2_noisy.png ├── 2_noisy_37.73dB_0.9347.png ├── 30_gt.png ├── 30_kpn_5x5_37.83dB_0.9713.png ├── 30_noisy_26.21dB_0.6279.png ├── 31_gt.png ├── 31_kpn_5x5_37.32dB_0.9682.png ├── 31_noisy_28.10dB_0.6873.png ├── 32_gt.png ├── 32_kpn_5x5_37.64dB_0.9700.png ├── 32_noisy_32.31dB_0.7877.png ├── 33_gt.png ├── 33_kpn_5x5_39.45dB_0.9817.png ├── 33_noisy_32.41dB_0.7184.png ├── 34_gt.png ├── 34_kpn_5x5_38.20dB_0.9803.png ├── 34_noisy_32.51dB_0.7680.png ├── 35_gt.png ├── 35_kpn_5x5_39.29dB_0.9842.png ├── 35_noisy_32.38dB_0.8153.png ├── 36_gt.png ├── 36_kpn_5x5_37.80dB_0.9754.png ├── 36_noisy_32.38dB_0.8101.png ├── 37_gt.png ├── 37_kpn_5x5_37.77dB_0.9826.png ├── 37_noisy_32.21dB_0.8376.png ├── 38_gt.png ├── 38_kpn_5x5_36.79dB_0.9648.png ├── 38_noisy_32.33dB_0.8203.png ├── 39_gt.png ├── 39_kpn_5x5_36.01dB_0.9710.png ├── 39_noisy_27.63dB_0.7201.png ├── 3_gt.png ├── 3_kpn.png ├── 3_kpn_5x5_39.14dB_0.9938.png ├── 3_noisy.png ├── 3_noisy_37.67dB_0.9583.png ├── 40_gt.png ├── 40_kpn_5x5_38.57dB_0.9918.png ├── 40_noisy_36.00dB_0.9021.png ├── 41_gt.png ├── 41_kpn_5x5_38.56dB_0.9867.png ├── 41_noisy_36.12dB_0.9290.png ├── 42_gt.png ├── 42_kpn_5x5_39.17dB_0.9878.png ├── 42_noisy_36.09dB_0.9311.png ├── 43_gt.png ├── 43_kpn_5x5_39.89dB_0.9728.png ├── 43_noisy_36.18dB_0.8960.png ├── 44_gt.png ├── 44_kpn_5x5_37.81dB_0.9755.png ├── 44_noisy_36.06dB_0.9542.png ├── 45_gt.png ├── 45_kpn_5x5_37.07dB_0.9824.png ├── 45_noisy_36.21dB_0.9775.png ├── 46_gt.png ├── 46_kpn_5x5_41.12dB_0.9823.png ├── 46_noisy_31.75dB_0.7773.png ├── 47_gt.png ├── 47_kpn_5x5_38.98dB_0.9849.png ├── 47_noisy_35.49dB_0.9071.png ├── 48_gt.png ├── 48_kpn_5x5_37.68dB_0.9244.png ├── 48_noisy_30.25dB_0.7162.png ├── 49_gt.png ├── 49_kpn_5x5_30.79dB_0.7580.png ├── 49_noisy_26.10dB_0.5892.png ├── 4_gt.png ├── 4_kpn.png ├── 4_kpn_5x5_39.25dB_0.9824.png ├── 4_noisy.png ├── 4_noisy_37.99dB_0.9598.png ├── 50_gt.png ├── 50_kpn_5x5_39.02dB_0.9887.png ├── 50_noisy_36.48dB_0.9212.png ├── 51_gt.png ├── 51_kpn_5x5_39.68dB_0.9903.png ├── 51_noisy_36.14dB_0.9540.png ├── 52_gt.png ├── 52_kpn_5x5_39.06dB_0.9819.png ├── 52_noisy_36.34dB_0.9172.png ├── 53_gt.png ├── 53_kpn_5x5_39.40dB_0.9871.png ├── 53_noisy_34.24dB_0.9062.png ├── 54_gt.png ├── 54_kpn_5x5_39.98dB_0.9886.png ├── 54_noisy_36.25dB_0.9036.png ├── 55_gt.png ├── 55_kpn_5x5_38.73dB_0.9822.png ├── 55_noisy_37.04dB_0.9654.png ├── 56_gt.png ├── 56_kpn_5x5_39.51dB_0.9936.png ├── 56_noisy_38.16dB_0.9714.png ├── 57_gt.png ├── 57_kpn_5x5_37.89dB_0.9887.png ├── 57_noisy_35.77dB_0.9747.png ├── 58_gt.png ├── 58_kpn_5x5_39.52dB_0.9840.png ├── 58_noisy_38.25dB_0.9738.png ├── 59_gt.png ├── 59_kpn_5x5_38.02dB_0.9827.png ├── 59_noisy_33.73dB_0.9389.png ├── 5_gt.png ├── 5_kpn.png ├── 5_kpn_5x5_39.33dB_0.9888.png ├── 5_noisy.png ├── 5_noisy_37.77dB_0.9749.png ├── 60_gt.png ├── 60_kpn_5x5_38.65dB_0.9853.png ├── 60_noisy_37.21dB_0.9718.png ├── 61_gt.png ├── 61_kpn_5x5_40.22dB_0.9922.png ├── 61_noisy_37.93dB_0.9604.png ├── 62_gt.png ├── 62_kpn_5x5_36.89dB_0.9901.png ├── 62_noisy_34.35dB_0.9645.png ├── 63_gt.png ├── 63_kpn_5x5_34.38dB_0.9104.png ├── 63_noisy_25.35dB_0.4702.png ├── 64_gt.png ├── 64_kpn_5x5_35.10dB_0.9525.png ├── 64_noisy_25.87dB_0.5550.png ├── 65_gt.png ├── 65_kpn_5x5_34.76dB_0.9458.png ├── 65_noisy_25.34dB_0.5893.png ├── 66_gt.png ├── 66_kpn_5x5_34.08dB_0.8902.png ├── 66_noisy_25.16dB_0.5120.png ├── 67_gt.png ├── 67_kpn_5x5_39.37dB_0.9908.png ├── 67_noisy_38.30dB_0.9847.png ├── 68_gt.png ├── 68_kpn_5x5_35.75dB_0.9324.png ├── 68_noisy_25.69dB_0.4059.png ├── 69_gt.png ├── 69_kpn_5x5_32.32dB_0.9361.png ├── 69_noisy_25.68dB_0.7298.png ├── 6_gt.png ├── 6_kpn_5x5_38.36dB_0.9825.png ├── 6_noisy_36.65dB_0.9624.png ├── 70_gt.png ├── 70_kpn_5x5_39.91dB_0.9910.png ├── 70_noisy_38.50dB_0.9572.png ├── 71_gt.png ├── 71_kpn_5x5_35.55dB_0.9774.png ├── 71_noisy_32.97dB_0.8975.png ├── 72_gt.png ├── 72_kpn_5x5_37.08dB_0.9755.png ├── 72_noisy_32.98dB_0.8527.png ├── 73_gt.png ├── 73_kpn_5x5_39.25dB_0.9713.png ├── 73_noisy_31.69dB_0.7565.png ├── 74_gt.png ├── 74_kpn_5x5_37.40dB_0.9598.png ├── 74_noisy_31.40dB_0.7913.png ├── 75_gt.png ├── 75_kpn_5x5_34.89dB_0.9287.png ├── 75_noisy_25.27dB_0.6376.png ├── 76_gt.png ├── 76_kpn_5x5_38.18dB_0.9756.png ├── 76_noisy_32.34dB_0.8855.png ├── 77_gt.png ├── 77_kpn_5x5_36.31dB_0.9730.png ├── 77_noisy_32.73dB_0.9155.png ├── 78_gt.png ├── 78_kpn_5x5_38.33dB_0.9751.png ├── 78_noisy_25.95dB_0.4840.png ├── 79_gt.png ├── 79_kpn_5x5_37.68dB_0.9554.png ├── 79_noisy_32.21dB_0.7821.png ├── 7_gt.png ├── 7_kpn_5x5_40.65dB_0.9868.png ├── 7_noisy_37.47dB_0.9308.png ├── 80_gt.png ├── 80_kpn_5x5_36.39dB_0.9837.png ├── 80_noisy_32.55dB_0.9169.png ├── 81_gt.png ├── 81_kpn_5x5_37.33dB_0.9816.png ├── 81_noisy_32.67dB_0.8759.png ├── 82_gt.png ├── 82_kpn_5x5_36.23dB_0.9788.png ├── 82_noisy_32.49dB_0.8655.png ├── 83_gt.png ├── 83_kpn_5x5_36.86dB_0.9658.png ├── 83_noisy_31.89dB_0.8717.png ├── 84_gt.png ├── 84_kpn_5x5_37.53dB_0.9908.png ├── 84_noisy_32.70dB_0.8561.png ├── 85_gt.png ├── 85_kpn_5x5_37.93dB_0.9557.png ├── 85_noisy_32.45dB_0.7962.png ├── 86_gt.png ├── 86_kpn_5x5_34.20dB_0.9634.png ├── 86_noisy_30.81dB_0.9148.png ├── 87_gt.png ├── 87_kpn_5x5_39.10dB_0.9747.png ├── 87_noisy_30.31dB_0.5805.png ├── 88_gt.png ├── 88_kpn_5x5_37.34dB_0.9650.png ├── 88_noisy_29.50dB_0.6900.png ├── 89_gt.png ├── 89_kpn_5x5_36.06dB_0.9816.png ├── 89_noisy_30.16dB_0.7799.png ├── 8_gt.png ├── 8_kpn_5x5_38.48dB_0.9800.png ├── 8_noisy_35.23dB_0.9459.png ├── 90_gt.png ├── 90_kpn_5x5_34.73dB_0.9713.png ├── 90_noisy_29.64dB_0.8340.png ├── 91_gt.png ├── 91_kpn_5x5_39.03dB_0.9762.png ├── 91_noisy_32.42dB_0.7615.png ├── 92_gt.png ├── 92_kpn_5x5_38.21dB_0.9746.png ├── 92_noisy_30.11dB_0.7182.png ├── 93_gt.png ├── 93_kpn_5x5_36.61dB_0.9652.png ├── 93_noisy_29.85dB_0.6761.png ├── 94_gt.png ├── 94_kpn_5x5_37.38dB_0.9859.png ├── 94_noisy_32.82dB_0.8374.png ├── 95_gt.png ├── 95_kpn_5x5_37.35dB_0.9740.png ├── 95_noisy_32.39dB_0.8629.png ├── 96_gt.png ├── 96_kpn_5x5_35.64dB_0.9681.png ├── 96_noisy_29.94dB_0.7946.png ├── 97_gt.png ├── 97_kpn_5x5_36.06dB_0.9525.png ├── 97_noisy_32.48dB_0.8883.png ├── 98_gt.png ├── 98_kpn_5x5_36.41dB_0.9654.png ├── 98_noisy_31.56dB_0.8672.png ├── 99_gt.png ├── 99_kpn_5x5_38.40dB_0.9825.png ├── 99_noisy_32.58dB_0.8636.png ├── 9_gt.png ├── 9_kpn_5x5_37.53dB_0.9877.png ├── 9_noisy_34.32dB_0.8747.png └── thumbs.db ├── eval_images_RGB ├── 0_gt.png ├── 0_noisy_29.17dB_0.911.png ├── 0_pred_31.86dB_0.957.png ├── 10_gt.png ├── 10_noisy_31.95dB_0.776.png ├── 10_pred_39.27dB_0.965.png ├── 11_gt.png ├── 11_noisy_23.93dB_0.549.png ├── 11_pred_29.44dB_0.801.png ├── 12_gt.png ├── 12_noisy_18.70dB_0.308.png ├── 12_pred_34.02dB_0.954.png ├── 13_gt.png ├── 13_noisy_32.41dB_0.916.png ├── 13_pred_37.87dB_0.987.png ├── 1_gt.png ├── 1_noisy_22.25dB_0.544.png ├── 1_pred_32.77dB_0.912.png ├── 2_gt.png ├── 2_noisy_19.42dB_0.496.png ├── 2_pred_29.73dB_0.900.png ├── 3_gt.png ├── 3_noisy_25.74dB_0.701.png ├── 3_pred_35.85dB_0.960.png ├── 4_gt.png ├── 4_noisy_19.34dB_0.595.png ├── 4_pred_29.69dB_0.937.png ├── 5_gt.png ├── 5_noisy_27.50dB_0.641.png ├── 5_pred_33.03dB_0.840.png ├── 6_gt.png ├── 6_noisy_16.37dB_0.283.png ├── 6_pred_29.25dB_0.844.png ├── 7_gt.png ├── 7_noisy_27.71dB_0.634.png ├── 7_pred_38.33dB_0.961.png ├── 8_gt.png ├── 8_noisy_30.11dB_0.870.png ├── 8_pred_32.60dB_0.942.png ├── 9_gt.png ├── 9_noisy_24.02dB_0.597.png ├── 9_pred_33.24dB_0.914.png └── thumbs.db ├── kpn_data_provider.py ├── kpn_specs ├── configspec.conf └── kpn_config.conf ├── models └── kpn │ └── checkpoint │ └── README.md ├── train_eval_syn.py └── utils ├── __init__.py ├── image_utils.py └── training_util.py /.gitattributes: -------------------------------------------------------------------------------- 1 | models/kpn/checkpoint/model_best.pth.tar filter=lfs diff=lfs merge=lfs -text 2 | -------------------------------------------------------------------------------- /.gitignore: -------------------------------------------------------------------------------- 1 | # Byte-compiled / optimized / DLL files 2 | __pycache__/ 3 | *.py[cod] 4 | *$py.class 5 | 6 | # C extensions 7 | *.so 8 | 9 | # Distribution / packaging 10 | .Python 11 | build/ 12 | develop-eggs/ 13 | dist/ 14 | downloads/ 15 | eggs/ 16 | .eggs/ 17 | lib/ 18 | lib64/ 19 | parts/ 20 | sdist/ 21 | var/ 22 | wheels/ 23 | share/python-wheels/ 24 | *.egg-info/ 25 | .installed.cfg 26 | *.egg 27 | MANIFEST 28 | .idea 29 | 30 | # PyInstaller 31 | # Usually these files are written by a python script from a template 32 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 33 | *.manifest 34 | *.spec 35 | 36 | # Installer logs 37 | pip-log.txt 38 | pip-delete-this-directory.txt 39 | 40 | # Unit test / coverage reports 41 | htmlcov/ 42 | .tox/ 43 | .nox/ 44 | .coverage 45 | .coverage.* 46 | .cache 47 | nosetests.xml 48 | coverage.xml 49 | *.cover 50 | *.py,cover 51 | .hypothesis/ 52 | .pytest_cache/ 53 | cover/ 54 | 55 | # Translations 56 | *.mo 57 | *.pot 58 | 59 | # Django stuff: 60 | *.log 61 | local_settings.py 62 | db.sqlite3 63 | db.sqlite3-journal 64 | 65 | # Flask stuff: 66 | instance/ 67 | .webassets-cache 68 | 69 | # Scrapy stuff: 70 | .scrapy 71 | 72 | # Sphinx documentation 73 | docs/_build/ 74 | 75 | # PyBuilder 76 | .pybuilder/ 77 | target/ 78 | 79 | # Jupyter Notebook 80 | .ipynb_checkpoints 81 | 82 | # IPython 83 | profile_default/ 84 | ipython_config.py 85 | 86 | # pyenv 87 | # For a library or package, you might want to ignore these files since the code is 88 | # intended to run in multiple environments; otherwise, check them in: 89 | # .python-version 90 | 91 | # pipenv 92 | # According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control. 93 | # However, in case of collaboration, if having platform-specific dependencies or dependencies 94 | # having no cross-platform support, pipenv may install dependencies that don't work, or not 95 | # install all needed dependencies. 96 | #Pipfile.lock 97 | 98 | # poetry 99 | # Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control. 100 | # This is especially recommended for binary packages to ensure reproducibility, and is more 101 | # commonly ignored for libraries. 102 | # https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control 103 | #poetry.lock 104 | 105 | # PEP 582; used by e.g. github.com/David-OConnor/pyflow 106 | __pypackages__/ 107 | 108 | # Celery stuff 109 | celerybeat-schedule 110 | celerybeat.pid 111 | 112 | # SageMath parsed files 113 | *.sage.py 114 | 115 | # Environments 116 | .env 117 | .venv 118 | env/ 119 | venv/ 120 | ENV/ 121 | env.bak/ 122 | venv.bak/ 123 | 124 | # Spyder project settings 125 | .spyderproject 126 | .spyproject 127 | 128 | # Rope project settings 129 | .ropeproject 130 | 131 | # mkdocs documentation 132 | /site 133 | 134 | # mypy 135 | .mypy_cache/ 136 | .dmypy.json 137 | dmypy.json 138 | 139 | # Pyre type checker 140 | .pyre/ 141 | 142 | # pytype static type analyzer 143 | .pytype/ 144 | 145 | # Cython debug symbols 146 | cython_debug/ 147 | 148 | # PyCharm 149 | # JetBrains specific template is maintainted in a separate JetBrains.gitignore that can 150 | # be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore 151 | # and can be added to the global gitignore or merged into this file. For a more nuclear 152 | # option (not recommended) you can uncomment the following to ignore the entire idea folder. 153 | #.idea/ 154 | -------------------------------------------------------------------------------- /KPN.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | import numpy as np 4 | import torch.nn.functional as F 5 | from torchsummary import summary 6 | import torchvision.models as models 7 | 8 | 9 | # KPN基本网路单元 10 | class Basic(nn.Module): 11 | def __init__(self, in_ch, out_ch, g=16, channel_att=False, 12 | spatial_att=False): 13 | super(Basic, self).__init__() 14 | self.channel_att = channel_att 15 | self.spatial_att = spatial_att 16 | self.conv1 = nn.Sequential( 17 | nn.Conv2d(in_channels=in_ch, out_channels=out_ch, kernel_size=3, 18 | stride=1, padding=1), 19 | # nn.BatchNorm2d(out_ch), 20 | nn.ReLU(), 21 | nn.Conv2d(in_channels=out_ch, out_channels=out_ch, kernel_size=3, 22 | stride=1, padding=1), 23 | # nn.BatchNorm2d(out_ch), 24 | nn.ReLU(), 25 | nn.Conv2d(in_channels=out_ch, out_channels=out_ch, kernel_size=3, 26 | stride=1, padding=1), 27 | # nn.BatchNorm2d(out_ch), 28 | nn.ReLU() 29 | ) 30 | 31 | if channel_att: 32 | self.att_c = nn.Sequential( 33 | nn.Conv2d(2 * out_ch, out_ch // g, 1, 1, 0), 34 | nn.ReLU(), 35 | nn.Conv2d(out_ch // g, out_ch, 1, 1, 0), 36 | nn.Sigmoid() 37 | ) 38 | if spatial_att: 39 | self.att_s = nn.Sequential( 40 | nn.Conv2d(in_channels=2, out_channels=1, kernel_size=7, 41 | stride=1, padding=3), 42 | nn.Sigmoid() 43 | ) 44 | 45 | def forward(self, data): 46 | """ 47 | Forward function. 48 | :param data: 49 | :return: tensor 50 | """ 51 | fm = self.conv1(data) 52 | if self.channel_att: 53 | # fm_pool = F.adaptive_avg_pool2d(fm, (1, 1)) + F.adaptive_max_pool2d(fm, (1, 1)) 54 | fm_pool = torch.cat([F.adaptive_avg_pool2d(fm, (1, 1)), 55 | F.adaptive_max_pool2d(fm, (1, 1))], dim=1) 56 | att = self.att_c(fm_pool) 57 | fm = fm * att 58 | if self.spatial_att: 59 | fm_pool = torch.cat([torch.mean(fm, dim=1, keepdim=True), 60 | torch.max(fm, dim=1, keepdim=True)[0]], dim=1) 61 | att = self.att_s(fm_pool) 62 | fm = fm * att 63 | return fm 64 | 65 | 66 | class KPN(nn.Module): 67 | def __init__(self, color=True, burst_length=8, blind_est=False, 68 | kernel_size=[5], sep_conv=False, 69 | channel_att=False, spatial_att=False, upMode='bilinear', 70 | core_bias=False): 71 | super(KPN, self).__init__() 72 | self.upMode = upMode 73 | self.burst_length = burst_length 74 | self.core_bias = core_bias 75 | self.color_channel = 3 if color else 1 76 | in_channel = (3 if color else 1) * ( 77 | burst_length if blind_est else burst_length + 1) 78 | out_channel = (3 if color else 1) * ( 79 | 2 * sum(kernel_size) if sep_conv else np.sum( 80 | np.array(kernel_size) ** 2)) * burst_length 81 | if core_bias: 82 | out_channel += (3 if color else 1) * burst_length 83 | # 各个卷积层定义 84 | # 2~5层都是均值池化+3层卷积 85 | self.conv1 = Basic(in_channel, 64, channel_att=False, spatial_att=False) 86 | self.conv2 = Basic(64, 128, channel_att=False, spatial_att=False) 87 | self.conv3 = Basic(128, 256, channel_att=False, spatial_att=False) 88 | self.conv4 = Basic(256, 512, channel_att=False, spatial_att=False) 89 | self.conv5 = Basic(512, 512, channel_att=False, spatial_att=False) 90 | # 6~8层要先上采样再卷积 91 | self.conv6 = Basic(512 + 512, 512, channel_att=channel_att, 92 | spatial_att=spatial_att) 93 | self.conv7 = Basic(256 + 512, 256, channel_att=channel_att, 94 | spatial_att=spatial_att) 95 | self.conv8 = Basic(256 + 128, out_channel, channel_att=channel_att, 96 | spatial_att=spatial_att) 97 | self.outc = nn.Conv2d(out_channel, out_channel, 1, 1, 0) 98 | 99 | self.kernel_pred = KernelConv(kernel_size, sep_conv, self.core_bias) 100 | 101 | self.apply(self._init_weights) 102 | 103 | @staticmethod 104 | def _init_weights(m): 105 | if isinstance(m, nn.Conv2d): 106 | nn.init.xavier_normal_(m.weight.data) 107 | nn.init.constant_(m.bias.data, 0.0) 108 | elif isinstance(m, nn.Linear): 109 | nn.init.xavier_normal_(m.weight.data) 110 | nn.init.constant_(m.bias.data, 0.0) 111 | 112 | @staticmethod 113 | def pad_before_cat(x1, x2): 114 | """Prevent the failure of skip concatenation between the corresponding 115 | layers of the encoder and decoder due to the automatic edge trimming of 116 | nn.Conv2d when the input image size is odd.""" 117 | diffY = x1.size()[-2] - x2.size()[-2] 118 | diffX = x1.size()[-1] - x2.size()[-1] 119 | x2 = F.pad(x2, [diffX // 2, diffX - diffX // 2, 120 | diffY // 2, diffY - diffY // 2]) 121 | return x2 122 | 123 | # 前向传播函数 124 | def forward(self, data_with_est, data, white_level=1.0): 125 | """ 126 | forward and obtain pred image directly 127 | :param data_with_est: if not blind estimation, it is same as data 128 | :param data: 129 | :return: pred_img_i and img_pred 130 | """ 131 | conv1 = self.conv1(data_with_est) 132 | conv2 = self.conv2(F.avg_pool2d(conv1, kernel_size=2, stride=2)) 133 | conv3 = self.conv3(F.avg_pool2d(conv2, kernel_size=2, stride=2)) 134 | conv4 = self.conv4(F.avg_pool2d(conv3, kernel_size=2, stride=2)) 135 | conv5 = self.conv5(F.avg_pool2d(conv4, kernel_size=2, stride=2)) 136 | # 开始上采样 同时要进行skip connection 137 | # Padding before skip concatenation is needed, only works if the input 138 | # image size is odd. 139 | conv6 = self.conv6(torch.cat([conv4, self.pad_before_cat(conv4, 140 | F.interpolate( 141 | conv5, 142 | scale_factor=2, 143 | mode=self.upMode))], 144 | dim=1)) 145 | conv7 = self.conv7(torch.cat([conv3, self.pad_before_cat(conv3, 146 | F.interpolate( 147 | conv6, 148 | scale_factor=2, 149 | mode=self.upMode))], 150 | dim=1)) 151 | conv8 = self.conv8(torch.cat([conv2, self.pad_before_cat(conv2, 152 | F.interpolate( 153 | conv7, 154 | scale_factor=2, 155 | mode=self.upMode))], 156 | dim=1)) 157 | 158 | # return channel K*K*N 159 | core = self.outc(F.interpolate(conv8, scale_factor=2, mode=self.upMode)) 160 | 161 | return self.kernel_pred(data, core, white_level) 162 | 163 | 164 | class KernelConv(nn.Module): 165 | """ 166 | the class of computing prediction 167 | """ 168 | 169 | def __init__(self, kernel_size=[5], sep_conv=False, core_bias=False): 170 | super(KernelConv, self).__init__() 171 | self.kernel_size = sorted(kernel_size) 172 | self.sep_conv = sep_conv 173 | self.core_bias = core_bias 174 | 175 | def _sep_conv_core(self, core, batch_size, N, color, height, width): 176 | """ 177 | convert the sep_conv core to conv2d core 178 | 2p --> p^2 179 | :param core: shape: batch*(N*2*K)*height*width 180 | :return: 181 | """ 182 | kernel_total = sum(self.kernel_size) 183 | core = core.view(batch_size, N, -1, color, height, width) 184 | if not self.core_bias: 185 | core_1, core_2 = torch.split(core, kernel_total, dim=2) 186 | else: 187 | core_1, core_2, core_3 = torch.split(core, kernel_total, dim=2) 188 | # output core 189 | core_out = {} 190 | cur = 0 191 | for K in self.kernel_size: 192 | t1 = core_1[:, :, cur:cur + K, ...].view(batch_size, N, K, 1, color, 193 | height, width) 194 | t2 = core_2[:, :, cur:cur + K, ...].view(batch_size, N, 1, K, color, 195 | height, width) 196 | core_out[K] = torch.einsum('ijklnou,ijlmnou->ijkmnou', 197 | [t1, t2]).view( 198 | batch_size, N, K * K, color, height, width) 199 | cur += K 200 | # it is a dict 201 | return core_out, None if not self.core_bias else core_3.squeeze() 202 | 203 | def _convert_dict(self, core, batch_size, N, color, height, width): 204 | """ 205 | make sure the core to be a dict, generally, only one kind of kernel size is suitable for the func. 206 | :param core: shape: batch_size*(N*K*K)*height*width 207 | :return: core_out, a dict 208 | """ 209 | core_out = {} 210 | core = core.view(batch_size, N, -1, color, height, width) 211 | core_out[self.kernel_size[0]] = core[:, :, 0:self.kernel_size[0] ** 2, 212 | ...] 213 | bias = None if not self.core_bias else core[:, :, -1, ...] 214 | return core_out, bias 215 | 216 | def forward(self, frames, core, white_level=1.0): 217 | """ 218 | compute the pred image according to core and frames 219 | :param frames: [batch_size, N, 3, height, width] 220 | :param core: [batch_size, N, dict(kernel), 3, height, width] 221 | :return: 222 | """ 223 | if len(frames.size()) == 5: 224 | batch_size, N, color, height, width = frames.size() 225 | else: 226 | batch_size, N, height, width = frames.size() 227 | color = 1 228 | frames = frames.view(batch_size, N, color, height, width) 229 | if self.sep_conv: 230 | core, bias = self._sep_conv_core(core, batch_size, N, color, height, 231 | width) 232 | else: 233 | core, bias = self._convert_dict(core, batch_size, N, color, height, 234 | width) 235 | img_stack = [] 236 | pred_img = [] 237 | kernel = self.kernel_size[::-1] 238 | for index, K in enumerate(kernel): 239 | if len(img_stack) == 0: 240 | frame_pad = F.pad(frames, [K // 2, K // 2, K // 2, K // 2]) 241 | for i in range(K): 242 | for j in range(K): 243 | img_stack.append( 244 | frame_pad[..., i:i + height, j:j + width]) 245 | img_stack = torch.stack(img_stack, dim=2) 246 | else: 247 | k_diff = (kernel[index - 1] - kernel[index]) // 2 248 | img_stack = img_stack.view(batch_size, N, kernel[index - 1], 249 | kernel[index - 1], color, height, 250 | width) 251 | img_stack = img_stack[:, :, k_diff:-k_diff, k_diff:-k_diff, ...] 252 | img_stack = img_stack.reshape(batch_size, N, K ** 2, color, 253 | height, width) 254 | # print('img_stack:', img_stack.size()) 255 | pred_img.append(torch.sum( 256 | core[K].mul(img_stack), dim=2, keepdim=False 257 | )) 258 | pred_img = torch.stack(pred_img, dim=0) 259 | # print('pred_stack:', pred_img.size()) 260 | # pred_img_i = torch.mean(pred_img, dim=0, keepdim=False).squeeze() 261 | pred_img_i = torch.mean(pred_img, dim=0, keepdim=False) 262 | # if bias is permitted 263 | if self.core_bias: 264 | if bias is None: 265 | raise ValueError('The bias should not be None.') 266 | pred_img_i += bias 267 | # print('white_level', white_level.size()) 268 | pred_img_i = pred_img_i / white_level 269 | pred_img = torch.mean(pred_img_i, dim=1, keepdim=False) 270 | # print('pred_img:', pred_img.size()) 271 | # print('pred_img_i:', pred_img_i.size()) 272 | return pred_img_i, pred_img 273 | 274 | 275 | class LossFunc(nn.Module): 276 | """ 277 | loss function of KPN 278 | """ 279 | 280 | def __init__(self, coeff_basic=1.0, coeff_anneal=1.0, gradient_L1=True, 281 | alpha=0.9998, beta=100): 282 | super(LossFunc, self).