├── .DS_Store ├── LICENSE ├── MR.png ├── README.md ├── __pycache__ ├── ds_fusion.cpython-37.pyc ├── full_arrange.cpython-37.pyc ├── global_var.cpython-37.pyc └── test.cpython-37.pyc ├── cft.png ├── data ├── GlobalWheat2020.yaml ├── VisDrone.yaml ├── argoverse_hd.yaml ├── coco.yaml ├── coco128.yaml ├── hyp.finetune.yaml ├── hyp.scratch.yaml ├── images │ ├── bus.jpg │ └── zidane.jpg ├── multispectral │ ├── FLIR_aligned.yaml │ ├── LLVIP.yaml │ └── vedai_color_2.yaml ├── scripts │ ├── get_argoverse_hd.sh │ ├── get_coco.sh │ └── get_voc.sh └── voc.yaml ├── detect_twostream.py ├── example.png ├── global_var.py ├── hubconf.py ├── models ├── __init__.py ├── __pycache__ │ ├── __init__.cpython-37.pyc │ ├── common.cpython-37.pyc │ ├── experimental.cpython-37.pyc │ ├── yolo.cpython-37.pyc │ └── yolo_test.cpython-37.pyc ├── common.py ├── experimental.py ├── export.py ├── hub │ ├── anchors.yaml │ ├── yolov3-spp.yaml │ ├── yolov3-tiny.yaml │ ├── yolov3.yaml │ ├── yolov5-fpn.yaml │ ├── yolov5-p2.yaml │ ├── yolov5-p6.yaml │ ├── yolov5-p7.yaml │ ├── yolov5-panet.yaml │ ├── yolov5l6.yaml │ ├── yolov5m6.yaml │ ├── yolov5s-transformer.yaml │ ├── yolov5s6.yaml │ └── yolov5x6.yaml ├── transformer │ ├── yolov5l_fusion_add_FLIR_aligned.yaml │ ├── yolov5l_fusion_add_llvip.yaml │ ├── yolov5l_fusion_transformer_FLIR.yaml │ ├── yolov5l_fusion_transformer_FLIR_aligned.yaml │ ├── yolov5l_fusion_transformer_llvip.yaml │ ├── yolov5l_fusion_transformer_vedai.yaml │ ├── yolov5l_fusion_transformerx3_FLIR_aligned.yaml │ ├── yolov5l_fusion_transformerx3_llvip.yaml │ ├── yolov5s_fusion_add_vedai.yaml │ ├── yolov5s_fusion_transformer_vedai.yaml │ ├── yolov5s_fusion_transformerx3_vedai.yaml │ ├── yolov5x_fusion_transformer_FLIR.yaml │ └── yolov5x_fusion_transformer_FLIR_aligned.yaml ├── yolo.py ├── yolo_test.py ├── yolov5l.yaml ├── yolov5m.yaml ├── yolov5s.yaml └── yolov5x.yaml ├── requirements.txt ├── test.py ├── train.py ├── utils ├── __init__.py ├── __pycache__ │ ├── __init__.cpython-37.pyc │ ├── autoanchor.cpython-37.pyc │ ├── datasets.cpython-37.pyc │ ├── ds_fusion.cpython-37.pyc │ ├── general.cpython-37.pyc │ ├── google_utils.cpython-37.pyc │ ├── gradcam.cpython-37.pyc │ ├── loss.cpython-37.pyc │ ├── metrics.cpython-37.pyc │ ├── plots.cpython-37.pyc │ └── torch_utils.cpython-37.pyc ├── activations.py ├── autoanchor.py ├── aws │ ├── __init__.py │ ├── mime.sh │ ├── resume.py │ └── userdata.sh ├── datasets.py ├── ds_fusion.py ├── flask_rest_api │ ├── README.md │ ├── example_request.py │ └── restapi.py ├── general.py ├── google_app_engine │ ├── Dockerfile │ ├── additional_requirements.txt │ └── app.yaml ├── google_utils.py ├── gradcam.py ├── loss.py ├── metrics.py ├── plots.py ├── torch_utils.py └── wandb_logging │ ├── __init__.py │ ├── __pycache__ │ ├── __init__.cpython-37.pyc │ └── wandb_utils.cpython-37.pyc │ ├── log_dataset.py │ └── wandb_utils.py └── video ├── demo.gif └── demo1.gif /.DS_Store: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/.DS_Store -------------------------------------------------------------------------------- /MR.