__init__() 283 | self.coeff_basic = coeff_basic 284 | self.coeff_anneal = coeff_anneal 285 | self.loss_basic = LossBasic(gradient_L1) 286 | self.loss_anneal = LossAnneal(alpha, beta) 287 | 288 | def forward(self, pred_img_i, pred_img, ground_truth, global_step): 289 | """ 290 | forward function of loss_func 291 | :param frames: frame_1 ~ frame_N, shape: [batch, N, 3, height, width] 292 | :param core: a dict coverted by ...... 293 | :param ground_truth: shape [batch, 3, height, width] 294 | :param global_step: int 295 | :return: loss 296 | """ 297 | return self.coeff_basic * self.loss_basic(pred_img, 298 | ground_truth), self.coeff_anneal * self.loss_anneal( 299 | global_step, pred_img_i, ground_truth) 300 | 301 | 302 | class LossBasic(nn.Module): 303 | """ 304 | Basic loss function. 305 | """ 306 | 307 | def __init__(self, gradient_L1=True): 308 | super(LossBasic, self).__init__() 309 | self.l1_loss = nn.L1Loss() 310 | self.l2_loss = nn.MSELoss() 311 | self.gradient = TensorGradient(gradient_L1) 312 | 313 | def forward(self, pred, ground_truth): 314 | return self.l2_loss(pred, ground_truth) + \ 315 | self.l1_loss(self.gradient(pred), self.gradient(ground_truth)) 316 | 317 | 318 | class LossAnneal(nn.Module): 319 | """ 320 | anneal loss function 321 | """ 322 | 323 | def __init__(self, alpha=0.9998, beta=100): 324 | super(LossAnneal, self).__init__() 325 | self.global_step = 0 326 | self.loss_func = LossBasic(gradient_L1=True) 327 | self.alpha = alpha 328 | self.beta = beta 329 | 330 | def forward(self, global_step, pred_i, ground_truth): 331 | """ 332 | :param global_step: int 333 | :param pred_i: [batch_size, N, 3, height, width] 334 | :param ground_truth: [batch_size, 3, height, width] 335 | :return: 336 | """ 337 | loss = 0 338 | for i in range(pred_i.size(1)): 339 | loss += self.loss_func(pred_i[:, i, ...], ground_truth) 340 | loss /= pred_i.size(1) 341 | return self.beta * self.alpha ** global_step * loss 342 | 343 | 344 | class TensorGradient(nn.Module): 345 | """ 346 | the gradient of tensor 347 | """ 348 | 349 | def __init__(self, L1=True): 350 | super(TensorGradient, self).__init__() 351 | self.L1 = L1 352 | 353 | def forward(self, img): 354 | w, h = img.size(-2), img.size(-1) 355 | l = F.pad(img, [1, 0, 0, 0]) 356 | r = F.pad(img, [0, 1, 0, 0]) 357 | u = F.pad(img, [0, 0, 1, 0]) 358 | d = F.pad(img, [0, 0, 0, 1]) 359 | if self.L1: 360 | return torch.abs((l - r)[..., 0:w, 0:h]) + torch.abs( 361 | (u - d)[..., 0:w, 0:h]) 362 | else: 363 | return torch.sqrt( 364 | torch.pow((l - r)[..., 0:w, 0:h], 2) + torch.pow( 365 | (u - d)[..., 0:w, 0:h], 2) 366 | ) 367 | 368 | 369 | if __name__ == '__main__': 370 | kpn = KPN(6, 5 * 5 * 6, True, True).cuda() 371 | print(summary(kpn, (6, 224, 224), batch_size=4)) 372 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | ## Kernel Prediction Networks and Multi-Kernel Prediction Networks 2 | Reimplement of [Burst Denoising with Kernel Prediction Networks](https://arxiv.org/pdf/1712.02327.pdf) and [Multi-Kernel Prediction Networks for Denoising of Image Burst](https://arxiv.org/pdf/1902.05392.pdf) by using PyTorch. 3 | 4 | The partial work is following [https://github.com/12dmodel/camera_sim](https://github.com/12dmodel/camera_sim). 5 | 6 | ## TODO 7 | Write the documents. 8 | 9 | ## Requirements 10 | - Python3 11 | - PyTorch >= 1.0.0 12 | - Scikit-image 13 | - Numpy 14 | - TensorboardX (needed tensorflow support) 15 | 16 | ## How to use this repo? 17 | 18 | Firstly, you can clone this repo. including train and test codes. Download pretrained model for grayscale images at [https://drive.google.com/open?id=1Xnpllr1dinAU7BIN21L3LkEP5AqMNWso](https://drive.google.com/open?id=1Xnpllr1dinAU7BIN21L3LkEP5AqMNWso), and for color images at [https://drive.google.com/file/d/1Il-n7un_u8wWizjQ5ZKQ5hns7S27b0HW/view?usp=sharing](https://drive.google.com/file/d/1Il-n7un_u8wWizjQ5ZKQ5hns7S27b0HW/view?usp=sharing). 19 | 20 | The repo. supports multiple GPUs to train and validate, and the default setting is multi-GPUs. In other words, the pretrained model is obtained by training on multi-GPUs. 21 | 22 | - If you want to restart the train process by yourself, the command you should type is that 23 | ```angular2html 24 | CUDA_VISIBLE_DEVICES=x,y train_eval_sym.py --cuda --mGPU -nw 4 --config_file ./kpn_specs/kpn_config.conf --restart 25 | ``` 26 | If no option of `--restart`, the train process could be resumed from when it was broken. 27 | 28 | - If you want to evaluate the network by pre-trained model directly, you could use 29 | ```angular2html 30 | CUDA_VISIBLE_DEVICES=x,y train_eval_syn.py --cuda --mGPU -nw 4 --eval 31 | ``` 32 | If else option `-ckpt` is choosen, you can select the other models you trained. 33 | 34 | - Anything else. 35 | - The code for single image is not released now, I will program it in few weeks. 36 | 37 | ## Results 38 | ### on grayscale images: 39 | The following images and more examples can be found at [here](https://github.com/z-bingo/kernel-prediction-networks-PyTorch/blob/master/eval_images/). 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | 69 |
Ground Truth
Noisy
Denoised
Ground Truth
Noisy
Denoised
70 | 71 | ### on color images: 72 | The following images and more examples can be found at [here](https://github.com/LujiaJin/kernel-prediction-networks-PyTorch/tree/master/eval_images_RGB/). 73 | 74 | 75 | 76 | 77 | 78 | 79 | 80 | 81 | 82 | 83 | 84 | 85 | 86 | 87 | 88 | 89 | 90 | 91 | 92 | 93 | 94 | 95 | 96 | 97 | 98 | 99 | 100 | 101 | 102 |
Ground Truth
Noisy (PSNR: 19.34dB, SSIM: 0.595)
Denoised (PSNR: 29.69dB, SSIM: 0.937)
Ground Truth
Noisy (PSNR: 18.70dB, SSIM: 0.308)
Denoised (PSNR: 34.02dB, SSIM: 0.954)
103 | 104 | ##### If you like this repo, Star or Fork to support my work. Thank you. 105 | -------------------------------------------------------------------------------- /data_generation/__init__.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import inspect, os 3 | 4 | # Need this to import halide. See: 5 | # https://stackoverflow.com/questions/6323860/sibling-package-imports 6 | sys.path.insert(0, os.path.join( 7 | os.path.dirname(inspect.getfile(inspect.currentframe())), 8 | '..')) 9 | sys.path.insert(0, os.path.join( 10 | os.path.dirname(inspect.getfile(inspect.currentframe())))) 11 | -------------------------------------------------------------------------------- /data_generation/ahd_demosaicking.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import scipy 3 | from scipy.io import savemat 4 | from .constants import RGB2YUV 5 | from scipy.interpolate import interp2d 6 | 7 | 8 | _RGB2YUV = RGB2YUV.cpu().data.numpy() 9 | 10 | 11 | def ahd_demosaicking(mosaic, delta=1, sobel_sz=3, avg_sz=3): 12 | """Demosaicking using AHD algorithm. 13 | 14 | No median filtering, assume GRBG format. 15 | Args: 16 | delta: neighborhood size for calculating homogeneity. 17 | sobel_sz: size of sobel kernels. 18 | avg_sz: size of averaging kernel for homogeneity. 19 | """ 20 | Yx = _demosaickX(mosaic) 21 | Yy = _demosaickY(mosaic) 22 | 23 | YxYUV = _rgb2YUV(Yx) 24 | YyYUV = _rgb2YUV(Yy) 25 | 26 | epsL, epsCsq = _adaptive_param(YxYUV, YyYUV, sobel_sz) 27 | 28 | Hx = _homogeniety(YxYUV, delta, epsL, epsCsq) 29 | Hy = _homogeniety(YyYUV, delta, epsL, epsCsq) 30 | 31 | Hx = _conv2(Hx, np.ones((avg_sz, avg_sz)) / float(avg_sz**2)) 32 | Hy = _conv2(Hy, np.ones((avg_sz, avg_sz)) / float(avg_sz**2)) 33 | 34 | mask = (Hx > Hy).astype('float') 35 | mask = np.expand_dims(mask, -1) 36 | output = mask * Yx + (1.0 - mask) * Yy 37 | return np.clip(output, 0.0, 1.0) 38 | 39 | 40 | # https://stackoverflow.com/questions/9567882/sobel-filter-kernel-of-large-size/41065243#41065243 41 | def _sobel_kernel(sz): 42 | if (sz % 2) == 0: 43 | raise ValueError("Kernel size must be odd ({} received)".format(sz)) 44 | kernel = np.zeros((sz, sz)) 45 | for i in range(sz): 46 | for j in range(sz): 47 | ii = i - (sz // 2) 48 | jj = j - (sz // 2) 49 | kernel[i, j] = ii / (ii**2 + jj**2) if ii != 0 else 0 50 | return kernel 51 | 52 | 53 | def _interp2d(arr, new_sz): 54 | f = interp2d(x=np.linspace(0, 1, arr.shape[1]), 55 | y=np.linspace(0, 1, arr.shape[0]), 56 | z=arr) 57 | return f(np.linspace(0, 1, new_sz[1]), np.linspace(0, 1, new_sz[0])) 58 | 59 | 60 | def _interp_kernel(m=5, n=3): 61 | # Duplicate row so it works with bilinear interpolation 62 | Hg = np.array([[-0.25, 0.5, 0.5, 0.5, -0.25],[-0.25, 0.5, 0.5, 0.5, -0.25]]) 63 | Hr = np.array([[0.25, 0.5, 0.25], [0.5, 1.0, 0.5], [0.25, 0.5, 0.25]]) 64 | if m != 5: 65 | Hg = _interp2d(Hg, (2, m)) 66 | if n != 3: 67 | Hr = _interp2d(Hr, (n, n)) 68 | Hg = Hg[0:1, :] 69 | Hg = Hg / np.sum(Hg[:]) 70 | Hr = Hr / np.sum(Hr[:]) * 4 71 | return Hg, Hr 72 | 73 | 74 | def _conv2(x, k): 75 | return scipy.ndimage.filters.convolve(x, k, mode='reflect') 76 | 77 | 78 | def _demosaickX(X, transposed=False): 79 | Mr = np.zeros(X.shape) 80 | Mg = np.ones(X.shape) 81 | Mb = np.zeros(X.shape) 82 | 83 | Mr[0::2, 1::2] = 1.0 84 | Mb[1::2, 0::2] = 1.0 85 | Mg = Mg - Mr - Mb 86 | # Switch R and B (which got swapped when we transpose X). 87 | if transposed: 88 | Mr, Mb = Mb, Mr 89 | 90 | Hg, Hr = _interp_kernel(5, 3) 91 | G = Mg * X + (Mr + Mb) * _conv2(X, Hg) 92 | 93 | R = G + _conv2(Mr * (X - G), Hr) 94 | B = G + _conv2(Mb * (X - G), Hr) 95 | R = np.expand_dims(R, -1) 96 | G = np.expand_dims(G, -1) 97 | B = np.expand_dims(B, -1) 98 | return np.concatenate((R,G,B), axis=2) 99 | 100 | 101 | def _demosaickY(X): 102 | X = X.T 103 | Y = _demosaickX(X, transposed=True) 104 | Y = np.swapaxes(Y, 0, 1) 105 | return Y 106 | 107 | 108 | def _adaptive_param(X, Y, sz): 109 | sobel_y = _sobel_kernel(sz) 110 | sobel_x = sobel_y.T 111 | eL = np.minimum(abs(_conv2(X[:,:,0], sobel_x)), 112 | abs(_conv2(Y[:,:,0], sobel_y))) 113 | eCsq = np.minimum(_conv2(X[:,:,1], sobel_x)**2 + _conv2(X[:,:,2], sobel_x)**2, 114 | _conv2(Y[:,:,1], sobel_y)**2 + _conv2(Y[:,:,2], sobel_y)**2) 115 | return eL, eCsq 116 | 117 | 118 | def _rgb2YUV(X): 119 | return np.einsum("ijk,lk->ijl", X, _RGB2YUV) 120 | 121 | 122 | def _ballset(delta): 123 | index = int(np.ceil(delta)) 124 | # initialize 125 | H = np.zeros((index*2+1, index*2+1, (index*2+1)**2)) 126 | k = 0; 127 | for i in range(-index, index): 128 | for j in range(-index,index): 129 | if np.sqrt(i**2 + j**2) <= delta: 130 | # included 131 | H[index+i, index+j, k] = 1 132 | k = k + 1 133 | H = H[:,:,:k]; 134 | return H 135 | 136 | 137 | def _homogeniety(X, delta, epsL, epsC_sq): 138 | H = _ballset(delta); 139 | 140 | K = np.zeros(X.shape[:2]) 141 | 142 | for i in range(H.shape[-1]): 143 | # level set 144 | L = abs(_conv2(X[:,:,0], H[:,:,i]) - X[:,:,0]) <= epsL 145 | # color set 146 | C = ((_conv2(X[:,:,1], H[:,:,i]) - X[:,:,1])**2 + \ 147 | (_conv2(X[:,:,2], H[:,:,i]) - X[:,:,2])**2) <= epsC_sq; 148 | # metric neighborhood 149 | U = C * L 150 | # homogeneity 151 | K = K + U 152 | return K 153 | -------------------------------------------------------------------------------- /data_generation/constants.py: -------------------------------------------------------------------------------- 1 | import math 2 | import torch 3 | from torch import FloatTensor 4 | 5 | 6 | XYZ2sRGB = FloatTensor([[ 3.2406, -1.5372, -0.4986], 7 | [-0.9689, 1.8758, 0.0415], 8 | [ 0.0557, -0.2040, 1.0570]]) 9 | 10 | # http://brucelindbloom.com/index.html?Eqn_RGB_XYZ_Matrix.html 11 | ProPhotoRGB2XYZ = FloatTensor([[0.7976749, 0.1351917, 0.0313534], 12 | [0.2880402, 0.7118741, 0.0000857], 13 | [0.0000000, 0.0000000, 0.8252100]]) 14 | 15 | RGB2YUV = FloatTensor([[0.29900, 0.5870, 0.1140], 16 | [-.33750, -.6625, 1.0000], 17 | [1.00000, -.8374, -.1626]]) 18 | 19 | YUV2RGB = FloatTensor([[1.0, 0.0000, 0.7010], 20 | [1.0, -.1721, -.3571], 21 | [1.0, 0.8860, 0.0]]) 22 | 23 | xyz_color_matching = { 24 | "lambda": FloatTensor([390,395,400,405,410,415,420,425,430,435,440,445,450,455,460,465,470,475,480,485,490,495,500,505,510,515,520,525,530,535,540,545,550,555,560,565,570,575,580,585,590,595,600,605,610,615,620,625,630,635,640,645,650,655,660,665,670,675,680,685,690,695,700,705,710,715,720,725,730,735,740,745,750,755,760,765,770,775,780,785,790,795,800,805,810,815,820,825,830]), 25 | "xyz": FloatTensor([[0.003769647,0.009382967,0.02214302,0.04742986,0.08953803,0.1446214,0.2035729,0.2488523,0.2918246,0.3227087,0.3482554,0.3418483,0.3224637,0.2826646,0.2485254,0.2219781,0.1806905,0.129192,0.08182895,0.04600865,0.02083981,0.007097731,0.002461588,0.003649178,0.01556989,0.04315171,0.07962917,0.1268468,0.1818026,0.2405015,0.3098117,0.3804244,0.4494206,0.5280233,0.6133784,0.7016774,0.796775,0.8853376,0.9638388,1.051011,1.109767,1.14362,1.151033,1.134757,1.083928,1.007344,0.9142877,0.8135565,0.6924717,0.575541,0.4731224,0.3844986,0.2997374,0.2277792,0.1707914,0.1263808,0.09224597,0.0663996,0.04710606,0.03292138,0.02262306,0.01575417,0.01096778,0.00760875,0.005214608,0.003569452,0.002464821,0.001703876,0.001186238,0.000826954,0.00057583,0.00040583,0.000285658,0.000202185,0.000143827,0.000102469,7.34755E-05,5.25987E-05,3.80611E-05,2.75822E-05,2.00412E-05,1.45879E-05,1.06814E-05,7.85752E-06,5.76828E-06,4.25917E-06,3.16777E-06,2.35872E-06,1.76247E-06], 26 | [0.000414616,0.001059646,0.002452194,0.004971717,0.00907986,0.01429377,0.02027369,0.02612106,0.03319038,0.0415794,0.05033657,0.05743393,0.06472352,0.07238339,0.08514816,0.1060145,0.1298957,0.1535066,0.1788048,0.2064828,0.237916,0.285068,0.3483536,0.4277595,0.5204972,0.6206256,0.718089,0.7946448,0.8575799,0.9071347,0.9544675,0.9814106,0.9890228,0.9994608,0.9967737,0.9902549,0.9732611,0.9424569,0.8963613,0.8587203,0.8115868,0.7544785,0.6918553,0.6270066,0.5583746,0.489595,0.4229897,0.3609245,0.2980865,0.2416902,0.1943124,0.1547397,0.119312,0.08979594,0.06671045,0.04899699,0.03559982,0.02554223,0.01807939,0.01261573,0.008661284,0.006027677,0.004195941,0.002910864,0.001995557,0.001367022,0.000944727,0.000653705,0.000455597,0.000317974,0.000221745,0.000156557,0.000110393,7.82744E-05,5.57886E-05,3.98188E-05,2.86018E-05,2.05126E-05,1.48724E-05,1.08E-05,7.86392E-06,5.73694E-06,4.2116E-06,3.10656E-06,2.28679E-06,1.69315E-06,1.26256E-06,9.42251E-07,7.05386E-07], 27 | [0.0184726,0.04609784,0.109609,0.2369246,0.4508369,0.7378822,1.051821,1.305008,1.552826,1.74828,1.917479,1.918437,1.848545,1.664439,1.522157,1.42844,1.25061,0.9991789,0.7552379,0.5617313,0.4099313,0.3105939,0.2376753,0.1720018,0.1176796,0.08283548,0.05650407,0.03751912,0.02438164,0.01566174,0.00984647,0.006131421,0.003790291,0.002327186,0.001432128,0.000882253,0.000545242,0.000338674,0.000211777,0.000133503,8.49447E-05,5.46071E-05,3.54966E-05,2.33474E-05,1.55463E-05,1.04839E-05,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]]) 28 | } 29 | 30 | 31 | # default 50% quality 32 | default_jpeg_quantization_matrix = \ 33 | FloatTensor([[16, 11, 10, 16, 24, 40, 51, 61], 34 | [12, 12, 14, 19, 26, 58, 60, 55], 35 | [14, 13, 16, 24, 40, 57, 69, 56], 36 | [14, 17, 22, 29, 51, 87, 80, 62], 37 | [18, 22, 37, 56, 68, 109, 103, 77], 38 | [24, 35, 55, 64, 81, 104, 113, 92], 39 | [49, 64, 78, 87, 103, 121, 120, 101], 40 | [72, 92, 95, 98, 112, 100, 103, 99]]) 41 | 42 | # Photoshop quantization. 43 | # https://www.impulseadventure.com/photo/jpeg-quantization.html 44 | photoshop_jpeg_quantization_lum = \ 45 | [ 46 | # Luminance Level 0 47 | FloatTensor([ 48 | [32, 33, 51, 81, 66, 39, 34, 17], 49 | [33, 36, 48, 47, 28, 23, 12, 12], 50 | [51, 48, 47, 28, 23, 12, 12, 12], 51 | [81, 47, 28, 23, 12, 12, 12, 12], 52 | [66, 28, 23, 12, 12, 12, 12, 12], 53 | [39, 23, 12, 12, 12, 12, 12, 12], 54 | [34, 12, 12, 12, 12, 12, 12, 12], 55 | [17, 12, 12, 12, 12, 12, 12, 12], 56 | ]), 57 | # Luminance Level 1 58 | FloatTensor([ 59 | [27, 26, 41, 65, 66, 39, 34, 17], 60 | [26, 29, 38, 47, 28, 23, 12, 12], 61 | [41, 38, 47, 28, 23, 12, 12, 12], 62 | [65, 47, 28, 23, 12, 12, 12, 12], 63 | [66, 28, 23, 12, 12, 12, 12, 12], 64 | [39, 23, 12, 12, 12, 12, 12, 12], 65 | [34, 12, 12, 12, 12, 12, 12, 12], 66 | [17, 12, 12, 12, 12, 12, 12, 12], 67 | ]), 68 | # Luminance Level 2 69 | FloatTensor([ 70 | [20, 17, 26, 41, 51, 39, 34, 17], 71 | [17, 18, 24, 39, 28, 23, 12, 12], 72 | [26, 24, 32, 28, 23, 12, 12, 12], 73 | [41, 39, 28, 23, 12, 12, 12, 12], 74 | [51, 28, 23, 12, 12, 12, 12, 12], 75 | [39, 23, 12, 12, 12, 12, 12, 12], 76 | [34, 12, 12, 12, 12, 12, 12, 12], 77 | [17, 12, 12, 12, 12, 12, 12, 12], 78 | ]), 79 | # Luminance Level 3 80 | FloatTensor([ 81 | [18, 14, 22, 35, 44, 39, 34, 17], 82 | [14, 16, 21, 34, 28, 23, 12, 12], 83 | [22, 21, 27, 28, 23, 12, 12, 12], 84 | [35, 34, 28, 23, 12, 12, 12, 12], 85 | [44, 28, 23, 12, 12, 12, 12, 12], 86 | [39, 23, 12, 12, 12, 12, 12, 12], 87 | [34, 12, 12, 12, 12, 12, 12, 12], 88 | [17, 12, 12, 12, 12, 12, 12, 12], 89 | ]), 90 | # Luminance Level 4 91 | FloatTensor([ 92 | [16, 11, 17, 27, 34, 39, 34, 17], 93 | [11, 12, 16, 26, 28, 23, 12, 12], 94 | [17, 16, 21, 28, 23, 12, 12, 12], 95 | [27, 26, 28, 23, 12, 12, 12, 12], 96 | [34, 28, 23, 12, 12, 12, 12, 12], 97 | [39, 23, 12, 12, 12, 12, 12, 12], 98 | [34, 12, 12, 12, 12, 12, 12, 12], 99 | [17, 12, 12, 12, 12, 12, 12, 12], 100 | ]), 101 | # Luminance Level 5 102 | FloatTensor([ 103 | [12, 8, 13, 21, 26, 32, 34, 17], 104 | [8, 9, 12, 20, 27, 23, 12, 12], 105 | [13, 12, 16, 26, 23, 12, 12, 12], 106 | [21, 20, 26, 23, 12, 12, 12, 12], 107 | [26, 27, 23, 12, 12, 12, 12, 12], 108 | [32, 23, 12, 12, 12, 12, 12, 12], 109 | [34, 12, 12, 12, 12, 12, 12, 12], 110 | [17, 12, 12, 12, 12, 12, 12, 12], 111 | ]), 112 | # Luminance Level 6 113 | FloatTensor([ 114 | [8, 6, 9, 14, 17, 21, 28, 17], 115 | [6, 6, 8, 13, 18, 23, 12, 12], 116 | [9, 8, 11, 17, 23, 12, 12, 12], 117 | [14, 13, 17, 23, 12, 12, 12, 12], 118 | [17, 18, 23, 12, 12, 12, 12, 12], 119 | [21, 23, 12, 12, 12, 12, 12, 12], 120 | [28, 12, 12, 12, 12, 12, 12, 12], 121 | [17, 12, 12, 12, 12, 12, 12, 12], 122 | ]), 123 | # Luminance Level 7 124 | FloatTensor([ 125 | [10, 7, 11, 18, 22, 27, 34, 17], 126 | [7, 8, 10, 17, 23, 23, 12, 12], 127 | [11, 10, 14, 22, 23, 12, 12, 12], 128 | [18, 17, 22, 23, 12, 12, 12, 12], 129 | [22, 23, 23, 12, 12, 12, 12, 12], 130 | [27, 23, 12, 12, 12, 12, 12, 12], 131 | [34, 12, 12, 12, 12, 12, 12, 12], 132 | [17, 12, 12, 12, 12, 12, 12, 12], 133 | ]), 134 | # Luminance Level 8 135 | FloatTensor([ 136 | [6, 4, 7, 11, 14, 17, 22, 17], 137 | [4, 5, 6, 10, 14, 19, 12, 12], 138 | [7, 6, 8, 14, 19, 12, 12, 12], 139 | [11, 10, 14, 19, 12, 12, 12, 12], 140 | [14, 14, 19, 12, 12, 12, 12, 12], 141 | [17, 19, 12, 12, 12, 12, 12, 12], 142 | [22, 12, 12, 12, 12, 12, 12, 12], 143 | [17, 12, 12, 12, 12, 12, 12, 12], 144 | ]), 145 | # Luminance Level 9 146 | FloatTensor([ 147 | [4, 3, 4, 7, 9, 11, 14, 17], 148 | [3, 3, 4, 7, 9, 12, 12, 12], 149 | [4, 4, 5, 9, 12, 12, 12, 12], 150 | [7, 7, 9, 12, 12, 12, 12, 12], 151 | [9, 9, 12, 12, 12, 12, 12, 12], 152 | [11, 12, 12, 12, 12, 12, 12, 12], 153 | [14, 12, 12, 12, 12, 12, 12, 12], 154 | [17, 12, 12, 12, 12, 12, 12, 12], 155 | ]), 156 | # Luminance Level 10 157 | FloatTensor([ 158 | [2, 2, 3, 4, 5, 6, 8, 11], 159 | [2, 2, 2, 4, 5, 7, 9, 11], 160 | [3, 2, 3, 5, 7, 9, 11, 12], 161 | [4, 4, 5, 7, 9, 11, 12, 12], 162 | [5, 5, 7, 9, 11, 12, 12, 12], 163 | [6, 7, 9, 11, 12, 12, 12, 12], 164 | [8, 9, 11, 12, 12, 12, 12, 12], 165 | [11, 11, 12, 12, 12, 12, 12, 12], 166 | ]), 167 | # Luminance Level 11 168 | FloatTensor([ 169 | [1, 1, 1, 2, 3, 3, 4, 5], 170 | [1, 1, 1, 2, 3, 4, 4, 6], 171 | [1, 1, 2, 3, 4, 4, 5, 7], 172 | [2, 2, 3, 4, 4, 5, 7, 8], 173 | [3, 3, 4, 4, 5, 7, 8, 8], 174 | [3, 4, 4, 5, 7, 8, 8, 8], 175 | [4, 4, 5, 7, 8, 8, 8, 8], 176 | [5, 6, 7, 8, 8, 8, 8, 8], 177 | ]), 178 | # Luminance Level 12 179 | FloatTensor([ 180 | [1, 1, 1, 1, 1, 1, 1, 2], 181 | [1, 1, 1, 1, 1, 1, 1, 2], 182 | [1, 1, 1, 1, 1, 1, 2, 2], 183 | [1, 1, 1, 1, 1, 2, 2, 3], 184 | [1, 1, 1, 1, 2, 2, 3, 3], 185 | [1, 1, 1, 2, 2, 3, 3, 3], 186 | [1, 1, 2, 2, 3, 3, 3, 3], 187 | [2, 2, 2, 3, 3, 3, 3, 3], 188 | ]), 189 | ] 190 | 191 | photoshop_jpeg_quantization_chrom = \ 192 | [ 193 | # Chrominance Level 0 194 | FloatTensor([ 195 | [34, 51, 52, 34, 20, 20, 17, 17], 196 | [51, 38, 24, 14, 14, 12, 12, 12], 197 | [52, 24, 14, 14, 12, 12, 12, 12], 198 | [34, 14, 14, 12, 12, 12, 12, 12], 199 | [20, 14, 12, 12, 12, 12, 12, 12], 200 | [20, 12, 12, 12, 12, 12, 12, 12], 201 | [17, 12, 12, 12, 12, 12, 12, 12], 202 | [17, 12, 12, 12, 12, 12, 12, 12], 203 | ]), 204 | # Chrominance Level 1 205 | FloatTensor([ 206 | [29, 41, 52, 34, 20, 20, 17, 17], 207 | [41, 38, 24, 14, 14, 12, 12, 12], 208 | [52, 24, 14, 14, 12, 12, 12, 12], 209 | [34, 14, 14, 12, 12, 12, 12, 12], 210 | [20, 14, 12, 12, 12, 12, 12, 12], 211 | [20, 12, 12, 12, 12, 12, 12, 12], 212 | [17, 12, 12, 12, 12, 12, 12, 12], 213 | [17, 12, 12, 12, 12, 12, 12, 12], 214 | ]), 215 | # Chrominance Level 2 216 | FloatTensor([ 217 | [21, 26, 33, 34, 20, 20, 17, 17], 218 | [26, 29, 24, 14, 14, 12, 12, 12], 219 | [33, 24, 14, 14, 12, 12, 12, 12], 220 | [34, 14, 14, 12, 12, 12, 12, 12], 221 | [20, 14, 12, 12, 12, 12, 12, 12], 222 | [20, 12, 12, 12, 12, 12, 12, 12], 223 | [17, 12, 12, 12, 12, 12, 12, 12], 224 | [17, 12, 12, 12, 12, 12, 12, 12], 225 | ]), 226 | # Chrominance Level 3 227 | FloatTensor([ 228 | [20, 22, 29, 34, 20, 20, 17, 17], 229 | [22, 25, 24, 14, 14, 12, 12, 12], 230 | [29, 24, 14, 14, 12, 12, 12, 12], 231 | [34, 14, 14, 12, 12, 12, 12, 12], 232 | [20, 14, 12, 12, 12, 12, 12, 12], 233 | [20, 12, 12, 12, 12, 12, 12, 12], 234 | [17, 12, 12, 12, 12, 12, 12, 12], 235 | [17, 12, 12, 12, 12, 12, 12, 12], 236 | ]), 237 | # Chrominance Level 4 238 | FloatTensor([ 239 | [17, 17, 22, 34, 20, 20, 17, 17], 240 | [17, 19, 22, 14, 14, 12, 12, 12], 241 | [22, 22, 14, 14, 12, 12, 12, 12], 242 | [34, 14, 14, 12, 12, 12, 12, 12], 243 | [20, 14, 12, 12, 12, 12, 12, 12], 244 | [20, 12, 12, 12, 12, 12, 12, 12], 245 | [17, 12, 12, 12, 12, 12, 12, 12], 246 | [17, 12, 12, 12, 12, 12, 12, 12], 247 | ]), 248 | # Chrominance Level 5 249 | FloatTensor([ 250 | [13, 13, 17, 27, 20, 20, 17, 17], 251 | [13, 14, 17, 14, 14, 12, 12, 12], 252 | [17, 17, 14, 14, 12, 12, 12, 12], 253 | [27, 14, 14, 12, 12, 12, 12, 12], 254 | [20, 14, 12, 12, 12, 12, 12, 12], 255 | [20, 12, 12, 12, 12, 12, 12, 12], 256 | [17, 12, 12, 12, 12, 12, 12, 12], 257 | [17, 12, 12, 12, 12, 12, 12, 12], 258 | ]), 259 | # Chrominance Level 6 260 | FloatTensor([ 261 | [9, 9, 11, 18, 20, 20, 17, 17], 262 | [9, 10, 11, 14, 14, 12, 12, 12], 263 | [11, 11, 14, 14, 12, 12, 12, 12], 264 | [18, 14, 14, 12, 12, 12, 12, 12], 265 | [20, 14, 12, 12, 12, 12, 12, 12], 266 | [20, 12, 12, 12, 12, 12, 12, 12], 267 | [17, 12, 12, 12, 12, 12, 12, 12], 268 | [17, 12, 12, 12, 12, 12, 12, 12], 269 | ]), 270 | # Chrominance Level 7 271 | FloatTensor([ 272 | [11, 14, 31, 34, 20, 20, 17, 17], 273 | [14, 19, 24, 14, 14, 12, 12, 12], 274 | [31, 24, 14, 14, 12, 12, 12, 12], 275 | [34, 14, 14, 12, 12, 12, 12, 12], 276 | [20, 14, 12, 12, 12, 12, 12, 12], 277 | [20, 12, 12, 12, 12, 12, 12, 12], 278 | [17, 12, 12, 12, 12, 12, 12, 12], 279 | [17, 12, 12, 12, 12, 12, 12, 12], 280 | ]), 281 | # Chrominance Level 8 282 | FloatTensor([ 283 | [7, 9, 19, 34, 20, 20, 17, 17], 284 | [9, 12, 19, 14, 14, 12, 12, 12], 285 | [19, 19, 14, 14, 12, 12, 12, 12], 286 | [34, 14, 14, 12, 12, 12, 12, 12], 287 | [20, 14, 12, 12, 12, 12, 12, 12], 288 | [20, 12, 12, 12, 12, 12, 12, 12], 289 | [17, 12, 12, 12, 12, 12, 12, 12], 290 | [17, 12, 12, 12, 12, 12, 12, 12], 291 | ]), 292 | # Chrominance Level 9 293 | FloatTensor([ 294 | [4, 6, 12, 22, 20, 20, 17, 17], 295 | [6, 8, 12, 14, 14, 12, 12, 12], 296 | [12, 12, 14, 14, 12, 12, 12, 12], 297 | [22, 14, 14, 12, 12, 12, 12, 12], 298 | [20, 14, 12, 12, 12, 12, 12, 12], 299 | [20, 12, 12, 12, 12, 12, 12, 12], 300 | [17, 12, 12, 12, 12, 12, 12, 12], 301 | [17, 12, 12, 12, 12, 12, 12, 12], 302 | ]), 303 | # Chrominance Level 10 304 | FloatTensor([ 305 | [3, 3, 7, 13, 15, 15, 15, 15], 306 | [3, 4, 7, 13, 14, 12, 12, 12], 307 | [7, 7, 13, 14, 12, 12, 12, 12], 308 | [13, 13, 14, 12, 12, 12, 12, 12], 309 | [15, 14, 12, 12, 12, 12, 12, 12], 310 | [15, 12, 12, 12, 12, 12, 12, 12], 311 | [15, 12, 12, 12, 12, 12, 12, 12], 312 | [15, 12, 12, 12, 12, 12, 12, 12], 313 | ]), 314 | # Chrominance Level 11 315 | FloatTensor([ 316 | [1, 2, 4, 7, 8, 8, 8, 8], 317 | [2, 2, 4, 7, 8, 8, 8, 8], 318 | [4, 4, 7, 8, 8, 8, 8, 8], 319 | [7, 7, 8, 8, 8, 8, 8, 8], 320 | [8, 8, 8, 8, 8, 8, 8, 8], 321 | [8, 8, 8, 8, 8, 8, 8, 8], 322 | [8, 8, 8, 8, 8, 8, 8, 8], 323 | [8, 8, 8, 8, 8, 8, 8, 8], 324 | ]), 325 | # Chrominance Level 12 326 | FloatTensor([ 327 | [1, 1, 1, 2, 3, 3, 3, 3], 328 | [1, 1, 1, 2, 3, 3, 3, 3], 329 | [1, 1, 2, 3, 3, 3, 3, 3], 330 | [2, 2, 3, 3, 3, 3, 3, 3], 331 | [3, 3, 3, 3, 3, 3, 3, 3], 332 | [3, 3, 3, 3, 3, 3, 3, 3], 333 | [3, 3, 3, 3, 3, 3, 3, 3], 334 | [3, 3, 3, 3, 3, 3, 3, 3], 335 | ]), 336 | ] 337 | 338 | # 0-6 have subsampling, 7-12 don't. 339 | photoshop_chroma_subsampling = [True] * 7 + [False] * 6 340 | 341 | 342 | 343 | # DCT Coefficients 344 | # The inverse coefficient is the same. 345 | def _DCT_coeff(): 346 | v = torch.arange(8).unsqueeze(-1).unsqueeze(-1).unsqueeze(-1).expand((8, 8, 8, 8)).float() 347 | u = torch.arange(8).unsqueeze( 0).unsqueeze(-1).unsqueeze(-1).expand((8, 8, 8, 8)).float() 348 | y = torch.arange(8).unsqueeze( 0).unsqueeze( 0).unsqueeze(-1).expand((8, 8, 8, 8)).float() 349 | x = torch.arange(8).unsqueeze( 0).unsqueeze( 0).unsqueeze( 0).expand((8, 8, 8, 8)).float() 350 | au = torch.ones((8, 8, 8, 8)).float() 351 | av = torch.ones((8, 8, 8, 8)).float() 352 | av[0, :, ...] = 0.707 # 1 / sqrt(2) 353 | au[:, 0, ...] = 0.707 # 1 / sqrt(2) 354 | 355 | coeff = au * av * torch.cos((2*x + 1)*u*math.pi/16.0) \ 356 | * torch.cos((2*y + 1)*v*math.pi/16.0) 357 | return coeff * 0.25 358 | 359 | DCT_coeff = _DCT_coeff() 360 | -------------------------------------------------------------------------------- /data_generation/data_utils.py: -------------------------------------------------------------------------------- 1 | """ Utilities functions. 2 | """ 3 | import numbers 4 | import numpy as np 5 | import torch 6 | from torch import FloatTensor 7 | 8 | 9 | def random_crop(im, num_patches, w, h=None): 10 | h = w if h is None else h 11 | nw = im.size(-1) - w 12 | nh = im.size(-2) - h 13 | if nw < 0 or nh < 0: 14 | raise RuntimeError("Image is to small {} for the desired size {}". \ 15 | format((im.size(-1), im.size(-2)), (w, h)) 16 | ) 17 | 18 | idx_w = np.random.choice(nw + 1, size=num_patches) 19 | idx_h = np.random.choice(nh + 1, size=num_patches) 20 | 21 | result = [] 22 | for i in range(num_patches): 23 | result.append(im[..., 24 | idx_h[i]:(idx_h[i]+h), 25 | idx_w[i]:(idx_w[i]+w)]) 26 | return result 27 | 28 | 29 | def expand_to_4d_channel(arr): 30 | """ Expand Scalar or 1D dimension to 4D 31 | 32 | Assumes that a 1D list represent the channel dimension (2nd dim). 33 | 34 | Args: 35 | arr: A scalar or 1D tensor to be expanded to 4D 36 | """ 37 | # for scalar and 1D tensor, add batch dimensions. 38 | while len(arr.size()) < 2: 39 | arr = arr.unsqueeze(0) 40 | # regain spatial dimension 41 | while len(arr.size()) < 4: 42 | arr = arr.unsqueeze(-1) 43 | return arr 44 | 45 | 46 | def expand_to_4d_batch(arr): 47 | """ Expand Scalar or 1D dimension to 4D 48 | 49 | Assumes that a 1D list represent the batch dimension (1st dim). 50 | 51 | Args: 52 | arr: A scalar or 1D tensor to be expanded to 4D 53 | """ 54 | # regain spatial dimension and channel dimension 55 | while len(arr.size()) < 4: 56 | arr = arr.unsqueeze(-1) 57 | return arr 58 | 59 | 60 | def is_number(a): 61 | return isinstance(a, numbers.Number) 62 | 63 | 64 | def python_to_tensor(a): 65 | if isinstance(a, numbers.Number): 66 | return FloatTensor([a]) 67 | return a 68 | 69 | 70 | def number_to_list(a): 71 | if isinstance(a, numbers.Number): 72 | a = [a] 73 | return a 74 | 75 | 76 | def cuda_like(arr, src): 77 | """ Move arr on to GPU/CPU like src 78 | """ 79 | if src.is_cuda: 80 | return arr.cuda() 81 | else: 82 | return arr.cpu() 83 | 84 | 85 | def mosaick_multiply(mult, im, mosaick_pattern): 86 | """ mosaick pattern-aware multiply. 87 | 88 | Args: 89 | mult: n-list of multiplier, where n is number of image channel. 90 | A batch dimension is optional. 91 | im: tensor of size n_batch x n_channel x width x height. 92 | mosaick_pattern: None or string indicating the mosaick pattern. 93 | """ 94 | if mosaick_pattern is None: 95 | return im * expand_to_4d_channel(mult) 96 | elif mosaick_pattern == "bayer": 97 | # Assume GRGB format. 98 | mult = expand_to_4d_channel(mult) 99 | 100 | h, w = im.size(2), im.size(3) 101 | x = torch.arange(w).unsqueeze(0).expand(h, -1) 102 | y = torch.arange(h).unsqueeze(-1).expand(-1, w) 103 | x = x.unsqueeze(0).unsqueeze(0) 104 | y = y.unsqueeze(0).unsqueeze(0) 105 | 106 | if im.is_cuda: 107 | x = x.cuda() 108 | y = y.cuda() 109 | 110 | odd_x = torch.fmod(x, 2) 111 | odd_y = torch.fmod(y, 2) 112 | 113 | is_green = odd_x == odd_y 114 | is_red = odd_x * (1.0 - odd_y) 115 | is_blue = (1.0 - odd_x) * odd_y 116 | 117 | mult = mult.expand(-1, 3, -1, -1) 118 | 119 | return im * mult[:, 0:1, ...] * is_red.float() + \ 120 | im * mult[:, 1:2, ...] * is_green.float() + \ 121 | im * mult[:, 2:3, ...] * is_blue.float() 122 | else: 123 | raise ValueError("Mosaick pattern, {}, not supported." \ 124 | .format(mosaick_pattern)) 125 | -------------------------------------------------------------------------------- /data_generation/generate_dataset.py: -------------------------------------------------------------------------------- 1 | import tifffile 2 | import skimage 3 | import numpy as np 4 | import os 5 | import argparse 6 | import glob 7 | import json 8 | from tqdm import tqdm 9 | from sklearn.feature_extraction.image import extract_patches_2d 10 | 11 | import torch 12 | from torch.autograd import Variable 13 | from torch import FloatTensor 14 | 15 | from data_generation.pipeline import ImageDegradationPipeline 16 | from data_generation.constants import XYZ2sRGB, ProPhotoRGB2XYZ 17 | 18 | 19 | def numpy2tensor(arr): 20 | if len(arr.shape) < 3: 21 | arr = np.expand_dims(arr, -1) 22 | return FloatTensor(arr).permute(2, 0, 1).unsqueeze(0).float() / 255.0 23 | 24 | 25 | def tensor2numpy(t, idx=None): 26 | t = torch.clamp(t, 0, 1) 27 | if idx is None: 28 | t = t[0, ...] 29 | else: 30 | t = t[idx, ...] 31 | return t.permute(1, 2, 0).cpu().squeeze().numpy() 32 | 33 | 34 | parser = argparse.ArgumentParser(description='') 35 | parser.add_argument('--im_folder', required=True, help='path to input images') 36 | parser.add_argument('--out_dir', required=True, help='path to place output') 37 | parser.add_argument('--total_patch', type=int, required=True, help='total number of patches to generate') 38 | 39 | parser.add_argument('--patch_per_image', type=int, default=5, help='Number of patch to generate from a single degradation of an image') 40 | parser.add_argument('--patch_sz', type=int, default=256, help='Patch size (square patch for now)') 41 | parser.add_argument('--fraction_train', type=float, default=0.8, help='Fraction of images to use as training') 42 | 43 | parser.add_argument('--input_ext', default='tif', help='path to place output') 44 | parser.add_argument('--max_exposure', type=float, default=0.0, help='maximum exposure adjustment in stops') 45 | parser.add_argument('--min_exposure', type=float, default=0.0, help='minimum exposure adjustment in stops') 46 | parser.add_argument('--max_gaussian_noise', type=float, default=0.0, help='maximum gaussian noise std (on range 0 - 1)') 47 | parser.add_argument('--min_gaussian_noise', type=float, default=0.0, help='minimum gaussian noise std (on range 0 - 1)') 48 | parser.add_argument('--max_poisson_noise', type=float, default=0.0, help='maximum poisson noise mult (See image_processing.PoissonNoise for detail)') 49 | parser.add_argument('--min_poisson_noise', type=float, default=0.0, help='minimum poisson noise mult (See image_processing.PoissonNoise for detail)') 50 | parser.add_argument('--skip_degraded', action="store_true", help='Whether to skip degraded images.') 51 | parser.add_argument('--dwn_factor', type=float, default=4, help='Factor to downsample.') 52 | args = parser.parse_args() 53 | 54 | 55 | im_names = glob.glob(os.path.join(args.im_folder, '*.' + args.input_ext)) 56 | im_names = sorted([os.path.basename(i) for i in im_names]) 57 | 58 | # Create output folder 59 | os.makedirs(args.out_dir, exist_ok=True) 60 | train_dir = os.path.join(args.out_dir, 'train') 61 | test_dir = os.path.join(args.out_dir, 'test') 62 | os.makedirs(train_dir, exist_ok=True) 63 | os.makedirs(test_dir, exist_ok=True) 64 | for base_dir in [train_dir, test_dir]: 65 | target_dir = os.path.join(base_dir, 'images', 'target') 66 | degraded_dir = os.path.join(base_dir, 'images', 'degraded') 67 | meta_dir = os.path.join(base_dir, 'meta') 68 | os.makedirs(target_dir, exist_ok=True) 69 | os.makedirs(degraded_dir, exist_ok=True) 70 | os.makedirs(meta_dir, exist_ok=True) 71 | 72 | n_count = 0 73 | img_idx = 0 74 | 75 | progress_bar = tqdm(total=args.total_patch) 76 | while n_count < args.total_patch: 77 | if img_idx < args.fraction_train * len(im_names): 78 | base_dir = train_dir 79 | else: 80 | base_dir = test_dir 81 | 82 | target_dir = os.path.join(base_dir, 'images', 'target') 83 | degraded_dir = os.path.join(base_dir, 'images', 'degraded') 84 | meta_dir = os.path.join(base_dir, 'meta') 85 | 86 | name = im_names[img_idx] 87 | path = os.path.join(args.im_folder, name) 88 | # We know 5k dataset is 16-bits. 89 | raw_im = tifffile.imread(path).astype('float32') / 65536.0 90 | raw_im = FloatTensor(raw_im).permute(2, 0, 1).unsqueeze(0) 91 | 92 | # Define pipeline 93 | poisson_k = np.random.uniform(args.min_poisson_noise, args.max_poisson_noise) 94 | read_noise_sigma = np.random.uniform(args.min_gaussian_noise, args.max_gaussian_noise) 95 | dwn_factor = args.dwn_factor 96 | exp_adjustment = np.random.uniform(args.min_exposure, args.max_exposure) 97 | configs_prepreprocess = [ 98 | ('UndoProPhotoRGBGamma', {}), 99 | # Convert to sRGB 100 | ('ColorSpaceConversionMatrix', {'matrix': torch.matmul(XYZ2sRGB, ProPhotoRGB2XYZ)}), 101 | ] 102 | 103 | configs_preprocess = [ 104 | # Blur and downsample to reduce noise 105 | ('GaussianBlur', {'sigma_x': dwn_factor}), 106 | ('PytorchResizing', {'resizing_factor': 1.0/dwn_factor, 'mode': 'nearest'}) 107 | ] 108 | configs_degrade = [ 109 | ('ExposureAdjustment', {'nstops': exp_adjustment}), 110 | # ('MotionBlur', {'amt': [3, 2], 'direction': [0, 45,]}), 111 | ('BayerMosaicking', {}), 112 | # Add artificial noise. 113 | ('PoissonNoise',{'sigma': FloatTensor([poisson_k] * 3), 'mosaick_pattern': 'bayer'}), 114 | ('GaussianNoise',{'sigma': FloatTensor([read_noise_sigma] * 3), 'mosaick_pattern': 'bayer'}), 115 | ('PixelClip', {}), 116 | ('ExposureAdjustment', {'nstops': -exp_adjustment}), 117 | ('PixelClip', {}), 118 | ('NaiveDemosaicking', {}), 119 | ('PixelClip', {}), 120 | ] 121 | configs_denoise = [ 122 | ('DenoisingBilateral',{'sigma_s': 1.0, 'sigma_r': 0.1}), 123 | ('PixelClip', {}), 124 | ('sRGBGamma', {}), 125 | ] 126 | 127 | pipeline_prepreprocess = ImageDegradationPipeline(configs_prepreprocess) 128 | pipeline_preprocess = ImageDegradationPipeline(configs_preprocess) 129 | pipeline_degrade = ImageDegradationPipeline(configs_degrade) 130 | pipeline_denoise = ImageDegradationPipeline(configs_denoise) 131 | 132 | 133 | demosaicked = pipeline_prepreprocess(raw_im) 134 | preprocessed = pipeline_preprocess(demosaicked) 135 | degraded = pipeline_degrade(preprocessed) 136 | denoised = pipeline_denoise(degraded) 137 | 138 | denoised_numpy = tensor2numpy(denoised) 139 | preprocessed_numpy = tensor2numpy(preprocessed) 140 | stacked = np.concatenate((denoised_numpy, preprocessed_numpy), axis=-1) 141 | patches = extract_patches_2d(stacked, 142 | (args.patch_sz, args.patch_sz), 143 | args.patch_per_image) 144 | degraded_patches, target_patches = np.split(patches, 2, axis=-1) 145 | 146 | target_patches = np.split(target_patches, target_patches.shape[0]) 147 | degraded_patches = np.split(degraded_patches, degraded_patches.shape[0]) 148 | 149 | meta = dict(orig=name, 150 | poisson_k=poisson_k, 151 | read_noise_sigma=read_noise_sigma, 152 | exp_adjustment=exp_adjustment, 153 | dwn_factor=dwn_factor) 154 | n_patches = len(degraded_patches) 155 | for i in range(n_patches): 156 | patch_idx = n_count + i + 1 157 | 158 | degraded = np.clip(degraded_patches[i] * 255.0, 0, 255).astype('uint8') 159 | if not args.skip_degraded: 160 | skimage.io.imsave(os.path.join(degraded_dir, 161 | "{:06d}.png".format(patch_idx) 162 | ), 163 | np.squeeze(degraded)) 164 | np.save(os.path.join(target_dir, 165 | "{:06d}.npy".format(patch_idx) 166 | ), 167 | np.squeeze(target_patches[i])) 168 | with open(os.path.join(meta_dir, 169 | '{:06d}.json'.format(patch_idx)), 170 | 'w') as f: 171 | json.dump(meta, f) 172 | n_count += n_patches 173 | img_idx = (img_idx + 1) % len(im_names) 174 | progress_bar.update(n_patches) 175 | progress_bar.close() 176 | -------------------------------------------------------------------------------- /data_generation/image_io.py: -------------------------------------------------------------------------------- 1 | """ I/O module 2 | 3 | This unit deals with the nitty gritty of reading in DSLR raw camera and 4 | various other formats. 5 | """ 6 | import numpy as np 7 | import rawpy 8 | 9 | 10 | def read_raw(path, n_bits=None): 11 | with rawpy.imread(path) as raw: 12 | im_ = raw.raw_image_visible.copy() 13 | 14 | # subtract black level 15 | im = np.zeros(im_.shape, dtype='float32') 16 | for i in range(len(raw.black_level_per_channel)): 17 | im += (im_ - raw.black_level_per_channel[i]) * (raw.raw_colors_visible == i).astype('float32') 18 | if n_bits is None: 19 | im /= np.amax(im) 20 | else: 21 | im /= np.power(2, n_bits) 22 | 23 | # shift bayer pattern 24 | red_idx = raw.color_desc.find(b'R') 25 | if red_idx == -1: 26 | print("Warning: Red is not in color description.") 27 | red_idx = 0 28 | 29 | raw_pattern = raw.raw_colors_visible[:8, :8].copy() 30 | red_pos = np.asarray(np.where(raw_pattern == red_idx))[:,0] 31 | row_offset = red_pos[0] 32 | # So that we start with GR 33 | col_offset = red_pos[1] + 1 34 | im = im[row_offset:, col_offset:] 35 | return im, \ 36 | raw.rgb_xyz_matrix, \ 37 | raw.camera_whitebalance 38 | -------------------------------------------------------------------------------- /data_generation/kernel.py: -------------------------------------------------------------------------------- 1 | import torch 2 | 3 | 4 | def gausskern1d(sig, sz=None): 5 | """ 1D Gaussian kernel. 6 | 7 | Args: 8 | sz: kernel size. 9 | sig: stdev of the kernel 10 | """ 11 | if sz is None: 12 | sz = int(2*int(sig) + 1) 13 | sz = max(sz, 3) 14 | half_sz = int(sz / 2) 15 | neg_half_sz = half_sz - sz + 1 16 | neg_half_sz = float(neg_half_sz) 17 | half_sz = float(half_sz) 18 | x = torch.linspace(neg_half_sz, half_sz, int(sz)) / sig 19 | x = x ** 2 20 | kern = torch.exp(-x/2.0) 21 | kern = kern / kern.sum() 22 | return kern 23 | 24 | def gausskern2d(sz_x, sig_x, sz_y=None, sig_y=None): 25 | """Returns a 2D Gaussian kernel array. 26 | 27 | Modified from https://stackoverflow.com/questions/29731726/how-to-calculate-a-gaussian-kernel-matrix-efficiently-in-numpy 28 | 29 | Args: 30 | sz_{x,y}: kernel size. 31 | sig_{x,y}: stdev of kernel in each direction 32 | """ 33 | if sz_y is None: 34 | sz_y = sz_x 35 | if sig_y is None: 36 | sig_y = sig_x 37 | 38 | kern1d_x = gausskern1d(sz_x, sig_x) 39 | kern1d_y = gausskern1d(sz_y, sig_y) 40 | kernel_raw = torch.einsum('i,j->ij', kern1d_x, kern1d_y) 41 | # This einsum is equivalent to outer product (no repeated indices). 42 | # For future reference 43 | # kernel_raw = np.sqrt(np.einsum('ij,k', kernel_raw, kern_r)) 44 | kernel = kernel_raw/kernel_raw.sum() 45 | return kernel 46 | -------------------------------------------------------------------------------- /data_generation/pipeline.py: -------------------------------------------------------------------------------- 1 | import torch.nn as nn 2 | import torch 3 | from . import image_processing 4 | 5 | 6 | class ImageDegradationPipeline(nn.Module): 7 | 8 | def __init__(self, configs): 9 | """ Image Degradation Pipeline. 10 | 11 | Args: 12 | configs: list of modules to be implemented and their parameters. 13 | The list should contain tuple of a form (str, dict), 14 | where str indicate module class name (see 15 | image_processing.py), and dict contain the key-value of 16 | the parameter of such module. 17 | """ 18 | super().__init__() 19 | self.initialize_pipeline(configs) 20 | 21 | def initialize_pipeline(self, configs): 22 | pipeline = [] 23 | # initialize module. 24 | for c in configs: 25 | class_ = getattr(image_processing, c[0]) 26 | module = class_(**c[1]) 27 | pipeline.append(module) 28 | self._pipeline = nn.Sequential(*pipeline) 29 | # self._pipeline = tuple(pipeline) 30 | def forward(self, image): 31 | # import torchvision.transforms as transforms 32 | # trans = transforms.ToPILImage() 33 | # for index, func in enumerate(self._pipeline): 34 | # image = func(image) 35 | # # save images 36 | # # image_trans = trans((torch.clamp(image, 0.0, 1.0)).squeeze()) 37 | # # image_trans.save('./train_images/tmp_{:02d}.png'.format(index), quality=100) 38 | # return image 39 | return self._pipeline(image) 40 | 41 | -------------------------------------------------------------------------------- /dataset_specs/data_configspec.conf: -------------------------------------------------------------------------------- 1 | [dataset_configs] 2 | dataset_dir = string(default=None) 3 | burst_length = integer(default=8) 4 | patch_size = integer(default=200) 5 | big_jitter = integer(default=16) 6 | small_jitter = integer(default=2) 7 | down_sample = integer(default=4) 8 | color = boolean(default=False) 9 | 10 | [pipeline_configs] 11 | denoise = boolean(default=None) 12 | demosaick = boolean(default=None) 13 | jpeg_compression = boolean(default=None) 14 | use_unsharp_mask = boolean(default=None) 15 | use_motion_blur = boolean(default=None) 16 | use_chromatic_aberration = boolean(default=None) 17 | denoise_color_only = boolean(default=None) 18 | exposure_correction = boolean(default=None) 19 | quantize = boolean(default=None) 20 | denoise_guide_transform = float(default=None) 21 | denoise_n_iter = integer(default=1) 22 | demosaick_use_median = boolean(default=False) 23 | demosaick_n_iter = integer(default=0) 24 | use_median_denoise = boolean(default=False) 25 | use_wavelet = boolean(default=False) 26 | wavelet_family = string(default="db2") 27 | wavelet_th_method = string(default="BayesShrink") 28 | wavelet_levels = integer(default=None) 29 | bayer_crop_phase = integer(default=None) 30 | use_log_uniform = boolean(default=False) 31 | median_before_bilateral = boolean(default=False) 32 | repair_hotdead_pixel = boolean(default=False) 33 | hot_px_th = float(default=0.2) 34 | [pipeline_param_ranges] 35 | min_exposure_adjustment = float(default=0.0) 36 | max_exposure_adjustment = float(default=0.0) 37 | min_gaussian_noise = float(default=0.0) 38 | max_gaussian_noise = float(default=0.0) 39 | min_poisson_noise = float(default=0.0) 40 | max_poisson_noise = float(default=0.0) 41 | min_motion_blur = float(default=0) 42 | max_motion_blur = float(default=0) 43 | min_chromatic_aberration = float(default=1) 44 | max_chromatic_aberration = float(default=1) 45 | min_jpeg_quality = integer(default=1) 46 | max_jpeg_quality = integer(default=2) 47 | min_denoise_sigma_s = float(default=1) 48 | max_denoise_sigma_s = float(default=1) 49 | min_denoise_sigma_r = float(default=1) 50 | max_denoise_sigma_r = float(default=1) 51 | min_denoise_color_sigma_ratio = float(default=1) 52 | max_denoise_color_sigma_ratio = float(default=1) 53 | min_denoise_color_range_ratio = float(default=1) 54 | max_denoise_color_range_ratio = float(default=1) 55 | min_unsharp_amount = float(default=0) 56 | max_unsharp_amount = float(default=0) 57 | min_denoise_median_sz = integer(default=0) 58 | max_denoise_median_sz = integer(default=1) 59 | min_quantize_bits = integer(default=14) 60 | max_quantize_bits = integer(default=15) 61 | min_wavelet_sigma = float(default=0.003) 62 | max_wavelet_sigma = float(default=0.015) 63 | min_motionblur_th = float(default=0.99) 64 | max_motionblur_th = float(default=0.997) 65 | min_motionblur_boost = float(default=5) 66 | max_motionblur_boost = float(default=100) 67 | min_motionblur_segment = integer(default=1) 68 | max_motionblur_segment = integer(default=4) 69 | -------------------------------------------------------------------------------- /dataset_specs/full_dataset.conf: -------------------------------------------------------------------------------- 1 | [dataset_configs] 2 | dataset_dir = ../dataset/Adobe5K/ 3 | burst_length = 8 4 | patch_size = 128 5 | big_jitter = 16 6 | small_jitter = 2 7 | down_sample = 4 8 | color = True 9 | 10 | # From 6.3, use log-uniform randomization. 