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/MR.png -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Multispectral-Object-Detection 2 | 3 | [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/cross-modality-fusion-transformer-for/multispectral-object-detection-on-flir)](https://paperswithcode.com/sota/multispectral-object-detection-on-flir?p=cross-modality-fusion-transformer-for) 4 | 5 | [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/cross-modality-fusion-transformer-for/pedestrian-detection-on-llvip)](https://paperswithcode.com/sota/pedestrian-detection-on-llvip?p=cross-modality-fusion-transformer-for) 6 | 7 | [![New](https://img.shields.io/badge/2021-NEW-brightgreen.svg)](https://github.com/DocF/multispectral-object-detection/) 8 | ![Visitors](https://visitor-badge.glitch.me/badge?page_id=DocF.multispectral-object-detection) 9 | [![GitHub stars](https://img.shields.io/github/stars/DocF/multispectral-object-detection.svg?style=social&label=Stars)](https://github.com/DocF/multispectral-object-detection) 10 | 11 | 12 | ## Intro 13 | Official Code for [Cross-Modality Fusion Transformer for Multispectral Object Detection](https://arxiv.org/abs/2111.00273). 14 | 15 | Multispectral Object Detection with Transformer and Yolov5 16 | 17 | ## Abstract 18 | Multispectral image pairs can provide the combined information, making object detection applications more reliable and robust in the open world. 19 | To fully exploit the different modalities, we present a simple yet effective cross-modality feature fusion approach, named Cross-Modality Fusion Transformer (CFT) in this paper. 20 | Unlike prior CNNs-based works, guided by the Transformer scheme, our network learns long-range dependencies and integrates global contextual information in the feature extraction stage. 21 | More importantly, by leveraging the self attention of the Transformer, the network can naturally carry out simultaneous intra-modality and inter-modality fusion, and robustly capture the latent interactions between RGB and Thermal domains, thereby significantly improving the performance of multispectral object detection. 22 | Extensive experiments and ablation studies on multiple datasets demonstrate that our approach is effective and achieves state-of-the-art detection performance. 23 | ### Demo 24 | **Night Scene** 25 |
26 | 27 |
28 | 29 | **Day Scene** 30 |
31 | 32 |
33 | 34 | 35 | ### Overview 36 |
37 | 38 |
39 | 40 | ## Citation 41 | If you use this repo for your research, please cite our paper: 42 | 43 | ``` 44 | @article{qingyun2022cross, 45 | title={Cross-modality attentive feature fusion for object detection in multispectral remote sensing imagery}, 46 | author={Qingyun, Fang and Zhaokui, Wang}, 47 | journal={Pattern Recognition}, 48 | volume={130}, 49 | pages={108786}, 50 | year={2022}, 51 | publisher={Elsevier} 52 | } 53 | @article{fang2021cross, 54 | title={Cross-Modality Fusion Transformer for Multispectral Object Detection}, 55 | author={Fang Qingyun and Han Dapeng and Wang Zhaokui}, 56 | journal={arXiv preprint arXiv:2111.00273}, 57 | year={2021} 58 | } 59 | ``` 60 | 61 | 62 | 63 | ## Installation 64 | Python>=3.6.0 is required with all requirements.txt installed including PyTorch>=1.7 (The same as yolov5 https://github.com/ultralytics/yolov5 ). 65 | 66 | #### Clone the repo 67 | git clone https://github.com/DocF/multispectral-object-detection 68 | 69 | #### Install requirements 70 | ```bash 71 | $ cd multispectral-object-detection 72 | $ pip install -r requirements.txt 73 | ``` 74 | 75 | ## Dataset 76 | -[FLIR] [[Google Drive]](http://shorturl.at/ahAY4) [[Baidu Drive]](https://pan.baidu.com/s/1z2GHVD3WVlGsVzBR1ajSrQ?pwd=qwer) ```extraction code:qwer``` 77 | 78 | A new aligned version. 