11 | [pipeline_configs] 12 | denoise = True 13 | demosaick = True 14 | jpeg_compression = True 15 | use_unsharp_mask = True 16 | use_motion_blur = False 17 | use_chromatic_aberration = False 18 | denoise_color_only = False 19 | exposure_correction = False 20 | quantize = True 21 | denoise_guide_transform = 0.5 22 | use_median_denoise = True 23 | use_wavelet = False 24 | use_log_uniform = True 25 | median_before_bilateral = True 26 | 27 | [pipeline_param_ranges] 28 | min_gaussian_noise = 0.002 29 | max_gaussian_noise = 0.1 30 | min_poisson_noise = 0.02 31 | max_poisson_noise = 0.2 32 | min_jpeg_quality = 4 33 | max_jpeg_quality = 8 34 | min_denoise_sigma_s = 0.25 35 | max_denoise_sigma_s = 1.0 36 | min_denoise_sigma_r = 0.1 37 | max_denoise_sigma_r = 1.0 38 | min_denoise_color_sigma_ratio = 4 39 | max_denoise_color_sigma_ratio = 32 40 | min_denoise_color_range_ratio = 0.1 41 | max_denoise_color_range_ratio = 0.5 42 | min_unsharp_amount = 0.0 43 | max_unsharp_amount = 0.25 44 | min_denoise_median_sz = -1 45 | max_denoise_median_sz = 0 46 | -------------------------------------------------------------------------------- /eval_images/10_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/10_gt.png -------------------------------------------------------------------------------- /eval_images/10_kpn_5x5_38.68dB_0.9798.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/10_kpn_5x5_38.68dB_0.9798.png -------------------------------------------------------------------------------- /eval_images/10_noisy_34.50dB_0.8800.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/10_noisy_34.50dB_0.8800.png -------------------------------------------------------------------------------- /eval_images/11_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/11_gt.png -------------------------------------------------------------------------------- /eval_images/11_kpn_5x5_38.84dB_0.9909.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/11_kpn_5x5_38.84dB_0.9909.png -------------------------------------------------------------------------------- /eval_images/11_noisy_34.61dB_0.8585.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/11_noisy_34.61dB_0.8585.png -------------------------------------------------------------------------------- /eval_images/12_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/12_gt.png -------------------------------------------------------------------------------- /eval_images/12_kpn_5x5_38.04dB_0.9860.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/12_kpn_5x5_38.04dB_0.9860.png -------------------------------------------------------------------------------- /eval_images/12_noisy_34.56dB_0.8722.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/12_noisy_34.56dB_0.8722.png -------------------------------------------------------------------------------- /eval_images/13_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/13_gt.png -------------------------------------------------------------------------------- /eval_images/13_kpn_5x5_38.88dB_0.9890.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/13_kpn_5x5_38.88dB_0.9890.png -------------------------------------------------------------------------------- /eval_images/13_noisy_34.48dB_0.8867.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/13_noisy_34.48dB_0.8867.png -------------------------------------------------------------------------------- /eval_images/14_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/14_gt.png -------------------------------------------------------------------------------- /eval_images/14_kpn_5x5_37.04dB_0.9497.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/14_kpn_5x5_37.04dB_0.9497.png -------------------------------------------------------------------------------- /eval_images/14_noisy_34.69dB_0.8965.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/14_noisy_34.69dB_0.8965.png -------------------------------------------------------------------------------- /eval_images/15_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/15_gt.png -------------------------------------------------------------------------------- /eval_images/15_kpn_5x5_38.31dB_0.9893.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/15_kpn_5x5_38.31dB_0.9893.png -------------------------------------------------------------------------------- /eval_images/15_noisy_34.50dB_0.9219.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/15_noisy_34.50dB_0.9219.png -------------------------------------------------------------------------------- /eval_images/16_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/16_gt.png -------------------------------------------------------------------------------- /eval_images/16_kpn_5x5_37.81dB_0.9653.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/16_kpn_5x5_37.81dB_0.9653.png -------------------------------------------------------------------------------- /eval_images/16_noisy_34.38dB_0.8607.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/16_noisy_34.38dB_0.8607.png -------------------------------------------------------------------------------- /eval_images/17_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/17_gt.png -------------------------------------------------------------------------------- /eval_images/17_kpn_5x5_34.16dB_0.9625.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/17_kpn_5x5_34.16dB_0.9625.png -------------------------------------------------------------------------------- /eval_images/17_noisy_28.32dB_0.7489.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/17_noisy_28.32dB_0.7489.png -------------------------------------------------------------------------------- /eval_images/18_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/18_gt.png -------------------------------------------------------------------------------- /eval_images/18_kpn_5x5_36.83dB_0.9698.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/18_kpn_5x5_36.83dB_0.9698.png -------------------------------------------------------------------------------- /eval_images/18_noisy_28.09dB_0.6567.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/18_noisy_28.09dB_0.6567.png -------------------------------------------------------------------------------- /eval_images/19_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/19_gt.png -------------------------------------------------------------------------------- /eval_images/19_kpn_5x5_39.11dB_0.9738.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/19_kpn_5x5_39.11dB_0.9738.png -------------------------------------------------------------------------------- /eval_images/19_noisy_28.13dB_0.5510.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/19_noisy_28.13dB_0.5510.png -------------------------------------------------------------------------------- /eval_images/1_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/1_gt.png -------------------------------------------------------------------------------- /eval_images/1_kpn.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/1_kpn.png -------------------------------------------------------------------------------- /eval_images/1_kpn_5x5_39.35dB_0.9866.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/1_kpn_5x5_39.35dB_0.9866.png -------------------------------------------------------------------------------- /eval_images/1_noisy.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/1_noisy.png -------------------------------------------------------------------------------- /eval_images/1_noisy_36.73dB_0.9366.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/1_noisy_36.73dB_0.9366.png -------------------------------------------------------------------------------- /eval_images/20_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/20_gt.png -------------------------------------------------------------------------------- /eval_images/20_kpn_5x5_33.79dB_0.9656.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/20_kpn_5x5_33.79dB_0.9656.png -------------------------------------------------------------------------------- /eval_images/20_noisy_28.72dB_0.7307.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/20_noisy_28.72dB_0.7307.png -------------------------------------------------------------------------------- /eval_images/21_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/21_gt.png -------------------------------------------------------------------------------- /eval_images/21_kpn.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/21_kpn.png -------------------------------------------------------------------------------- /eval_images/21_kpn_5x5_37.36dB_0.9316.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/21_kpn_5x5_37.36dB_0.9316.png -------------------------------------------------------------------------------- /eval_images/21_noisy.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/21_noisy.png -------------------------------------------------------------------------------- /eval_images/21_noisy_27.61dB_0.5016.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/21_noisy_27.61dB_0.5016.png -------------------------------------------------------------------------------- /eval_images/22_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/22_gt.png -------------------------------------------------------------------------------- /eval_images/22_kpn.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/22_kpn.png -------------------------------------------------------------------------------- /eval_images/22_kpn_5x5_33.98dB_0.9473.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/22_kpn_5x5_33.98dB_0.9473.png -------------------------------------------------------------------------------- /eval_images/22_noisy.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/22_noisy.png -------------------------------------------------------------------------------- /eval_images/22_noisy_28.10dB_0.7991.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/22_noisy_28.10dB_0.7991.png -------------------------------------------------------------------------------- /eval_images/23_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/23_gt.png -------------------------------------------------------------------------------- /eval_images/23_kpn.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/23_kpn.png -------------------------------------------------------------------------------- /eval_images/23_kpn_5x5_35.85dB_0.9759.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/23_kpn_5x5_35.85dB_0.9759.png -------------------------------------------------------------------------------- /eval_images/23_noisy.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/23_noisy.png -------------------------------------------------------------------------------- /eval_images/23_noisy_28.39dB_0.6966.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/23_noisy_28.39dB_0.6966.png -------------------------------------------------------------------------------- /eval_images/24_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/24_gt.png -------------------------------------------------------------------------------- /eval_images/24_kpn.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/24_kpn.png -------------------------------------------------------------------------------- /eval_images/24_kpn_5x5_34.84dB_0.9692.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/24_kpn_5x5_34.84dB_0.9692.png -------------------------------------------------------------------------------- /eval_images/24_noisy.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/24_noisy.png -------------------------------------------------------------------------------- /eval_images/24_noisy_28.60dB_0.6402.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/24_noisy_28.60dB_0.6402.png -------------------------------------------------------------------------------- /eval_images/25_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/25_gt.png -------------------------------------------------------------------------------- /eval_images/25_kpn_5x5_40.11dB_0.9853.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/25_kpn_5x5_40.11dB_0.9853.png -------------------------------------------------------------------------------- /eval_images/25_noisy_29.38dB_0.5669.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/25_noisy_29.38dB_0.5669.png -------------------------------------------------------------------------------- /eval_images/26_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/26_gt.png -------------------------------------------------------------------------------- /eval_images/26_kpn_5x5_32.85dB_0.9452.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/26_kpn_5x5_32.85dB_0.9452.png -------------------------------------------------------------------------------- /eval_images/26_noisy_26.66dB_0.7940.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/26_noisy_26.66dB_0.7940.png -------------------------------------------------------------------------------- /eval_images/27_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/27_gt.png -------------------------------------------------------------------------------- /eval_images/27_kpn_5x5_35.52dB_0.9546.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/27_kpn_5x5_35.52dB_0.9546.png -------------------------------------------------------------------------------- /eval_images/27_noisy_27.42dB_0.7241.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/27_noisy_27.42dB_0.7241.png -------------------------------------------------------------------------------- /eval_images/28_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/28_gt.png -------------------------------------------------------------------------------- /eval_images/28_kpn_5x5_36.13dB_0.9708.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/28_kpn_5x5_36.13dB_0.9708.png -------------------------------------------------------------------------------- /eval_images/28_noisy_28.70dB_0.6872.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/28_noisy_28.70dB_0.6872.png -------------------------------------------------------------------------------- /eval_images/29_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/29_gt.png -------------------------------------------------------------------------------- /eval_images/29_kpn_5x5_35.68dB_0.9565.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/29_kpn_5x5_35.68dB_0.9565.png -------------------------------------------------------------------------------- /eval_images/29_noisy_27.39dB_0.7316.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/29_noisy_27.39dB_0.7316.png -------------------------------------------------------------------------------- /eval_images/2_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/2_gt.png -------------------------------------------------------------------------------- /eval_images/2_kpn.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/2_kpn.png -------------------------------------------------------------------------------- /eval_images/2_kpn_5x5_39.48dB_0.9919.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/2_kpn_5x5_39.48dB_0.9919.png -------------------------------------------------------------------------------- /eval_images/2_noisy.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/2_noisy.png -------------------------------------------------------------------------------- /eval_images/2_noisy_37.73dB_0.9347.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/2_noisy_37.73dB_0.9347.png -------------------------------------------------------------------------------- /eval_images/30_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/30_gt.png -------------------------------------------------------------------------------- /eval_images/30_kpn_5x5_37.83dB_0.9713.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/30_kpn_5x5_37.83dB_0.9713.png -------------------------------------------------------------------------------- /eval_images/30_noisy_26.21dB_0.6279.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/30_noisy_26.21dB_0.6279.png -------------------------------------------------------------------------------- /eval_images/31_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/31_gt.png -------------------------------------------------------------------------------- /eval_images/31_kpn_5x5_37.32dB_0.9682.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/31_kpn_5x5_37.32dB_0.9682.png -------------------------------------------------------------------------------- /eval_images/31_noisy_28.10dB_0.6873.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/31_noisy_28.10dB_0.6873.png -------------------------------------------------------------------------------- /eval_images/32_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/32_gt.png -------------------------------------------------------------------------------- /eval_images/32_kpn_5x5_37.64dB_0.9700.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/32_kpn_5x5_37.64dB_0.9700.png -------------------------------------------------------------------------------- /eval_images/32_noisy_32.31dB_0.7877.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/32_noisy_32.31dB_0.7877.png -------------------------------------------------------------------------------- /eval_images/33_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/33_gt.png -------------------------------------------------------------------------------- /eval_images/33_kpn_5x5_39.45dB_0.9817.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/33_kpn_5x5_39.45dB_0.9817.png -------------------------------------------------------------------------------- /eval_images/33_noisy_32.41dB_0.7184.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/33_noisy_32.41dB_0.7184.png -------------------------------------------------------------------------------- /eval_images/34_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/34_gt.png -------------------------------------------------------------------------------- /eval_images/34_kpn_5x5_38.20dB_0.9803.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/34_kpn_5x5_38.20dB_0.9803.png -------------------------------------------------------------------------------- /eval_images/34_noisy_32.51dB_0.7680.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/34_noisy_32.51dB_0.7680.png -------------------------------------------------------------------------------- /eval_images/35_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/35_gt.png -------------------------------------------------------------------------------- /eval_images/35_kpn_5x5_39.29dB_0.9842.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/35_kpn_5x5_39.29dB_0.9842.png -------------------------------------------------------------------------------- /eval_images/35_noisy_32.38dB_0.8153.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/35_noisy_32.38dB_0.8153.png -------------------------------------------------------------------------------- /eval_images/36_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/36_gt.png -------------------------------------------------------------------------------- /eval_images/36_kpn_5x5_37.80dB_0.9754.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/36_kpn_5x5_37.80dB_0.9754.png -------------------------------------------------------------------------------- /eval_images/36_noisy_32.38dB_0.8101.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/36_noisy_32.38dB_0.8101.png -------------------------------------------------------------------------------- /eval_images/37_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/37_gt.png -------------------------------------------------------------------------------- /eval_images/37_kpn_5x5_37.77dB_0.9826.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/37_kpn_5x5_37.77dB_0.9826.png -------------------------------------------------------------------------------- /eval_images/37_noisy_32.21dB_0.8376.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/37_noisy_32.21dB_0.8376.png -------------------------------------------------------------------------------- /eval_images/38_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/38_gt.png -------------------------------------------------------------------------------- /eval_images/38_kpn_5x5_36.79dB_0.9648.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/38_kpn_5x5_36.79dB_0.9648.png -------------------------------------------------------------------------------- /eval_images/38_noisy_32.33dB_0.8203.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/38_noisy_32.33dB_0.8203.png -------------------------------------------------------------------------------- /eval_images/39_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/39_gt.png -------------------------------------------------------------------------------- /eval_images/39_kpn_5x5_36.01dB_0.9710.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/39_kpn_5x5_36.01dB_0.9710.png -------------------------------------------------------------------------------- /eval_images/39_noisy_27.63dB_0.7201.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/39_noisy_27.63dB_0.7201.png -------------------------------------------------------------------------------- /eval_images/3_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/3_gt.png -------------------------------------------------------------------------------- /eval_images/3_kpn.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/3_kpn.png -------------------------------------------------------------------------------- /eval_images/3_kpn_5x5_39.14dB_0.9938.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/3_kpn_5x5_39.14dB_0.9938.png -------------------------------------------------------------------------------- /eval_images/3_noisy.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/3_noisy.png -------------------------------------------------------------------------------- /eval_images/3_noisy_37.67dB_0.9583.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/3_noisy_37.67dB_0.9583.png -------------------------------------------------------------------------------- /eval_images/40_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/40_gt.png -------------------------------------------------------------------------------- /eval_images/40_kpn_5x5_38.57dB_0.9918.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/40_kpn_5x5_38.57dB_0.9918.png -------------------------------------------------------------------------------- /eval_images/40_noisy_36.00dB_0.9021.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/40_noisy_36.00dB_0.9021.png -------------------------------------------------------------------------------- /eval_images/41_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/41_gt.png -------------------------------------------------------------------------------- /eval_images/41_kpn_5x5_38.56dB_0.9867.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/41_kpn_5x5_38.56dB_0.9867.png -------------------------------------------------------------------------------- /eval_images/41_noisy_36.12dB_0.9290.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/41_noisy_36.12dB_0.9290.png -------------------------------------------------------------------------------- /eval_images/42_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/42_gt.png -------------------------------------------------------------------------------- /eval_images/42_kpn_5x5_39.17dB_0.9878.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/42_kpn_5x5_39.17dB_0.9878.png -------------------------------------------------------------------------------- /eval_images/42_noisy_36.09dB_0.9311.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/42_noisy_36.09dB_0.9311.png -------------------------------------------------------------------------------- /eval_images/43_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/43_gt.png -------------------------------------------------------------------------------- /eval_images/43_kpn_5x5_39.89dB_0.9728.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/43_kpn_5x5_39.89dB_0.9728.png -------------------------------------------------------------------------------- /eval_images/43_noisy_36.18dB_0.8960.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/43_noisy_36.18dB_0.8960.png -------------------------------------------------------------------------------- /eval_images/44_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/44_gt.png -------------------------------------------------------------------------------- /eval_images/44_kpn_5x5_37.81dB_0.9755.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/44_kpn_5x5_37.81dB_0.9755.png -------------------------------------------------------------------------------- /eval_images/44_noisy_36.06dB_0.9542.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/44_noisy_36.06dB_0.9542.png -------------------------------------------------------------------------------- /eval_images/45_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/45_gt.png -------------------------------------------------------------------------------- /eval_images/45_kpn_5x5_37.07dB_0.9824.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/45_kpn_5x5_37.07dB_0.9824.png -------------------------------------------------------------------------------- /eval_images/45_noisy_36.21dB_0.9775.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/45_noisy_36.21dB_0.9775.png -------------------------------------------------------------------------------- /eval_images/46_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/46_gt.png -------------------------------------------------------------------------------- /eval_images/46_kpn_5x5_41.12dB_0.9823.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/46_kpn_5x5_41.12dB_0.9823.png -------------------------------------------------------------------------------- /eval_images/46_noisy_31.75dB_0.7773.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/46_noisy_31.75dB_0.7773.png -------------------------------------------------------------------------------- /eval_images/47_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/47_gt.png -------------------------------------------------------------------------------- /eval_images/47_kpn_5x5_38.98dB_0.9849.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/47_kpn_5x5_38.98dB_0.9849.png -------------------------------------------------------------------------------- /eval_images/47_noisy_35.49dB_0.9071.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/47_noisy_35.49dB_0.9071.png -------------------------------------------------------------------------------- /eval_images/48_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/48_gt.png -------------------------------------------------------------------------------- /eval_images/48_kpn_5x5_37.68dB_0.9244.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/48_kpn_5x5_37.68dB_0.9244.png -------------------------------------------------------------------------------- /eval_images/48_noisy_30.25dB_0.7162.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/48_noisy_30.25dB_0.7162.png -------------------------------------------------------------------------------- /eval_images/49_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/49_gt.png -------------------------------------------------------------------------------- /eval_images/49_kpn_5x5_30.79dB_0.7580.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/49_kpn_5x5_30.79dB_0.7580.png -------------------------------------------------------------------------------- /eval_images/49_noisy_26.10dB_0.5892.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/49_noisy_26.10dB_0.5892.png -------------------------------------------------------------------------------- /eval_images/4_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/4_gt.png -------------------------------------------------------------------------------- /eval_images/4_kpn.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/4_kpn.png -------------------------------------------------------------------------------- /eval_images/4_kpn_5x5_39.25dB_0.9824.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/4_kpn_5x5_39.25dB_0.9824.png -------------------------------------------------------------------------------- /eval_images/4_noisy.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/4_noisy.png -------------------------------------------------------------------------------- /eval_images/4_noisy_37.99dB_0.9598.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/4_noisy_37.99dB_0.9598.png -------------------------------------------------------------------------------- /eval_images/50_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/50_gt.png -------------------------------------------------------------------------------- /eval_images/50_kpn_5x5_39.02dB_0.9887.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/50_kpn_5x5_39.02dB_0.9887.png -------------------------------------------------------------------------------- /eval_images/50_noisy_36.48dB_0.9212.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/50_noisy_36.48dB_0.9212.png -------------------------------------------------------------------------------- /eval_images/51_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/51_gt.png -------------------------------------------------------------------------------- /eval_images/51_kpn_5x5_39.68dB_0.9903.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/51_kpn_5x5_39.68dB_0.9903.png -------------------------------------------------------------------------------- /eval_images/51_noisy_36.14dB_0.9540.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/51_noisy_36.14dB_0.9540.png -------------------------------------------------------------------------------- /eval_images/52_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/52_gt.png -------------------------------------------------------------------------------- /eval_images/52_kpn_5x5_39.06dB_0.9819.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/52_kpn_5x5_39.06dB_0.9819.png -------------------------------------------------------------------------------- /eval_images/52_noisy_36.34dB_0.9172.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/52_noisy_36.34dB_0.9172.png -------------------------------------------------------------------------------- /eval_images/53_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/53_gt.png -------------------------------------------------------------------------------- /eval_images/53_kpn_5x5_39.40dB_0.9871.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/53_kpn_5x5_39.40dB_0.9871.png -------------------------------------------------------------------------------- /eval_images/53_noisy_34.24dB_0.9062.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/53_noisy_34.24dB_0.9062.png -------------------------------------------------------------------------------- /eval_images/54_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/54_gt.png -------------------------------------------------------------------------------- /eval_images/54_kpn_5x5_39.98dB_0.9886.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/54_kpn_5x5_39.98dB_0.9886.png -------------------------------------------------------------------------------- /eval_images/54_noisy_36.25dB_0.9036.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/54_noisy_36.25dB_0.9036.png -------------------------------------------------------------------------------- /eval_images/55_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/55_gt.png -------------------------------------------------------------------------------- /eval_images/55_kpn_5x5_38.73dB_0.9822.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/55_kpn_5x5_38.73dB_0.9822.png -------------------------------------------------------------------------------- /eval_images/55_noisy_37.04dB_0.9654.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/55_noisy_37.04dB_0.9654.png -------------------------------------------------------------------------------- /eval_images/56_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/56_gt.png -------------------------------------------------------------------------------- /eval_images/56_kpn_5x5_39.51dB_0.9936.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/56_kpn_5x5_39.51dB_0.9936.png -------------------------------------------------------------------------------- /eval_images/56_noisy_38.16dB_0.9714.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/56_noisy_38.16dB_0.9714.png -------------------------------------------------------------------------------- /eval_images/57_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/57_gt.png -------------------------------------------------------------------------------- /eval_images/57_kpn_5x5_37.89dB_0.9887.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/57_kpn_5x5_37.89dB_0.9887.png -------------------------------------------------------------------------------- /eval_images/57_noisy_35.77dB_0.9747.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/57_noisy_35.77dB_0.9747.png -------------------------------------------------------------------------------- /eval_images/58_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/58_gt.png -------------------------------------------------------------------------------- /eval_images/58_kpn_5x5_39.52dB_0.9840.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/58_kpn_5x5_39.52dB_0.9840.png -------------------------------------------------------------------------------- /eval_images/58_noisy_38.25dB_0.9738.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/58_noisy_38.25dB_0.9738.png -------------------------------------------------------------------------------- /eval_images/59_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/59_gt.png -------------------------------------------------------------------------------- /eval_images/59_kpn_5x5_38.02dB_0.9827.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/59_kpn_5x5_38.02dB_0.9827.png -------------------------------------------------------------------------------- /eval_images/59_noisy_33.73dB_0.9389.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/59_noisy_33.73dB_0.9389.png -------------------------------------------------------------------------------- /eval_images/5_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/5_gt.png -------------------------------------------------------------------------------- /eval_images/5_kpn.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/5_kpn.png -------------------------------------------------------------------------------- /eval_images/5_kpn_5x5_39.33dB_0.9888.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/5_kpn_5x5_39.33dB_0.9888.png -------------------------------------------------------------------------------- /eval_images/5_noisy.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/5_noisy.png -------------------------------------------------------------------------------- /eval_images/5_noisy_37.77dB_0.9749.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/5_noisy_37.77dB_0.9749.png -------------------------------------------------------------------------------- /eval_images/60_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/60_gt.png -------------------------------------------------------------------------------- /eval_images/60_kpn_5x5_38.65dB_0.9853.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/60_kpn_5x5_38.65dB_0.9853.png -------------------------------------------------------------------------------- /eval_images/60_noisy_37.21dB_0.9718.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/60_noisy_37.21dB_0.9718.png -------------------------------------------------------------------------------- /eval_images/61_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/61_gt.png -------------------------------------------------------------------------------- /eval_images/61_kpn_5x5_40.22dB_0.9922.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/61_kpn_5x5_40.22dB_0.9922.png -------------------------------------------------------------------------------- /eval_images/61_noisy_37.93dB_0.9604.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/61_noisy_37.93dB_0.9604.png -------------------------------------------------------------------------------- /eval_images/62_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/62_gt.png -------------------------------------------------------------------------------- /eval_images/62_kpn_5x5_36.89dB_0.9901.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/62_kpn_5x5_36.89dB_0.9901.png -------------------------------------------------------------------------------- /eval_images/62_noisy_34.35dB_0.9645.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/62_noisy_34.35dB_0.9645.png -------------------------------------------------------------------------------- /eval_images/63_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/63_gt.png -------------------------------------------------------------------------------- /eval_images/63_kpn_5x5_34.38dB_0.9104.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/63_kpn_5x5_34.38dB_0.9104.png -------------------------------------------------------------------------------- /eval_images/63_noisy_25.35dB_0.4702.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/63_noisy_25.35dB_0.4702.png -------------------------------------------------------------------------------- /eval_images/64_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/64_gt.png -------------------------------------------------------------------------------- /eval_images/64_kpn_5x5_35.10dB_0.9525.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/64_kpn_5x5_35.10dB_0.9525.png -------------------------------------------------------------------------------- /eval_images/64_noisy_25.87dB_0.5550.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/64_noisy_25.87dB_0.5550.png -------------------------------------------------------------------------------- /eval_images/65_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/65_gt.png -------------------------------------------------------------------------------- /eval_images/65_kpn_5x5_34.76dB_0.9458.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/65_kpn_5x5_34.76dB_0.9458.png -------------------------------------------------------------------------------- /eval_images/65_noisy_25.34dB_0.5893.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/65_noisy_25.34dB_0.5893.png -------------------------------------------------------------------------------- /eval_images/66_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/66_gt.png -------------------------------------------------------------------------------- /eval_images/66_kpn_5x5_34.08dB_0.8902.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/66_kpn_5x5_34.08dB_0.8902.png -------------------------------------------------------------------------------- /eval_images/66_noisy_25.16dB_0.5120.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/66_noisy_25.16dB_0.5120.png -------------------------------------------------------------------------------- /eval_images/67_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/67_gt.png -------------------------------------------------------------------------------- /eval_images/67_kpn_5x5_39.37dB_0.9908.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/67_kpn_5x5_39.37dB_0.9908.png -------------------------------------------------------------------------------- /eval_images/67_noisy_38.30dB_0.9847.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/67_noisy_38.30dB_0.9847.png -------------------------------------------------------------------------------- /eval_images/68_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/68_gt.png -------------------------------------------------------------------------------- /eval_images/68_kpn_5x5_35.75dB_0.9324.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/68_kpn_5x5_35.75dB_0.9324.png -------------------------------------------------------------------------------- /eval_images/68_noisy_25.69dB_0.4059.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/68_noisy_25.69dB_0.4059.png -------------------------------------------------------------------------------- /eval_images/69_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/69_gt.png -------------------------------------------------------------------------------- /eval_images/69_kpn_5x5_32.32dB_0.9361.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/69_kpn_5x5_32.32dB_0.9361.png -------------------------------------------------------------------------------- /eval_images/69_noisy_25.68dB_0.7298.