79 | 80 | -[LLVIP] [download](https://github.com/bupt-ai-cz/LLVIP) 81 | 82 | -[VEDAI] [download](https://downloads.greyc.fr/vedai/) 83 | 84 | 85 | You need to convert all annotations to YOLOv5 format. 86 | 87 | Refer: https://github.com/ultralytics/yolov5/wiki/Train-Custom-Data 88 | 89 | ## Run 90 | #### Download the pretrained weights 91 | yolov5 weights (pre-train) 92 | 93 | -[yolov5s] [google drive](https://drive.google.com/file/d/1UGAsaOvV7jVrk0RvFVYL6Vq0K7NQLD8H/view?usp=sharing) 94 | 95 | -[yolov5m] [google drive](https://drive.google.com/file/d/1qB7L2vtlGppGjHp5xpXCKw14YHhbV4s1/view?usp=sharing) 96 | 97 | -[yolov5l] [google drive](https://drive.google.com/file/d/12OFGLF73CqTgOCMJAycZ8lB4eW19D0nb/view?usp=sharing) 98 | 99 | -[yolov5x] [google drive](https://drive.google.com/file/d/1e9xiQImx84KFQ_a7XXpn608I3rhRmKEn/view?usp=sharing) 100 | 101 | CFT weights 102 | 103 | -[LLVIP] [google drive](https://drive.google.com/file/d/18yLDUOxNXQ17oypQ-fAV9OS9DESOZQtV/view?usp=sharing) 104 | 105 | -[FLIR] [google drive](https://drive.google.com/file/d/1PwEOgT5ZOTjoKT2LpOzvCsxsVgwP8NIJ/view) 106 | 107 | 108 | #### Change the data cfg 109 | some example in data/multispectral/ 110 | 111 | #### Change the model cfg 112 | some example in models/transformer/ 113 | 114 | note!!! we used xxxx_transfomerx3_dataset.yaml in our paper. 115 | 116 | ### Train Test and Detect 117 | train: ``` python train.py``` 118 | 119 | test: ``` python test.py``` 120 | 121 | detect: ``` python detect_twostream.py``` 122 | 123 | ## Results 124 | 125 | |Dataset|CFT|mAP50|mAP75|mAP| 126 | |:---------: |------------|:-----:|:-----------------:|:-------------:| 127 | |FLIR||73.0|32.0|37.4| 128 | |FLIR| ✔️ |**78.7 (Δ5.7)**|**35.5 (Δ3.5)**|**40.2 (Δ2.8)**| 129 | |LLVIP||95.8|71.4|62.3| 130 | |LLVIP| ✔️ |**97.5 (Δ1.7)**|**72.9 (Δ1.5)**|**63.6 (Δ1.3)**| 131 | |VEDAI||79.7 | 47.7 | 46.8 132 | |VEDAI| ✔️ |**85.3 (Δ5.6)**|**65.9(Δ18.2)**|**56.0 (Δ9.2)**| 133 | 134 | 135 | ### LLVIP 136 | Log Average Miss Rate 137 | |Model| Log Average Miss Rate | 138 | |:---------: |:--------------:| 139 | |YOLOv3-RGB|37.70%| 140 | |YOLOv3-IR|17.73%| 141 | |YOLOv5-RGB|22.59%| 142 | |YOLOv5-IR|10.66%| 143 | |Baseline(Ours)|**6.91%**| 144 | |CFT(Ours)|**5.40%**| 145 | 146 | Miss Rate - FPPI curve 147 |
148 | 149 |
150 | 151 | #### References 152 | 153 | https://github.com/ultralytics/yolov5 154 | 155 | 156 | -------------------------------------------------------------------------------- /__pycache__/ds_fusion.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/__pycache__/ds_fusion.cpython-37.pyc -------------------------------------------------------------------------------- /__pycache__/full_arrange.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/__pycache__/full_arrange.cpython-37.pyc -------------------------------------------------------------------------------- /__pycache__/global_var.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/__pycache__/global_var.cpython-37.pyc -------------------------------------------------------------------------------- /__pycache__/test.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/__pycache__/test.cpython-37.pyc -------------------------------------------------------------------------------- /cft.