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/69_noisy_25.68dB_0.7298.png -------------------------------------------------------------------------------- /eval_images/6_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/6_gt.png -------------------------------------------------------------------------------- /eval_images/6_kpn_5x5_38.36dB_0.9825.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/6_kpn_5x5_38.36dB_0.9825.png -------------------------------------------------------------------------------- /eval_images/6_noisy_36.65dB_0.9624.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/6_noisy_36.65dB_0.9624.png -------------------------------------------------------------------------------- /eval_images/70_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/70_gt.png -------------------------------------------------------------------------------- /eval_images/70_kpn_5x5_39.91dB_0.9910.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/70_kpn_5x5_39.91dB_0.9910.png -------------------------------------------------------------------------------- /eval_images/70_noisy_38.50dB_0.9572.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/70_noisy_38.50dB_0.9572.png -------------------------------------------------------------------------------- /eval_images/71_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/71_gt.png -------------------------------------------------------------------------------- /eval_images/71_kpn_5x5_35.55dB_0.9774.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/71_kpn_5x5_35.55dB_0.9774.png -------------------------------------------------------------------------------- /eval_images/71_noisy_32.97dB_0.8975.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/71_noisy_32.97dB_0.8975.png -------------------------------------------------------------------------------- /eval_images/72_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/72_gt.png -------------------------------------------------------------------------------- /eval_images/72_kpn_5x5_37.08dB_0.9755.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/72_kpn_5x5_37.08dB_0.9755.png -------------------------------------------------------------------------------- /eval_images/72_noisy_32.98dB_0.8527.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/72_noisy_32.98dB_0.8527.png -------------------------------------------------------------------------------- /eval_images/73_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/73_gt.png -------------------------------------------------------------------------------- /eval_images/73_kpn_5x5_39.25dB_0.9713.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/73_kpn_5x5_39.25dB_0.9713.png -------------------------------------------------------------------------------- /eval_images/73_noisy_31.69dB_0.7565.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/73_noisy_31.69dB_0.7565.png -------------------------------------------------------------------------------- /eval_images/74_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/74_gt.png -------------------------------------------------------------------------------- /eval_images/74_kpn_5x5_37.40dB_0.9598.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/74_kpn_5x5_37.40dB_0.9598.png -------------------------------------------------------------------------------- /eval_images/74_noisy_31.40dB_0.7913.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/74_noisy_31.40dB_0.7913.png -------------------------------------------------------------------------------- /eval_images/75_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/75_gt.png -------------------------------------------------------------------------------- /eval_images/75_kpn_5x5_34.89dB_0.9287.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/75_kpn_5x5_34.89dB_0.9287.png -------------------------------------------------------------------------------- /eval_images/75_noisy_25.27dB_0.6376.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/75_noisy_25.27dB_0.6376.png -------------------------------------------------------------------------------- /eval_images/76_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/76_gt.png -------------------------------------------------------------------------------- /eval_images/76_kpn_5x5_38.18dB_0.9756.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/76_kpn_5x5_38.18dB_0.9756.png -------------------------------------------------------------------------------- /eval_images/76_noisy_32.34dB_0.8855.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/76_noisy_32.34dB_0.8855.png -------------------------------------------------------------------------------- /eval_images/77_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/77_gt.png -------------------------------------------------------------------------------- /eval_images/77_kpn_5x5_36.31dB_0.9730.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/77_kpn_5x5_36.31dB_0.9730.png -------------------------------------------------------------------------------- /eval_images/77_noisy_32.73dB_0.9155.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/77_noisy_32.73dB_0.9155.png -------------------------------------------------------------------------------- /eval_images/78_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/78_gt.png -------------------------------------------------------------------------------- /eval_images/78_kpn_5x5_38.33dB_0.9751.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/78_kpn_5x5_38.33dB_0.9751.png -------------------------------------------------------------------------------- /eval_images/78_noisy_25.95dB_0.4840.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/78_noisy_25.95dB_0.4840.png -------------------------------------------------------------------------------- /eval_images/79_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/79_gt.png -------------------------------------------------------------------------------- /eval_images/79_kpn_5x5_37.68dB_0.9554.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/79_kpn_5x5_37.68dB_0.9554.png -------------------------------------------------------------------------------- /eval_images/79_noisy_32.21dB_0.7821.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/79_noisy_32.21dB_0.7821.png -------------------------------------------------------------------------------- /eval_images/7_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/7_gt.png -------------------------------------------------------------------------------- /eval_images/7_kpn_5x5_40.65dB_0.9868.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/7_kpn_5x5_40.65dB_0.9868.png -------------------------------------------------------------------------------- /eval_images/7_noisy_37.47dB_0.9308.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/7_noisy_37.47dB_0.9308.png -------------------------------------------------------------------------------- /eval_images/80_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/80_gt.png -------------------------------------------------------------------------------- /eval_images/80_kpn_5x5_36.39dB_0.9837.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/80_kpn_5x5_36.39dB_0.9837.png -------------------------------------------------------------------------------- /eval_images/80_noisy_32.55dB_0.9169.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/80_noisy_32.55dB_0.9169.png -------------------------------------------------------------------------------- /eval_images/81_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/81_gt.png -------------------------------------------------------------------------------- /eval_images/81_kpn_5x5_37.33dB_0.9816.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/81_kpn_5x5_37.33dB_0.9816.png -------------------------------------------------------------------------------- /eval_images/81_noisy_32.67dB_0.8759.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/81_noisy_32.67dB_0.8759.png -------------------------------------------------------------------------------- /eval_images/82_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/82_gt.png -------------------------------------------------------------------------------- /eval_images/82_kpn_5x5_36.23dB_0.9788.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/82_kpn_5x5_36.23dB_0.9788.png -------------------------------------------------------------------------------- /eval_images/82_noisy_32.49dB_0.8655.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/82_noisy_32.49dB_0.8655.png -------------------------------------------------------------------------------- /eval_images/83_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/83_gt.png -------------------------------------------------------------------------------- /eval_images/83_kpn_5x5_36.86dB_0.9658.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/83_kpn_5x5_36.86dB_0.9658.png -------------------------------------------------------------------------------- /eval_images/83_noisy_31.89dB_0.8717.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/83_noisy_31.89dB_0.8717.png -------------------------------------------------------------------------------- /eval_images/84_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/84_gt.png -------------------------------------------------------------------------------- /eval_images/84_kpn_5x5_37.53dB_0.9908.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/84_kpn_5x5_37.53dB_0.9908.png -------------------------------------------------------------------------------- /eval_images/84_noisy_32.70dB_0.8561.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/84_noisy_32.70dB_0.8561.png -------------------------------------------------------------------------------- /eval_images/85_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/85_gt.png -------------------------------------------------------------------------------- /eval_images/85_kpn_5x5_37.93dB_0.9557.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/85_kpn_5x5_37.93dB_0.9557.png -------------------------------------------------------------------------------- /eval_images/85_noisy_32.45dB_0.7962.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/85_noisy_32.45dB_0.7962.png -------------------------------------------------------------------------------- /eval_images/86_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/86_gt.png -------------------------------------------------------------------------------- /eval_images/86_kpn_5x5_34.20dB_0.9634.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/86_kpn_5x5_34.20dB_0.9634.png -------------------------------------------------------------------------------- /eval_images/86_noisy_30.81dB_0.9148.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/86_noisy_30.81dB_0.9148.png -------------------------------------------------------------------------------- /eval_images/87_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/87_gt.png -------------------------------------------------------------------------------- /eval_images/87_kpn_5x5_39.10dB_0.9747.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/87_kpn_5x5_39.10dB_0.9747.png -------------------------------------------------------------------------------- /eval_images/87_noisy_30.31dB_0.5805.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/87_noisy_30.31dB_0.5805.png -------------------------------------------------------------------------------- /eval_images/88_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/88_gt.png -------------------------------------------------------------------------------- /eval_images/88_kpn_5x5_37.34dB_0.9650.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/88_kpn_5x5_37.34dB_0.9650.png -------------------------------------------------------------------------------- /eval_images/88_noisy_29.50dB_0.6900.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/88_noisy_29.50dB_0.6900.png -------------------------------------------------------------------------------- /eval_images/89_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/89_gt.png -------------------------------------------------------------------------------- /eval_images/89_kpn_5x5_36.06dB_0.9816.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/89_kpn_5x5_36.06dB_0.9816.png -------------------------------------------------------------------------------- /eval_images/89_noisy_30.16dB_0.7799.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/89_noisy_30.16dB_0.7799.png -------------------------------------------------------------------------------- /eval_images/8_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/8_gt.png -------------------------------------------------------------------------------- /eval_images/8_kpn_5x5_38.48dB_0.9800.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/8_kpn_5x5_38.48dB_0.9800.png -------------------------------------------------------------------------------- /eval_images/8_noisy_35.23dB_0.9459.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/8_noisy_35.23dB_0.9459.png -------------------------------------------------------------------------------- /eval_images/90_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/90_gt.png -------------------------------------------------------------------------------- /eval_images/90_kpn_5x5_34.73dB_0.9713.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/90_kpn_5x5_34.73dB_0.9713.png -------------------------------------------------------------------------------- /eval_images/90_noisy_29.64dB_0.8340.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/90_noisy_29.64dB_0.8340.png -------------------------------------------------------------------------------- /eval_images/91_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/91_gt.png -------------------------------------------------------------------------------- /eval_images/91_kpn_5x5_39.03dB_0.9762.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/91_kpn_5x5_39.03dB_0.9762.png -------------------------------------------------------------------------------- /eval_images/91_noisy_32.42dB_0.7615.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/91_noisy_32.42dB_0.7615.png -------------------------------------------------------------------------------- /eval_images/92_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/92_gt.png -------------------------------------------------------------------------------- /eval_images/92_kpn_5x5_38.21dB_0.9746.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/92_kpn_5x5_38.21dB_0.9746.png -------------------------------------------------------------------------------- /eval_images/92_noisy_30.11dB_0.7182.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/92_noisy_30.11dB_0.7182.png -------------------------------------------------------------------------------- /eval_images/93_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/93_gt.png -------------------------------------------------------------------------------- /eval_images/93_kpn_5x5_36.61dB_0.9652.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/93_kpn_5x5_36.61dB_0.9652.png -------------------------------------------------------------------------------- /eval_images/93_noisy_29.85dB_0.6761.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/93_noisy_29.85dB_0.6761.png -------------------------------------------------------------------------------- /eval_images/94_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/94_gt.png -------------------------------------------------------------------------------- /eval_images/94_kpn_5x5_37.38dB_0.9859.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/94_kpn_5x5_37.38dB_0.9859.png -------------------------------------------------------------------------------- /eval_images/94_noisy_32.82dB_0.8374.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/94_noisy_32.82dB_0.8374.png -------------------------------------------------------------------------------- /eval_images/95_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/95_gt.png -------------------------------------------------------------------------------- /eval_images/95_kpn_5x5_37.35dB_0.9740.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/95_kpn_5x5_37.35dB_0.9740.png -------------------------------------------------------------------------------- /eval_images/95_noisy_32.39dB_0.8629.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/95_noisy_32.39dB_0.8629.png -------------------------------------------------------------------------------- /eval_images/96_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/96_gt.png -------------------------------------------------------------------------------- /eval_images/96_kpn_5x5_35.64dB_0.9681.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/96_kpn_5x5_35.64dB_0.9681.png -------------------------------------------------------------------------------- /eval_images/96_noisy_29.94dB_0.7946.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/96_noisy_29.94dB_0.7946.png -------------------------------------------------------------------------------- /eval_images/97_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/97_gt.png -------------------------------------------------------------------------------- /eval_images/97_kpn_5x5_36.06dB_0.9525.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/97_kpn_5x5_36.06dB_0.9525.png -------------------------------------------------------------------------------- /eval_images/97_noisy_32.48dB_0.8883.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/97_noisy_32.48dB_0.8883.png -------------------------------------------------------------------------------- /eval_images/98_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/98_gt.png -------------------------------------------------------------------------------- /eval_images/98_kpn_5x5_36.41dB_0.9654.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/98_kpn_5x5_36.41dB_0.9654.png -------------------------------------------------------------------------------- /eval_images/98_noisy_31.56dB_0.8672.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/98_noisy_31.56dB_0.8672.png -------------------------------------------------------------------------------- /eval_images/99_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/99_gt.png -------------------------------------------------------------------------------- /eval_images/99_kpn_5x5_38.40dB_0.9825.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/99_kpn_5x5_38.40dB_0.9825.png -------------------------------------------------------------------------------- /eval_images/99_noisy_32.58dB_0.8636.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/99_noisy_32.58dB_0.8636.png -------------------------------------------------------------------------------- /eval_images/9_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/9_gt.png -------------------------------------------------------------------------------- /eval_images/9_kpn_5x5_37.53dB_0.9877.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/9_kpn_5x5_37.53dB_0.9877.png -------------------------------------------------------------------------------- /eval_images/9_noisy_34.32dB_0.8747.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images/9_noisy_34.32dB_0.8747.png -------------------------------------------------------------------------------- /eval_images_RGB/0_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/0_gt.png -------------------------------------------------------------------------------- /eval_images_RGB/0_noisy_29.17dB_0.911.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/0_noisy_29.17dB_0.911.png -------------------------------------------------------------------------------- /eval_images_RGB/0_pred_31.86dB_0.957.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/0_pred_31.86dB_0.957.png -------------------------------------------------------------------------------- /eval_images_RGB/10_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/10_gt.png -------------------------------------------------------------------------------- /eval_images_RGB/10_noisy_31.95dB_0.776.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/10_noisy_31.95dB_0.776.png -------------------------------------------------------------------------------- /eval_images_RGB/10_pred_39.27dB_0.965.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/10_pred_39.27dB_0.965.png -------------------------------------------------------------------------------- /eval_images_RGB/11_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/11_gt.png -------------------------------------------------------------------------------- /eval_images_RGB/11_noisy_23.93dB_0.549.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/11_noisy_23.93dB_0.549.png -------------------------------------------------------------------------------- /eval_images_RGB/11_pred_29.44dB_0.801.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/11_pred_29.44dB_0.801.png -------------------------------------------------------------------------------- /eval_images_RGB/12_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/12_gt.png -------------------------------------------------------------------------------- /eval_images_RGB/12_noisy_18.70dB_0.308.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/12_noisy_18.70dB_0.308.png -------------------------------------------------------------------------------- /eval_images_RGB/12_pred_34.02dB_0.954.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/12_pred_34.02dB_0.954.png -------------------------------------------------------------------------------- /eval_images_RGB/13_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/13_gt.png -------------------------------------------------------------------------------- /eval_images_RGB/13_noisy_32.41dB_0.916.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/13_noisy_32.41dB_0.916.png -------------------------------------------------------------------------------- /eval_images_RGB/13_pred_37.87dB_0.987.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/13_pred_37.87dB_0.987.png -------------------------------------------------------------------------------- /eval_images_RGB/1_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/1_gt.png -------------------------------------------------------------------------------- /eval_images_RGB/1_noisy_22.25dB_0.544.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/1_noisy_22.25dB_0.544.png -------------------------------------------------------------------------------- /eval_images_RGB/1_pred_32.77dB_0.912.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/1_pred_32.77dB_0.912.png -------------------------------------------------------------------------------- /eval_images_RGB/2_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/2_gt.png -------------------------------------------------------------------------------- /eval_images_RGB/2_noisy_19.42dB_0.496.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/2_noisy_19.42dB_0.496.png -------------------------------------------------------------------------------- /eval_images_RGB/2_pred_29.73dB_0.900.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/2_pred_29.73dB_0.900.png -------------------------------------------------------------------------------- /eval_images_RGB/3_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/3_gt.png -------------------------------------------------------------------------------- /eval_images_RGB/3_noisy_25.74dB_0.701.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/3_noisy_25.74dB_0.701.png -------------------------------------------------------------------------------- /eval_images_RGB/3_pred_35.85dB_0.960.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/3_pred_35.85dB_0.960.png -------------------------------------------------------------------------------- /eval_images_RGB/4_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/4_gt.png -------------------------------------------------------------------------------- /eval_images_RGB/4_noisy_19.34dB_0.595.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/4_noisy_19.34dB_0.595.png -------------------------------------------------------------------------------- /eval_images_RGB/4_pred_29.69dB_0.937.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/4_pred_29.69dB_0.937.png -------------------------------------------------------------------------------- /eval_images_RGB/5_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/5_gt.png -------------------------------------------------------------------------------- /eval_images_RGB/5_noisy_27.50dB_0.641.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/5_noisy_27.50dB_0.641.png -------------------------------------------------------------------------------- /eval_images_RGB/5_pred_33.03dB_0.840.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/5_pred_33.03dB_0.840.png -------------------------------------------------------------------------------- /eval_images_RGB/6_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/6_gt.png -------------------------------------------------------------------------------- /eval_images_RGB/6_noisy_16.37dB_0.283.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/6_noisy_16.37dB_0.283.png -------------------------------------------------------------------------------- /eval_images_RGB/6_pred_29.25dB_0.844.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/6_pred_29.25dB_0.844.png -------------------------------------------------------------------------------- /eval_images_RGB/7_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/7_gt.png -------------------------------------------------------------------------------- /eval_images_RGB/7_noisy_27.71dB_0.634.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/7_noisy_27.71dB_0.634.png -------------------------------------------------------------------------------- /eval_images_RGB/7_pred_38.33dB_0.961.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/7_pred_38.33dB_0.961.png -------------------------------------------------------------------------------- /eval_images_RGB/8_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/8_gt.png -------------------------------------------------------------------------------- /eval_images_RGB/8_noisy_30.11dB_0.870.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/8_noisy_30.11dB_0.870.png -------------------------------------------------------------------------------- /eval_images_RGB/8_pred_32.60dB_0.942.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/8_pred_32.60dB_0.942.png -------------------------------------------------------------------------------- /eval_images_RGB/9_gt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/9_gt.png -------------------------------------------------------------------------------- /eval_images_RGB/9_noisy_24.02dB_0.597.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/9_noisy_24.02dB_0.597.png -------------------------------------------------------------------------------- /eval_images_RGB/9_pred_33.24dB_0.914.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/LujiaJin/kernel-prediction-networks-PyTorch/209f45ff04a282267ccbb5323daa33717379daad/eval_images_RGB/9_pred_33.24dB_0.914.png -------------------------------------------------------------------------------- /kpn_data_provider.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | import torchvision.transforms as transforms 4 | import torch.nn.functional as F 5 | from torch.utils.data import Dataset, DataLoader 6 | import os 7 | from PIL import Image 8 | import numpy as np 9 | from skimage.color import rgb2xyz 10 | import inspect 11 | from utils.training_util import read_config 12 | from data_generation.data_utils import * 13 | import torch.nn.functional as F 14 | 15 | 16 | def sRGBGamma(tensor): 17 | threshold = 0.0031308 18 | a = 0.055 19 | mult = 12.92 20 | gamma = 2.4 21 | res = torch.zeros_like(tensor) 22 | mask = tensor > threshold 23 | # image_lo = tensor * mult 24 | # 0.001 is to avoid funny thing at 0. 25 | # image_hi = (1 + a) * torch.pow(tensor + 0.001, 1.0 / gamma) - a 26 | res[mask] = (1 + a) * torch.pow(tensor[mask], 1.0 / gamma) - a 27 | 28 | # res[1-mask] = tensor[1-mask] * mult 29 | res[~mask] = tensor[~mask] * mult 30 | # return mask * image_hi + (1 - mask) * image_lo 31 | return res 32 | 33 | 34 | def UndosRGBGamma(tensor): 35 | threshold = 0.0031308 36 | a = 0.055 37 | mult = 12.92 38 | gamma = 2.4 39 | res = torch.zeros_like(tensor) 40 | mask = tensor > threshold 41 | # image_lo = tensor / mult 42 | # image_hi = torch.pow(tensor + a, gamma) / (1 + a) 43 | # res[1-mask] = tensor[1-mask] / mult 44 | res[~mask] = tensor[~mask] / mult 45 | res[mask] = torch.pow(tensor[mask] + a, gamma) / (1 + a) 46 | # return mask * image_hi + (1 - mask) * image_lo 47 | return res 48 | 49 | 50 | class Random_Horizontal_Flip(object): 51 | def __init__(self, p=0.5): 52 | self.p = p 53 | 54 | def __call__(self, tensor): 55 | if np.random.rand() < self.p: 56 | return torch.flip(tensor, dims=[-1]) 57 | return tensor 58 | 59 | 60 | class Random_Vertical_Flip(object): 61 | def __init__(self, p=0.5): 62 | self.p = p 63 | 64 | def __call__(self, tensor): 65 | if np.random.rand() < self.p: 66 | return torch.flip(tensor, dims=[-2]) 67 | return tensor 68 | 69 | 70 | class TrainDataSet(torch.utils.data.Dataset): 71 | def __init__(self, config_file, 72 | config_spec=None, img_format='.bmp', degamma=True, color=True, 73 | blind=False, train=True): 74 | super(TrainDataSet, self).__init__() 75 | if config_spec is None: 76 | config_spec = self._configspec_path() 77 | config = read_config(config_file, config_spec) 78 | self.dataset_config = config['dataset_configs'] 79 | self.dataset_dir = self.dataset_config['dataset_dir'] 80 | self.images = list(filter(lambda x: True if img_format in x else False, 81 | os.listdir(self.dataset_dir))) 82 | self.burst_size = self.dataset_config['burst_length'] 83 | self.patch_size = self.dataset_config['patch_size'] 84 | 85 | self.upscale = self.dataset_config['down_sample'] 86 | self.big_jitter = self.dataset_config['big_jitter'] 87 | self.small_jitter = self.dataset_config['small_jitter'] 88 | # 对应下采样之前图像的最大偏移量 89 | self.jitter_upscale = self.big_jitter * self.upscale 90 | # 对应下采样之前的图像的patch尺寸 91 | self.size_upscale = self.patch_size * self.upscale + 2 * self.jitter_upscale 92 | # 产生大jitter和小jitter之间的delta 在下采样之前的尺度上 93 | self.delta_upscale = ( 94 | self.big_jitter - self.small_jitter) * self.upscale 95 | # 对应到原图的patch的尺寸 96 | self.patch_size_upscale = self.patch_size * self.upscale 97 | 98 | # 去伽马效应 99 | self.degamma = degamma 100 | # 是否用彩色图像进行处理 101 | self.color = color 102 | # 是否盲估计 盲估计即估计的噪声方差不会作为网络的输入 103 | self.blind = blind 104 | self.train = train 105 | 106 | self.vertical_flip = Random_Vertical_Flip(p=0.5) 107 | self.horizontal_flip = Random_Horizontal_Flip(p=0.5) 108 | 109 | @staticmethod 110 | def _configspec_path(): 111 | current_dir = os.path.dirname( 112 | os.path.abspath(inspect.getfile(inspect.currentframe())) 113 | ) # 获取当前正在执行的脚本在系统中的文件路径 114 | return os.path.join(current_dir, 115 | 'dataset_specs/data_configspec.conf') 116 | 117 | @staticmethod 118 | def crop_random(tensor, patch_size): 119 | return random_crop(tensor, 1, patch_size)[0] 120 | 121 | # get一个item 根据index检索 122 | def __getitem__(self, index): 123 | # print(index) 124 | image = Image.open( 125 | os.path.join(self.dataset_dir, self.images[index])).convert('RGB') 126 | 127 | # 先转换为Tensor进行degamma 128 | image = transforms.ToTensor()(image) 129 | 130 | # At testing time, if the image size is odd, the edge should be chopped 131 | # off to make the image size even. Otherwise, an error will be reported 132 | # because the image size before and after skip concatenation are 133 | # inconsistent. 134 | if not self.train: 135 | height = image.size()[-2] 136 | width = image.size()[-1] 137 | if height % 2 == 0 and width % 2 == 0: 138 | pass 139 | elif height % 2 == 1 and width % 2 == 0: 140 | image = image[..., :-1, :] 141 | elif height % 2 == 0 and width % 2 == 1: 142 | image = image[..., :, :-1] 143 | else: 144 | image = image[..., :-1, :-1] 145 | 146 | # if self.degamma: 147 | # image = UndosRGBGamma(tensor=image) 148 | image_crop = self.crop_random(image, self.size_upscale) 149 | # 3*H*W 对应于较小jitter下 150 | image_crop_small = image_crop[:, self.delta_upscale:-self.delta_upscale, 151 | self.delta_upscale:-self.delta_upscale] 152 | 153 | # 进一步进行random_crop所需的transform 154 | 155 | # burst中的第一个不做偏移 后期作为target 156 | # output shape: N*3*H*W 157 | img_burst = [] 158 | for i in range(self.burst_size): 159 | if i == 0: 160 | img_burst.append( 161 | image_crop[:, self.jitter_upscale:-self.jitter_upscale, 162 | self.jitter_upscale:-self.jitter_upscale] 163 | ) 164 | else: 165 | if np.random.binomial(1, min(1.0, np.random.poisson( 166 | lam=1.5) / self.burst_size)) == 0: 167 | img_burst.append( 168 | self.crop_random( 169 | image_crop_small, self.patch_size_upscale 170 | ) 171 | ) 172 | else: # big 173 | img_burst.append( 174 | self.crop_random(image_crop, self.patch_size_upscale) 175 | ) 176 | image_burst = torch.stack(img_burst, dim=0) 177 | image_burst = F.adaptive_avg_pool2d(image_burst, 178 | (self.patch_size, self.patch_size)) 179 | 180 | # label为patch中burst的第一个 181 | if not self.color: 182 | image_burst = 0.2989 * image_burst[:, 0, 183 | ...] + 0.5870 * image_burst[:, 1, 184 | ...] + 0.1140 * image_burst[ 185 | :, 2, ...] 186 | image_burst = torch.clamp(image_burst, 0.0, 1.0) 187 | 188 | if self.degamma: 189 | UndosRGBGamma(image_burst) 190 | 191 | if self.train: 192 | # data augment 193 | image_burst = self.horizontal_flip(image_burst) 194 | image_burst = self.vertical_flip(image_burst) 195 | 196 | gt = image_burst[0, ...] 197 | 198 | # 以上得到的patch size为burst*（3）*size*size 199 | """ 200 | 数据加噪声等一系列处理 全部基于rgb图像做 201 | """ 202 | # 要产生[log10(0.1), log10(1.0)]之间的均匀分布随机数 也就是[0,1加负号即可] 203 | # 产生pred之后 再除以white_level恢复原来的亮度 204 | # batch中的每一个burst 产生一个white_level 205 | white_level = torch.from_numpy( 206 | np.power(10, -np.random.rand(1, 1, 1))).type_as(image_burst) 207 | # 论文中对图像亮度赋值进行线性缩放[0.1, 1] 208 | image_burst = white_level * image_burst 209 | 210 | # gray image 211 | if not self.color: 212 | # 生成随机的read和shot噪声方差 213 | sigma_read = torch.from_numpy( 214 | np.power(10, np.random.uniform(-3.0, -1.5, (1, 1, 1)))).type_as( 215 | image_burst) 216 | sigma_shot = torch.from_numpy( 217 | np.power(10, np.random.uniform(-4.0, -2.0, (1, 1, 1)))).type_as( 218 | image_burst) 219 | 220 | # sigma_read = torch.from_numpy(2*np.power(10, np.array([[[-2.0]]]))).type_as(image_burst) 221 | # sigma_shot = torch.from_numpy(6.4 * np.power(10, np.array([[[-3.0]]]))).type_as(image_burst) 222 | 223 | # 产生噪声 依据论文中公式产生 224 | sigma_read_com = sigma_read.expand_as(image_burst) 225 | sigma_shot_com = sigma_shot.expand_as(image_burst) 226 | 227 | # generate noise 228 | burst_noise = torch.normal(image_burst, torch.sqrt( 229 | sigma_read_com ** 2 + image_burst * sigma_shot_com)).type_as( 230 | image_burst) 231 | 232 | # burst_noise 恢复到[0,1] 截去外面的值 233 | burst_noise = torch.clamp(burst_noise, 0.0, 1.0) 234 | 235 | # 非盲估计 就要估计噪声的方差 236 | if not self.blind: 237 | # 接下来就是根据两个sigma 将估计的噪声标准差也作为输入 用burst中的第一个进行估计 238 | # estimation shape: H*W 239 | sigma_read_est = sigma_read.view(1, 1).expand_as(gt) 240 | sigma_shot_est = sigma_shot.view(1, 1).expand_as(gt) 241 | sigma_estimate = torch.sqrt( 242 | sigma_read_est ** 2 + sigma_shot_est.mul( 243 | torch.max(torch.stack([burst_noise[0, ...], 244 | torch.zeros_like( 245 | burst_noise[0, ...])], 246 | dim=0), dim=0)[0])) 247 | 248 | # 把噪声的估计和burst图像连接在一起 249 | burst_noise = torch.cat( 250 | [burst_noise, sigma_estimate.unsqueeze(0)], dim=0) 251 | 252 | # 按照文章中的 ref Image作为target进行了训练 输出结果和ref很相似 没能起到太大的去噪作用 253 | # return patches_with_noise, patches_with_noise[:, 0, ...], white_level 254 | # 不含噪声的ref作为target进行测试 255 | 256 | return burst_noise, gt, white_level 257 | # color image 258 | else: 259 | # 生成随机的read和shot噪声方差 260 | sigma_read = torch.from_numpy( 261 | np.power(10, 262 | np.random.uniform(-3.0, -1.5, (1, 1, 1, 1)))).type_as( 263 | image_burst) 264 | sigma_shot = torch.from_numpy( 265 | np.power(10, 266 | np.random.uniform(-4.0, -2.0, (1, 1, 1, 1)))).type_as( 267 | image_burst) 268 | 269 | # 产生噪声 依据论文中公式产生 270 | sigma_read_com = sigma_read.expand_as(image_burst) 271 | sigma_shot_com = sigma_shot.expand_as(image_burst) 272 | 273 | # generate noise 274 | burst_noise = torch.normal(image_burst, 275 | torch.sqrt( 276 | sigma_read_com ** 2 + image_burst * sigma_shot_com)).type_as( 277 | image_burst) 278 | 279 | # burst_noise 恢复到[0,1] 截去外面的值 280 | burst_noise = torch.clamp(burst_noise, 0.0, 1.0) 281 | 282 | # 非盲估计 就要估计噪声的方差 283 | if not self.blind: 284 | # 接下来就是根据两个sigma 将估计的噪声标准差也作为输入 用burst中的第一个进行估计 285 | # estimation shape: H*W 286 | sigma_read_est = sigma_read.view(1, 1, 1).expand_as(gt) 287 | sigma_shot_est = sigma_shot.view(1, 1, 1).expand_as(gt) 288 | sigma_estimate = torch.sqrt( 289 | sigma_read_est ** 2 + sigma_shot_est.mul( 290 | torch.max(torch.stack([burst_noise[0, ...], 291 | torch.zeros_like( 292 | burst_noise[0, ...])], 293 | dim=0), dim=0)[0])) 294 | 295 | # 把噪声的估计和burst图像连接在一起 296 | burst_noise = torch.cat( 297 | [burst_noise, sigma_estimate.unsqueeze(0)], dim=0) 298 | 299 | white_level = white_level.unsqueeze(0) 300 | return burst_noise, gt, white_level 301 | 302 | def __len__(self): 303 | return len(self.images) 304 | 305 | 306 | if __name__ == '__main__': 307 | # path = 'F:/BinZhang/Codes/deep-burst-denoising/data/train' 308 | # dataset = TrainDataSet(path, '.jpg', 8, 128, 4, 16, 2, color=False) 309 | # dataloader = DataLoader(dataset, 310 | # batch_size=4, 311 | # shuffle=True, 312 | # num_workers=4) 313 | # dataloader = iter(dataloader) 314 | # a, b, c = next(dataloader) 315 | # print(a.size(), b.size(), c.size()) 316 | 317 | hf = Random_Horizontal_Flip(0.5) 318 | a = torch.randint(0, 10, (2, 2)) 319 | print(a, hf(a)) 320 | -------------------------------------------------------------------------------- /kpn_specs/configspec.conf: -------------------------------------------------------------------------------- 1 | exp_name = string 2 | 3 | [training] 4 | image_width = integer 5 | image_height = integer 6 | continue_train = boolean 7 | num_epochs = integer(default=100) 8 | 9 | use_cache = boolean(default=False) 10 | cache_dir = string(default=None) 11 | 12 | # IO 13 | dataset_configs = string(default=None) 14 | checkpoint_dir = string(default=None) 15 | logs_dir = string(default=None) 16 | eval_dir = string(default=None) 17 | restore_dir = string(default=None) 18 | save_freq = integer(default=6250) 19 | vis_freq = integer(default=100) 20 | ckpt_to_keep = integer(default=5) 21 | 22 | # Learning 23 | decay_steps = integer(default=3000) 24 | batch_size = integer(default=8) 25 | learning_rate = float(default=0.0002) 26 | lr_decay = float(default=0.97) 27 | beta1 = float(default=0.9) 28 | beta2 = float(default=0.999) 29 | n_loss_average = integer(default=1000) 30 | weight_decay = float(default=5e-4) 31 | optimizer = string(default="adam") 32 | 33 | [architecture] 34 | arch = string 35 | # 36 | kernel_size = string 37 | sep_conv = boolean(default=False) 38 | blind_est = boolean(default=False) 39 | # Spatial attention 40 | spatial_att = boolean(default=False) 41 | # Channel-wise attention 42 | channel_att = boolean(default=False) 43 | # 44 | alpha = float(default=0.9998) 45 | beta = float(default=100.0) 46 | # 47 | upMode = string(default=bilinear) 48 | # 49 | core_bias = boolean(default=False) 50 | 51 | 52 | 53 | 54 | -------------------------------------------------------------------------------- /kpn_specs/kpn_config.conf: -------------------------------------------------------------------------------- 1 | exp_name = kpn 2 | root_exp_dir = ./models/ 3 | exp_dir =%(root_exp_dir)s/%(exp_name)s 4 | 5 | [training] 6 | image_width = 128 7 | image_height = 128 8 | continue_train = False 9 | num_epochs = 500 10 | 11 | use_cache = True 12 | cache_dir = ../dataset/synthetic 13 | 14 | # IO 15 | dataset_configs = dataset_specs/full_dataset.conf 16 | checkpoint_dir = %(exp_dir)s/checkpoint 17 | logs_dir = %(exp_dir)s/logs 18 | eval_dir = %(exp_dir)s/eval_images 19 | save_freq = 200 20 | ckpt_to_keep = 20 21 | 22 | # Learning 23 | decay_steps = 1 24 | batch_size = 16 25 | lr_decay = 0.89125093813 # (10^-0.05) 26 | learning_rate = 2e-4 27 | weight_decay = 1e-4 28 | optimizer = adam 29 | 30 | [architecture] 31 | arch = "kpn" 32 | 33 | kernel_size = 5 34 | sep_conv = False 35 | blind_est = False 36 | alpha = 0.9998 37 | beta = 100.0 38 | 39 | upMode = bilinear 40 | core_bias = False 41 | 42 | -------------------------------------------------------------------------------- /models/kpn/checkpoint/README.md: -------------------------------------------------------------------------------- 1 | Due to the limitation of bandwidth of github, the pretrained model has uploaded to Google Drive at [https://drive.google.com/open?id=1Xnpllr1dinAU7BIN21L3LkEP5AqMNWso](https://drive.google.com/open?id=1Xnpllr1dinAU7BIN21L3LkEP5AqMNWso). 2 | -------------------------------------------------------------------------------- /train_eval_syn.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.optim as optim 3 | from torch.optim import lr_scheduler 4 | import torch.nn as nn 5 | from torch.utils.data import DataLoader 6 | 7 | import numpy as np 8 | import argparse 9 | 10 | import os, sys, time, shutil 11 | 12 | from data_provider import OnTheFlyDataset, _configspec_path 13 | from kpn_data_provider import TrainDataSet, UndosRGBGamma, sRGBGamma 14 | from KPN import KPN, LossFunc 15 | from utils.training_util import MovingAverage, save_checkpoint, load_checkpoint, read_config 16 | from utils.training_util import calculate_psnr, calculate_ssim 17 | 18 | from tensorboardX import SummaryWriter 19 | from PIL import Image 20 | from torchvision.transforms import transforms 21 | 22 | def train(config, num_workers, num_threads, cuda, restart_train, mGPU): 23 | # torch.set_num_threads(num_threads) 24 | 25 | train_config = config['training'] 26 | arch_config = config['architecture'] 27 | 28 | batch_size = train_config['batch_size'] 29 | lr = train_config['learning_rate'] 30 | weight_decay = train_config['weight_decay'] 31 | decay_step = train_config['decay_steps'] 32 | lr_decay = train_config['lr_decay'] 33 | 34 | n_epoch = train_config['num_epochs'] 35 | use_cache = train_config['use_cache'] 36 | 37 | print('Configs:', config) 38 | # checkpoint path 39 | checkpoint_dir = train_config['checkpoint_dir'] 40 | if not os.path.exists(checkpoint_dir): 41 | os.makedirs(checkpoint_dir) 42 | # logs path 43 | logs_dir = train_config['logs_dir'] 44 | if not os.path.exists(logs_dir): 45 | os.makedirs(logs_dir) 46 | # This sentence is usually commented to prevent the previous log from being 47 | # deleted when continuing training from a checkpoint. 