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/cft.png -------------------------------------------------------------------------------- /data/GlobalWheat2020.yaml: -------------------------------------------------------------------------------- 1 | # Global Wheat 2020 dataset http://www.global-wheat.com/ 2 | # Train command: python train.py --data GlobalWheat2020.yaml 3 | # Default dataset location is next to YOLOv5: 4 | # /parent_folder 5 | # /datasets/GlobalWheat2020 6 | # /yolov5 7 | 8 | 9 | # train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/] 10 | train: # 3422 images 11 | - ../datasets/GlobalWheat2020/images/arvalis_1 12 | - ../datasets/GlobalWheat2020/images/arvalis_2 13 | - ../datasets/GlobalWheat2020/images/arvalis_3 14 | - ../datasets/GlobalWheat2020/images/ethz_1 15 | - ../datasets/GlobalWheat2020/images/rres_1 16 | - ../datasets/GlobalWheat2020/images/inrae_1 17 | - ../datasets/GlobalWheat2020/images/usask_1 18 | 19 | val: # 748 images (WARNING: train set contains ethz_1) 20 | - ../datasets/GlobalWheat2020/images/ethz_1 21 | 22 | test: # 1276 23 | - ../datasets/GlobalWheat2020/images/utokyo_1 24 | - ../datasets/GlobalWheat2020/images/utokyo_2 25 | - ../datasets/GlobalWheat2020/images/nau_1 26 | - ../datasets/GlobalWheat2020/images/uq_1 27 | 28 | # number of classes 29 | nc: 1 30 | 31 | # class names 32 | names: [ 'wheat_head' ] 33 | 34 | 35 | # download command/URL (optional) -------------------------------------------------------------------------------------- 36 | download: | 37 | from utils.general import download, Path 38 | 39 | # Download 40 | dir = Path('../datasets/GlobalWheat2020') # dataset directory 41 | urls = ['https://zenodo.org/record/4298502/files/global-wheat-codalab-official.zip', 42 | 'https://github.com/ultralytics/yolov5/releases/download/v1.0/GlobalWheat2020_labels.zip'] 43 | download(urls, dir=dir) 44 | 45 | # Make Directories 46 | for p in 'annotations', 'images', 'labels': 47 | (dir / p).mkdir(parents=True, exist_ok=True) 48 | 49 | # Move 50 | for p in 'arvalis_1', 'arvalis_2', 'arvalis_3', 'ethz_1', 'rres_1', 'inrae_1', 'usask_1', \ 51 | 'utokyo_1', 'utokyo_2', 'nau_1', 'uq_1': 52 | (dir / p).rename(dir / 'images' / p) # move to /images 53 | f = (dir / p).with_suffix('.json') # json file 54 | if f.exists(): 55 | f.rename((dir / 'annotations' / p).with_suffix('.json')) # move to /annotations 56 | -------------------------------------------------------------------------------- /data/VisDrone.yaml: -------------------------------------------------------------------------------- 1 | # VisDrone2019-DET dataset https://github.com/VisDrone/VisDrone-Dataset 2 | # Train command: python train.py --data VisDrone.yaml 3 | # Default dataset location is next to YOLOv5: 4 | # /parent_folder 5 | # /VisDrone 6 | # /yolov5 7 | 8 | 9 | # train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/] 10 | train: ../VisDrone/VisDrone2019-DET-train/images # 6471 images 11 | val: ../VisDrone/VisDrone2019-DET-val/images # 548 images 12 | test: ../VisDrone/VisDrone2019-DET-test-dev/images # 1610 images 13 | 14 | # number of classes 15 | nc: 10 16 | 17 | # class names 18 | names: [ 'pedestrian', 'people', 'bicycle', 'car', 'van', 'truck', 'tricycle', 'awning-tricycle', 'bus', 'motor' ] 19 | 20 | 21 | # download command/URL (optional) -------------------------------------------------------------------------------------- 22 | download: | 23 | from utils.general import download, os, Path 24 | 25 | def visdrone2yolo(dir): 26 | from PIL import Image 27 | from tqdm import tqdm 28 | 29 | def convert_box(size, box): 30 | # Convert VisDrone box to YOLO xywh box 31 | dw = 1. / size[0] 32 | dh = 1. / size[1] 33 | return (box[0] + box[2] / 2) * dw, (box[1] + box[3] / 2) * dh, box[2] * dw, box[3] * dh 34 | 35 | (dir / 'labels').mkdir(parents=True, exist_ok=True) # make labels directory 36 | pbar = tqdm((dir / 'annotations').glob('*.txt'), desc=f'Converting {dir}') 37 | for f in pbar: 38 | img_size = Image.open((dir / 'images' / f.name).with_suffix('.jpg')).size 39 | lines = [] 40 | with open(f, 'r') as file: # read annotation.txt 41 | for row in [x.split(',') for x in file.read().strip().splitlines()]: 42 | if row[4] == '0': # VisDrone 'ignored regions' class 0 43 | continue 44 | cls = int(row[5]) - 1 45 | box = convert_box(img_size, tuple(map(int, row[:4]))) 46 | lines.append(f"{cls} {' '.join(f'{x:.6f}' for x in box)}\n") 47 | with open(str(f).replace(os.sep + 'annotations' + os.sep, os.sep + 'labels' + os.sep), 'w') as fl: 48 | fl.writelines(lines) # write label.txt 49 | 50 | 51 | # Download 52 | dir = Path('../VisDrone') # dataset directory 53 | urls = ['https://github.com/ultralytics/yolov5/releases/download/v1.0/VisDrone2019-DET-train.zip', 54 | 'https://github.com/ultralytics/yolov5/releases/download/v1.0/VisDrone2019-DET-val.zip', 55 | 'https://github.com/ultralytics/yolov5/releases/download/v1.0/VisDrone2019-DET-test-dev.zip', 56 | 'https://github.com/ultralytics/yolov5/releases/download/v1.0/VisDrone2019-DET-test-challenge.zip'] 57 | download(urls, dir=dir) 58 | 59 | # Convert 60 | for d in 'VisDrone2019-DET-train', 'VisDrone2019-DET-val', 'VisDrone2019-DET-test-dev': 61 | visdrone2yolo(dir / d) # convert VisDrone annotations to YOLO labels 62 | -------------------------------------------------------------------------------- /data/argoverse_hd.yaml: -------------------------------------------------------------------------------- 1 | # Argoverse-HD dataset (ring-front-center camera) http://www.cs.cmu.edu/~mengtial/proj/streaming/ 2 | # Train command: python train.py --data argoverse_hd.yaml 3 | # Default dataset location is next to YOLOv5: 4 | # /parent_folder 5 | # /argoverse 6 | # /yolov5 7 | 8 | 9 | # download command/URL (optional) 10 | download: bash data/scripts/get_argoverse_hd.sh 11 | 12 | # train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/] 13 | train: ../argoverse/Argoverse-1.1/images/train/ # 39384 images 14 | val: ../argoverse/Argoverse-1.1/images/val/ # 15062 iamges 15 | test: ../argoverse/Argoverse-1.1/images/test/ # Submit to: https://eval.ai/web/challenges/challenge-page/800/overview 16 | 17 | # number of classes 18 | nc: 8 19 | 20 | # class names 21 | names: [ 'person', 'bicycle', 'car', 'motorcycle', 'bus', 'truck', 'traffic_light', 'stop_sign' ] 22 | -------------------------------------------------------------------------------- /data/coco.yaml: -------------------------------------------------------------------------------- 1 | # COCO 2017 dataset http://cocodataset.org 2 | # Train command: python train.py --data coco.yaml 3 | # Default dataset location is next to YOLOv5: 4 | # /parent_folder 5 | # /coco 6 | # /yolov5 7 | 8 | 9 | # download command/URL (optional) 10 | download: bash data/scripts/get_coco.sh 11 | 12 | # train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/] 13 | train: ../coco/train2017.txt # 118287 images 14 | val: ../coco/val2017.txt # 5000 images 15 | test: ../coco/test-dev2017.txt # 20288 of 40670 images, submit to https://competitions.codalab.org/competitions/20794 16 | 17 | # number of classes 18 | nc: 80 19 | 20 | # class names 21 | names: [ 'person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train', 'truck', 'boat', 'traffic light', 22 | 'fire hydrant', 'stop sign', 'parking meter', 'bench', 'bird', 'cat', 'dog', 'horse', 'sheep', 'cow', 23 | 'elephant', 'bear', 'zebra', 'giraffe', 'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee', 24 | 'skis', 'snowboard', 'sports ball', 'kite', 'baseball bat', 'baseball glove', 'skateboard', 'surfboard', 25 | 'tennis racket', 'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple', 26 | 'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair', 'couch', 27 | 'potted plant', 'bed', 'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote', 'keyboard', 'cell phone', 28 | 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors', 'teddy bear', 29 | 'hair drier', 'toothbrush' ] 30 | 31 | # Print classes 32 | # with open('data/coco.yaml') as f: 33 | # d = yaml.safe_load(f) # dict 34 | # for i, x in enumerate(d['names']): 35 | # print(i, x) 36 | -------------------------------------------------------------------------------- /data/coco128.yaml: -------------------------------------------------------------------------------- 1 | # COCO 2017 dataset http://cocodataset.org - first 128 training images 2 | # Train command: python train.py --data coco128.yaml 3 | # Default dataset location is next to YOLOv5: 4 | # /parent_folder 5 | # /coco128 6 | # /yolov5 7 | 8 | 9 | # download command/URL (optional) 10 | download: https://github.com/ultralytics/yolov5/releases/download/v1.0/coco128.zip 11 | 12 | # train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/] 13 | train: ../coco128/images/train2017/ # 128 images 14 | val: ../coco128/images/train2017/ # 128 images 15 | 16 | # number of classes 17 | nc: 80 18 | 19 | # class names 20 | names: [ 'person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train', 'truck', 'boat', 'traffic light', 21 | 'fire hydrant', 'stop sign', 'parking meter', 'bench', 'bird', 'cat', 'dog', 'horse', 'sheep', 'cow', 22 | 'elephant', 'bear', 'zebra', 'giraffe', 'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee', 23 | 'skis', 'snowboard', 'sports ball', 'kite', 'baseball bat', 'baseball glove', 'skateboard', 'surfboard', 24 | 'tennis racket', 'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple', 25 | 'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair', 'couch', 26 | 'potted plant', 'bed', 'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote', 'keyboard', 'cell phone', 27 | 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors', 'teddy bear', 28 | 'hair drier', 'toothbrush' ] 29 | -------------------------------------------------------------------------------- /data/hyp.finetune.yaml: -------------------------------------------------------------------------------- 1 | # Hyperparameters for VOC finetuning 2 | # python train.py --batch 64 --weights yolov5m.pt --data voc.yaml --img 512 --epochs 50 3 | # See tutorials for hyperparameter evolution https://github.com/ultralytics/yolov5#tutorials 4 | 5 | 6 | # Hyperparameter Evolution Results 7 | # Generations: 306 8 | # P R mAP.5 mAP.5:.95 box obj cls 9 | # Metrics: 0.6 0.936 0.896 0.684 0.0115 0.00805 0.00146 10 | 11 | lr0: 0.02 #(0.0032) 12 | lrf: 0.12 13 | momentum: 0.843 14 | weight_decay: 0.00036 15 | warmup_epochs: 2.0 16 | warmup_momentum: 0.5 17 | warmup_bias_lr: 0.05 18 | box: 0.0296 19 | cls: 0.243 20 | cls_pw: 0.631 21 | obj: 0.301 22 | obj_pw: 0.911 23 | iou_t: 0.2 24 | anchor_t: 2.91 25 | # anchors: 3.63 26 | fl_gamma: 0.0 27 | hsv_h: 0.0138 28 | hsv_s: 0.664 29 | hsv_v: 0.464 30 | degrees: 0.373 31 | translate: 0.245 32 | scale: 0.898 33 | shear: 0.602 34 | perspective: 0.0 35 | flipud: 0.00856 36 | fliplr: 0.5 37 | mosaic: 1.0 38 | mixup: 0.243 39 | -------------------------------------------------------------------------------- /data/hyp.scratch.yaml: -------------------------------------------------------------------------------- 1 | # Hyperparameters for COCO training from scratch 2 | # python train.py --batch 40 --cfg yolov5m.yaml --weights '' --data coco.yaml --img 640 --epochs 300 3 | # See tutorials for hyperparameter evolution https://github.com/ultralytics/yolov5#tutorials 4 | 5 | 6 | lr0: 0.01 # initial learning rate ( original SGD=1E-2, Adam=1E-3) 7 | lrf: 0.2 # final OneCycleLR learning rate (lr0 * lrf) 8 | momentum: 0.937 # SGD momentum/Adam beta1 9 | weight_decay: 0.0005 # optimizer weight decay 5e-4 10 | warmup_epochs: 3.0 # warmup epochs (fractions ok) 11 | warmup_momentum: 0.8 # warmup initial momentum 12 | warmup_bias_lr: 0.1 # warmup initial bias lr 13 | box: 0.05 # box loss gain 14 | cls: 0.5 # cls loss gain 15 | cls_pw: 1.0 # cls BCELoss positive_weight 16 | obj: 1.0 # obj loss gain (scale with pixels) 17 | obj_pw: 1.0 # obj BCELoss positive_weight 18 | iou_t: 0.20 # IoU training threshold 19 | anchor_t: 4.0 # anchor-multiple threshold 20 | # anchors: 3 # anchors per output layer (0 to ignore) 21 | fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5) 22 | hsv_h: 0.015 # image HSV-Hue augmentation (fraction) 23 | hsv_s: 0.7 # image HSV-Saturation augmentation (fraction) 24 | hsv_v: 0.4 # image HSV-Value augmentation (fraction) 25 | degrees: 0.0 # image rotation (+/- deg) 26 | translate: 0.1 # image translation (+/- fraction) 27 | scale: 0.5 # image scale (+/- gain) 28 | shear: 0.0 # image shear (+/- deg) 29 | perspective: 0.0 # image perspective (+/- fraction), range 0-0.001 30 | flipud: 0.0 # image flip up-down (probability) 31 | fliplr: 0.5 # image flip left-right (probability) 32 | mosaic: 1.0 # image mosaic (probability) 33 | mixup: 0.0 # image mixup (probability) 34 | -------------------------------------------------------------------------------- /data/images/bus.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/data/images/bus.jpg -------------------------------------------------------------------------------- /data/images/zidane.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/data/images/zidane.jpg -------------------------------------------------------------------------------- /data/multispectral/FLIR_aligned.yaml: -------------------------------------------------------------------------------- 1 | # COCO 2017 dataset http://cocodataset.org - first 128 training images 2 | # Train command: python train.py --data coco128.yaml 3 | # Default dataset location is next to YOLOv5: 4 | # /parent_folder 5 | # /coco128 6 | # /yolov5 7 | 8 | 9 | # download command/URL (optional) 10 | download: https://github.com/ultralytics/yolov5/releases/download/v1.0/coco128.zip 11 | 12 | # train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/] 13 | train_rgb: /home/fqy/Data/FLIR_ADAS_1_3/align/yolo/rgb/align_train.txt # 128 images 14 | val_rgb: /home/fqy/Data/FLIR_ADAS_1_3/align/yolo/rgb/align_validation.txt # 128 images 15 | train_ir: /home/fqy/Data/FLIR_ADAS_1_3/align/yolo/ir/align_train.txt # 128 images 