48 | # shutil.rmtree(logs_dir) 49 | log_writer = SummaryWriter(logs_dir) 50 | 51 | dataset_config = \ 52 | read_config(train_config['dataset_configs'], _configspec_path())[ 53 | 'dataset_configs'] 54 | # dataset and dataloader 55 | data_set = TrainDataSet( 56 | train_config['dataset_configs'], 57 | img_format='.bmp', 58 | degamma=True, 59 | color=dataset_config['color'], 60 | blind=arch_config['blind_est'] 61 | ) 62 | data_loader = DataLoader( 63 | data_set, 64 | batch_size=batch_size, 65 | shuffle=True, 66 | num_workers=num_workers 67 | ) 68 | 69 | # model here 70 | model = KPN( 71 | color=dataset_config['color'], 72 | burst_length=dataset_config['burst_length'], 73 | blind_est=arch_config['blind_est'], 74 | kernel_size=list(map(int, arch_config['kernel_size'].split())), 75 | sep_conv=arch_config['sep_conv'], 76 | channel_att=arch_config['channel_att'], 77 | spatial_att=arch_config['spatial_att'], 78 | upMode=arch_config['upMode'], 79 | core_bias=arch_config['core_bias'] 80 | ) 81 | if cuda: 82 | model = model.cuda() 83 | 84 | if mGPU: 85 | model = nn.DataParallel(model) 86 | model.train() 87 | 88 | # loss function here 89 | loss_func = LossFunc( 90 | coeff_basic=1.0, 91 | coeff_anneal=1.0, 92 | gradient_L1=True, 93 | alpha=arch_config['alpha'], 94 | beta=arch_config['beta'] 95 | ) 96 | 97 | # Optimizer here 98 | if train_config['optimizer'] == 'adam': 99 | optimizer = optim.Adam( 100 | model.parameters(), 101 | lr=lr 102 | ) 103 | elif train_config['optimizer'] == 'sgd': 104 | optimizer = optim.SGD( 105 | model.parameters(), 106 | lr=lr, 107 | momentum=0.9, 108 | weight_decay=weight_decay 109 | ) 110 | else: 111 | raise ValueError("Optimizer must be 'sgd' or 'adam', but received {}.".format(train_config['optimizer'])) 112 | optimizer.zero_grad() 113 | 114 | # learning rate scheduler here 115 | scheduler = lr_scheduler.StepLR(optimizer, step_size=10, gamma=lr_decay) 116 | 117 | average_loss = MovingAverage(train_config['save_freq']) 118 | if not restart_train: 119 | try: 120 | checkpoint = load_checkpoint(checkpoint_dir, 'best') 121 | start_epoch = checkpoint['epoch'] 122 | global_step = checkpoint['global_iter'] 123 | best_loss = checkpoint['best_loss'] 124 | model.load_state_dict(checkpoint['state_dict']) 125 | optimizer.load_state_dict(checkpoint['optimizer']) 126 | scheduler.load_state_dict(checkpoint['lr_scheduler']) 127 | print('=> loaded checkpoint (epoch {}, global_step {})'.format(start_epoch, global_step)) 128 | except: 129 | start_epoch = 0 130 | global_step = 0 131 | best_loss = np.inf 132 | print('=> no checkpoint file to be loaded.') 133 | else: 134 | start_epoch = 0 135 | global_step = 0 136 | best_loss = np.inf 137 | if os.path.exists(checkpoint_dir): 138 | pass 139 | # files = os.listdir(checkpoint_dir) 140 | # for f in files: 141 | # os.remove(os.path.join(checkpoint_dir, f)) 142 | else: 143 | os.mkdir(checkpoint_dir) 144 | print('=> training') 145 | 146 | burst_length = dataset_config['burst_length'] 147 | data_length = burst_length if arch_config['blind_est'] else burst_length+1 148 | patch_size = dataset_config['patch_size'] 149 | 150 | for epoch in range(start_epoch, n_epoch): 151 | epoch_start_time = time.time() 152 | # decay the learning rate 153 | lr_cur = [param['lr'] for param in optimizer.param_groups] 154 | if lr_cur[0] > 5e-6: 155 | scheduler.step() 156 | else: 157 | for param in optimizer.param_groups: 158 | param['lr'] = 5e-6 159 | print('='*20, 'lr={}'.format([param['lr'] for param in optimizer.param_groups]), '='*20) 160 | t1 = time.time() 161 | for step, (burst_noise, gt, white_level) in enumerate(data_loader): 162 | if cuda: 163 | burst_noise = burst_noise.cuda() 164 | gt = gt.cuda() 165 | white_level = white_level.cuda() 166 | # print('white_level', white_level, white_level.size()) 167 | 168 | # 169 | pred_i, pred = model(torch.flatten(burst_noise, 1, 2) if dataset_config['color'] else burst_noise, burst_noise[:, 0:burst_length, ...], white_level) 170 | 171 | # 172 | loss_basic, loss_anneal = loss_func(sRGBGamma(pred_i), sRGBGamma(pred), sRGBGamma(gt), global_step) 173 | loss = loss_basic + loss_anneal 174 | # backward 175 | optimizer.zero_grad() 176 | loss.backward() 177 | optimizer.step() 178 | # update the average loss 179 | average_loss.update(loss) 180 | # calculate PSNR 181 | psnr = calculate_psnr(pred if dataset_config['color'] else pred.unsqueeze(1), gt if dataset_config['color'] else gt.unsqueeze(1)) 182 | ssim = calculate_ssim(pred if dataset_config['color'] else pred.unsqueeze(1), gt if dataset_config['color'] else gt.unsqueeze(1)) 183 | 184 | # add scalars to tensorboardX 185 | log_writer.add_scalar('loss_basic', loss_basic, global_step) 186 | log_writer.add_scalar('loss_anneal', loss_anneal, global_step) 187 | log_writer.add_scalar('loss_total', loss, global_step) 188 | log_writer.add_scalar('psnr', psnr, global_step) 189 | log_writer.add_scalar('ssim', ssim, global_step) 190 | 191 | # print 192 | print('{:-4d}\t| epoch {:2d}\t| step {:4d}\t| loss_basic: {:.4f}\t| loss_anneal: {:.4f}\t|' 193 | ' loss: {:.4f}\t| PSNR: {:.2f}dB\t| SSIM: {:.4f}\t| time:{:.2f} seconds.' 194 | .format(global_step, epoch, step, loss_basic, loss_anneal, loss, psnr, ssim, time.time()-t1)) 195 | t1 = time.time() 196 | # global_step 197 | global_step += 1 198 | 199 | if global_step % train_config['save_freq'] == 0: 200 | if average_loss.get_value() < best_loss: 201 | is_best = True 202 | best_loss = average_loss.get_value() 203 | else: 204 | is_best = False 205 | 206 | save_dict = { 207 | 'epoch': epoch, 208 | 'global_iter': global_step, 209 | 'state_dict': model.state_dict(), 210 | 'best_loss': best_loss, 211 | 'optimizer': optimizer.state_dict(), 212 | 'lr_scheduler': scheduler.state_dict() 213 | } 214 | save_checkpoint( 215 | save_dict, is_best, checkpoint_dir, global_step, max_keep=train_config['ckpt_to_keep'] 216 | ) 217 | 218 | print('Epoch {} is finished, time elapsed {:.2f} seconds.'.format(epoch, time.time()-epoch_start_time)) 219 | 220 | 221 | def eval(config, args): 222 | train_config = config['training'] 223 | arch_config = config['architecture'] 224 | 225 | use_cache = train_config['use_cache'] 226 | 227 | print('Eval Process......') 228 | 229 | checkpoint_dir = train_config['checkpoint_dir'] 230 | if not os.path.exists(checkpoint_dir) or len(os.listdir(checkpoint_dir)) == 0: 231 | print('There is no any checkpoint file in path:{}'.format(checkpoint_dir)) 232 | # the path for saving eval images 233 | eval_dir = train_config['eval_dir'] 234 | if not os.path.exists(eval_dir): 235 | os.mkdir(eval_dir) 236 | files = os.listdir(eval_dir) 237 | for f in files: 238 | os.remove(os.path.join(eval_dir, f)) 239 | 240 | dataset_config = \ 241 | read_config(train_config['dataset_configs'], _configspec_path())[ 242 | 'dataset_configs'] 243 | # dataset and dataloader 244 | 245 | # 固定随机种子 246 | setup_seed(42) 247 | 248 | data_set = TrainDataSet( 249 | train_config['dataset_configs'], 250 | img_format='.bmp', 251 | degamma=True, 252 | color=dataset_config['color'], 253 | blind=arch_config['blind_est'], 254 | train=False 255 | ) 256 | data_loader = DataLoader( 257 | data_set, 258 | batch_size=1, 259 | shuffle=False, 260 | num_workers=args.num_workers 261 | ) 262 | 263 | 264 | # model here 265 | model = KPN( 266 | color=dataset_config['color'], 267 | burst_length=dataset_config['burst_length'], 268 | blind_est=arch_config['blind_est'], 269 | kernel_size=list(map(int, arch_config['kernel_size'].split())), 270 | sep_conv=arch_config['sep_conv'], 271 | channel_att=arch_config['channel_att'], 272 | spatial_att=arch_config['spatial_att'], 273 | upMode=arch_config['upMode'], 274 | core_bias=arch_config['core_bias'] 275 | ) 276 | if args.cuda: 277 | model = model.cuda() 278 | 279 | if args.mGPU: 280 | model = nn.DataParallel(model) 281 | # load trained model 282 | ckpt = load_checkpoint(checkpoint_dir, args.checkpoint) 283 | model.load_state_dict(ckpt['state_dict']) 284 | print('The model has been loaded from epoch {}, n_iter {}.'.format(ckpt['epoch'], ckpt['global_iter'])) 285 | # switch the eval mode 286 | model.eval() 287 | 288 | # data_loader = iter(data_loader) 289 | burst_length = dataset_config['burst_length'] 290 | data_length = burst_length if arch_config['blind_est'] else burst_length + 1 291 | patch_size = dataset_config['patch_size'] 292 | 293 | trans = transforms.ToPILImage() 294 | 295 | with torch.no_grad(): 296 | psnr = [] 297 | ssim = [] 298 | test_num = 0 299 | for i, (burst_noise, gt, white_level) in enumerate(data_loader): 300 | # data = next(data_loader) 301 | if args.cuda: 302 | burst_noise = burst_noise.cuda() 303 | gt = gt.cuda() 304 | white_level = white_level.cuda() 305 | 306 | pred_i, pred = model(torch.flatten(burst_noise, 1, 2) if dataset_config['color'] else burst_noise, burst_noise[:, 0:burst_length, ...], white_level) 307 | 308 | pred_i = sRGBGamma(pred_i) 309 | pred = sRGBGamma(pred) 310 | gt = sRGBGamma(gt) 311 | burst_noise = sRGBGamma(burst_noise / white_level) 312 | 313 | psnr_t = calculate_psnr(pred, gt if dataset_config['color'] else gt.unsqueeze(1)) 314 | ssim_t = calculate_ssim(pred, gt if dataset_config['color'] else gt.unsqueeze(1)) 315 | psnr_noisy = calculate_psnr(burst_noise[:, 0, ...] if dataset_config['color'] else burst_noise[:, 0, ...].unsqueeze(1), gt if dataset_config['color'] else gt.unsqueeze(1)) 316 | ssim_noisy = calculate_ssim(burst_noise[:, 0, ...] if dataset_config['color'] else burst_noise[:, 0, ...].unsqueeze(1), gt if dataset_config['color'] else gt.unsqueeze(1)) 317 | psnr.append(psnr_t) 318 | ssim.append(ssim_t) 319 | 320 | pred = torch.clamp(pred, 0.0, 1.0) 321 | noise = torch.clamp(burst_noise[0, 0, ...], 0.0, 1.0) 322 | 323 | if args.cuda: 324 | pred = pred.cpu() 325 | gt = gt.cpu() 326 | burst_noise = burst_noise.cpu() 327 | 328 | trans(noise.squeeze()).save(os.path.join(eval_dir, '{}_noisy_{:.2f}dB_{:.3f}.png'.format(i, psnr_noisy, ssim_noisy)), quality=100) 329 | trans(pred.squeeze()).save(os.path.join(eval_dir, '{}_pred_{:.2f}dB_{:.3f}.png'.format(i, psnr_t, ssim_t)), quality=100) 330 | trans(gt.squeeze()).save(os.path.join(eval_dir, '{}_gt.png'.format(i)), quality=100) 331 | 332 | print('{}-th image is OK, with PSNR: {:.2f}dB, SSIM: {:.3f}'.format(i, psnr_t, ssim_t)) 333 | test_num += 1 334 | 335 | psnr_mean, psnr_std = np.mean(psnr), np.std(psnr) 336 | ssim_mean, ssim_std = np.mean(ssim), np.std(ssim) 337 | print('All {} images are OK, average PSNR: {:.2f} ± {:.2f}dB, SSIM: {:.3f} ± {:.3f}'.format(test_num, psnr_mean, psnr_std, ssim_mean, ssim_std)) 338 | print('({:.2f}±{:.2f}dB)-({:.3f}±{:.3f}))'.format(psnr_mean, psnr_std, ssim_mean, ssim_std)) 339 | 340 | 341 | if __name__ == '__main__': 342 | # argparse 343 | parser = argparse.ArgumentParser(description='parameters for training') 344 | parser.add_argument('--config_file', dest='config_file', default='kpn_specs/kpn_config.conf', help='path to config file') 345 | parser.add_argument('--config_spec', dest='config_spec', default='kpn_specs/configspec.conf', help='path to config spec file') 346 | parser.add_argument('--restart', action='store_true', help='Whether to remove all old files and restart the training process') 347 | parser.add_argument('--num_workers', '-nw', default=4, type=int, help='number of workers in data loader') 348 | parser.add_argument('--num_threads', '-nt', default=8, type=int, help='number of threads in data loader') 349 | parser.add_argument('--cuda', '-c', action='store_true', help='whether to train on the GPU') 350 | parser.add_argument('--mGPU', '-m', action='store_true', help='whether to train on multiple GPUs') 351 | parser.add_argument('--eval', action='store_true', help='whether to work on the evaluation mode') 352 | parser.add_argument('--checkpoint', '-ckpt', dest='checkpoint', type=str, default='best', 353 | help='the checkpoint to eval') 354 | args = parser.parse_args() 355 | 356 | for k in args.__dict__: 357 | print(k + ": " + str(args.__dict__[k])) 358 | # 359 | config = read_config(args.config_file, args.config_spec) 360 | if args.eval: 361 | eval(config, args) 362 | else: 363 | train(config, args.num_workers, args.num_threads, args.cuda, args.restart, args.mGPU) 364 | -------------------------------------------------------------------------------- /utils/__init__.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import inspect, os 3 | 4 | sys.path.insert(0, os.path.join( 5 | os.path.dirname(inspect.getfile(inspect.currentframe())), 6 | '..')) 7 | sys.path.insert(0, os.path.join( 8 | os.path.dirname(inspect.getfile(inspect.currentframe())))) 9 | -------------------------------------------------------------------------------- /utils/image_utils.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import torch 3 | 4 | 5 | def center_crop_tensor(tensor, w, h): 6 | tw = tensor.size(-1) 7 | th = tensor.size(-2) 8 | if tw < w or th < h: 9 | raise RuntimeError("Crop size is larger than image size.") 10 | h0 = int((th - h) / 2) 11 | w0 = int((tw - w) / 2) 12 | h1 = h0 + h 13 | w1 = w0 + w 14 | return tensor[..., h0:h1, w0:w1] 15 | 16 | 17 | def bayer_crop_tensor(tensor, w, h, mode="random"): 18 | """Crop that preserves Bayer phase""" 19 | tw = tensor.size(-1) 20 | th = tensor.size(-2) 21 | if tw < w or th < h: 22 | raise RuntimeError("Crop size ({}) is larger than image size ({})." \ 23 | .format((w, h), (tw, th))) 24 | if mode == "random": 25 | h0 = np.random.choice(th + 1 - h) 26 | w0 = np.random.choice(tw + 1 - w) 27 | elif mode == "center": 28 | h0 = int((th - h) / 2) 29 | w0 = int((tw - w) / 2) 30 | else: 31 | raise ValueError("Bayer crop: unrecognized mode ({}). Must be 'random' or 'center'.".format(mode)) 32 | # make sure start index is divisible by 2 33 | h0 = h0 - (h0 % 2) 34 | w0 = w0 - (w0 % 2) 35 | h1 = h0 + h 36 | w1 = w0 + w 37 | return tensor[..., h0:h1, w0:w1] 38 | 39 | 40 | def random_crop_tensor(tensor, w, h): 41 | tw = tensor.size(-1) 42 | th = tensor.size(-2) 43 | if tw < w or th < h: 44 | raise RuntimeError("Crop size is larger than image size.") 45 | h0 = np.random.randint(th - h) 46 | w0 = np.random.randint(tw - w) 47 | h1 = h0 + h 48 | w1 = w0 + w 49 | return tensor[..., h0:h1, w0:w1] 50 | 51 | 52 | def check_nan_tensor(x): 53 | return torch.isnan(x).any() 54 | -------------------------------------------------------------------------------- /utils/training_util.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import glob 3 | import torch 4 | import shutil 5 | import os 6 | import cv2 7 | import numbers 8 | import skimage 9 | import random 10 | from collections import OrderedDict 11 | from configobj import ConfigObj 12 | from validate import Validator 13 | from data_generation.pipeline import ImageDegradationPipeline 14 | 15 | 16 | class MovingAverage(object): 17 | def __init__(self, n): 18 | self.n = n 19 | self._cache = [] 20 | self.mean = 0 21 | 22 | def update(self, val): 23 | self._cache.append(val) 24 | if len(self._cache) > self.n: 25 | del self._cache[0] 26 | self.mean = sum(self._cache) / len(self._cache) 27 | 28 | def get_value(self): 29 | return self.mean 30 | 31 | 32 | def save_checkpoint(state, is_best, checkpoint_dir, n_iter, max_keep=10): 33 | filename = os.path.join(checkpoint_dir, "{:07d}.pth.tar".format(n_iter)) 34 | torch.save(state, filename) 35 | if is_best: 36 | shutil.copyfile(filename, 37 | os.path.join(checkpoint_dir, 38 | 'model_best.pth.tar')) 39 | files = sorted(os.listdir(checkpoint_dir)) 40 | rm_files = files[0:max(0, len(files) - max_keep)] 41 | for f in rm_files: 42 | os.remove(os.path.join(checkpoint_dir, f)) 43 | 44 | def _represent_int(s): 45 | try: 46 | int(s) 47 | return True 48 | except ValueError: 49 | return False 50 | 51 | 52 | def load_checkpoint(checkpoint_dir, best_or_latest='best'): 53 | if best_or_latest == 'best': 54 | checkpoint_file = os.path.join(checkpoint_dir, 'model_best.pth.tar') 55 | elif best_or_latest.isdigit(): 56 | best_or_latest = int(best_or_latest) 57 | checkpoint_file = os.path.join(checkpoint_dir, 58 | '{:07d}.pth.tar'.format(best_or_latest)) 59 | if not os.path.exists(checkpoint_file): 60 | files = glob.glob(os.path.join(checkpoint_dir, '*.pth.tar')) 61 | basenames = [os.path.basename(f).split('.')[0] for f in files] 62 | iters = sorted([int(b) for b in basenames if _represent_int(b)]) 63 | raise ValueError('Available iterations are ({} requested): {}'.format(best_or_latest, iters)) 64 | else: 65 | files = glob.glob(os.path.join(checkpoint_dir, '*.pth.tar')) 66 | basenames = [os.path.basename(f).split('.')[0] for f in files] 67 | iters = sorted([int(b) for b in basenames if _represent_int(b)]) 68 | checkpoint_file = os.path.join(checkpoint_dir, 69 | '{:07d}.pth.tar'.format(iters[-1])) 70 | return torch.load(checkpoint_file) 71 | 72 | 73 | def load_statedict_runtime(checkpoint_dir, best_or_latest='best'): 74 | # This function grabs state_dict from checkpoint, and do modification 75 | # to the weight name so that it can be load at runtime. 76 | # During training nn.DataParallel adds 'module.' to the name, 77 | # which doesn't exist at test time. 78 | ckpt = load_checkpoint(checkpoint_dir, best_or_latest) 79 | state_dict = ckpt['state_dict'] 80 | global_iter = ckpt['global_iter'] 81 | new_state_dict = OrderedDict() 82 | for k, v in state_dict.items(): 83 | # remove `module.` 84 | name = k[7:] 85 | new_state_dict[name] = v 86 | return new_state_dict, global_iter 87 | 88 | 89 | def prep_and_vis_flow(flow, flow_visualizer, max_flow=None): 90 | flow = flow_visualizer(flow[0, :, :, :], max_flow=max_flow) 91 | flow = flow.cpu().data.numpy() 92 | return flow 93 | 94 | 95 | def put_text_on_img(image, text, loc=(20, 100), color=(1, 0, 0)): 96 | """ Put text on flow 97 | 98 | Args: 99 | image: numpy array of dimension (3, h, w) 100 | text: text to put on. 101 | loc: ibottom-left location of text in (x, y) from top-left of image. 102 | color: color of the text. 103 | Returns: 104 | image with text written on it. 105 | """ 106 | image = np.array(np.moveaxis(image, 0, -1)).copy() 107 | cv2.putText(image, text, loc, cv2.FONT_HERSHEY_SIMPLEX, 1, color) 108 | return np.moveaxis(image, -1, 0) 109 | 110 | 111 | def read_config(config_file, config_spec): 112 | configspec = ConfigObj(config_spec, raise_errors=True) 113 | config = ConfigObj(config_file, 114 | configspec=configspec, 115 | raise_errors=True, 116 | file_error=True) 117 | config.validate(Validator()) 118 | return config 119 | 120 | 121 | def torch2numpy(tensor, gamma=None): 122 | tensor = torch.clamp(tensor, 0.0, 1.0) 123 | # Convert to 0 - 255 124 | if gamma is not None: 125 | tensor = torch.pow(tensor, gamma) 126 | tensor *= 255.0 127 | return tensor.permute(0, 2, 3, 1).cpu().data.numpy() 128 | 129 | 130 | def prep_for_vis(degraded_img, target_img, output_img, exposure=None): 131 | if exposure is not None: 132 | def adjust_exp(img, exp): 133 | configs = [ 134 | ('PixelClip', {}), 135 | ('ExposureAdjustment', {'nstops': exp}), 136 | ('PixelClip', {}), 137 | ] 138 | return ImageDegradationPipeline(configs)(img) 139 | degraded_img = adjust_exp(degraded_img, exposure) 140 | target_img = adjust_exp(target_img, exposure) 141 | output_img = adjust_exp(output_img, exposure) 142 | degraded_tf = torch2numpy(degraded_img, 1.0 / 2.2).astype('uint8') 143 | # Gamma encode output for illustration purpose 144 | target_tf = torch2numpy(target_img, 1.0 / 2.2).astype('uint8') 145 | output_tf = torch2numpy(output_img, 1.0 / 2.2).astype('uint8') 146 | return degraded_tf, target_tf, output_tf 147 | 148 | 149 | def prep_for_vis_arr(img_arr, exposure=None): 150 | if exposure is not None: 151 | configs = [ 152 | ('PixelClip', {}), 153 | ('ExposureAdjustment', {'nstops': exposure}), 154 | ('PixelClip', {}), 155 | ] 156 | exp_adj = ImageDegradationPipeline(configs) 157 | img_arr = [exp_adj(im) for im in img_arr] 158 | img_arr = [torch2numpy(im, 1.0 / 2.2).astype('uint8') for im in img_arr] 159 | return img_arr 160 | 161 | 162 | def create_vis_arr(img_arr, exposure=None): 163 | img_arr = prep_for_vis_arr(img_arr, exposure) 164 | return np.concatenate(img_arr, axis=-2) 165 | 166 | 167 | def create_vis(degraded_img, target_img, output_img, exposure=None): 168 | degraded_tf, target_tf, output_tf = prep_for_vis(degraded_img, 169 | target_img, 170 | output_img) 171 | img = np.concatenate((degraded_tf, 172 | target_tf, 173 | output_tf), 174 | axis=-2) 175 | return img 176 | 177 | 178 | def calculate_psnr(output_img, target_img): 179 | target_tf = torch2numpy(target_img) 180 | output_tf = torch2numpy(output_img) 181 | psnr = 0.0 182 | n = 0.0 183 | for im_idx in range(output_tf.shape[0]): 184 | # psnr += skimage.measure.compare_psnr(target_tf[im_idx, ...], 185 | # output_tf[im_idx, ...], 186 | # data_range=1.) 187 | psnr += skimage.metrics.peak_signal_noise_ratio(normalize(target_tf[im_idx, ...]), 188 | normalize(output_tf[im_idx, ...]), 189 | data_range=1.) 190 | n += 1.0 191 | return psnr / n 192 | 193 | 194 | def calculate_ssim(output_img, target_img): 195 | target_tf = torch2numpy(target_img) 196 | output_tf = torch2numpy(output_img) 197 | ssim = 0.0 198 | n = 0.0 199 | for im_idx in range(output_tf.shape[0]): 200 | ssim += skimage.metrics.structural_similarity(normalize(target_tf[im_idx, ...]), 201 | normalize(output_tf[im_idx, ...]), 202 | channel_axis=2, 203 | K1=0.01, K2=0.03, sigma=1.5, 204 | data_range=1.) 205 | n += 1.0 206 | return ssim / n 207 | --------------------------------------------------------------------------------