16 | val_ir: /home/fqy/Data/FLIR_ADAS_1_3/align/yolo/ir/align_validation.txt # 128 images 17 | 18 | # number of classes 19 | nc: 3 20 | 21 | # class names 22 | names: [ 'person','car','bicycle'] 23 | -------------------------------------------------------------------------------- /data/multispectral/LLVIP.yaml: -------------------------------------------------------------------------------- 1 | # COCO 2017 dataset http://cocodataset.org - first 128 training images 2 | # Train command: python train.py --data coco128.yaml 3 | # Default dataset location is next to YOLOv5: 4 | # /parent_folder 5 | # /coco128 6 | # /yolov5 7 | 8 | 9 | # train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/] 10 | train_rgb: /home/fqy/DATA/LLVIP/visible/train.txt 11 | val_rgb: /home/fqy/DATA/LLVIP/visible/test.txt 12 | train_ir: /home/fqy/DATA/LLVIP/infrared/train.txt 13 | val_ir: /home/fqy/DATA/LLVIP/infrared/test.txt 14 | 15 | # number of classes 16 | nc: 1 17 | 18 | # class names 19 | names: [ 'person'] 20 | -------------------------------------------------------------------------------- /data/multispectral/vedai_color_2.yaml: -------------------------------------------------------------------------------- 1 | # COCO 2017 dataset http://cocodataset.org - first 128 training images 2 | # Train command: python train.py --data coco128.yaml 3 | # Default dataset location is next to YOLOv5: 4 | # /parent_folder 5 | # /coco128 6 | # /yolov5 7 | 8 | 9 | # download command/URL (optional) 10 | download: https://github.com/ultralytics/yolov5/releases/download/v1.0/coco128.zip 11 | 12 | # train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/] 13 | train_rgb: /home/fqy/Data/vedai/Vehicules1024/Images/color/fold01.txt # 128 images 14 | val_rgb: /home/fqy/Data/vedai/Vehicules1024/Images/color/fold01test.txt # 128 images 15 | train_ir: /home/fqy/Data/vedai/Vehicules1024/Images/ir/fold01.txt # 128 images 16 | val_ir: /home/fqy/Data/vedai/Vehicules1024/Images/ir/fold01test.txt # 128 images 17 | 18 | # number of classes 19 | nc: 9 20 | 21 | # class names 22 | names: [ 'car', 'truck', 'pickup','tractor','camping car','boat','plane','van','other'] 23 | 24 | -------------------------------------------------------------------------------- /data/scripts/get_argoverse_hd.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | # Argoverse-HD dataset (ring-front-center camera) http://www.cs.cmu.edu/~mengtial/proj/streaming/ 3 | # Download command: bash data/scripts/get_argoverse_hd.sh 4 | # Train command: python train.py --data argoverse_hd.yaml 5 | # Default dataset location is next to YOLOv5: 6 | # /parent_folder 7 | # /argoverse 8 | # /yolov5 9 | 10 | # Download/unzip images 11 | d='../argoverse/' # unzip directory 12 | mkdir $d 13 | url=https://argoverse-hd.s3.us-east-2.amazonaws.com/ 14 | f=Argoverse-HD-Full.zip 15 | curl -L $url$f -o $f && unzip -q $f -d $d && rm $f &# download, unzip, remove in background 16 | wait # finish background tasks 17 | 18 | cd ../argoverse/Argoverse-1.1/ 19 | ln -s tracking images 20 | 21 | cd ../Argoverse-HD/annotations/ 22 | 23 | python3 - "$@" <train.txt 84 | cat 2007_train.txt 2007_val.txt 2007_test.txt 2012_train.txt 2012_val.txt >train.all.txt 85 | 86 | mkdir ../VOC ../VOC/images ../VOC/images/train ../VOC/images/val 87 | mkdir ../VOC/labels ../VOC/labels/train ../VOC/labels/val 88 | 89 | python3 - "$@" <= 1 97 | p, s, im0, frame = path[i], '%g: ' % i, im0s[i].copy(), dataset.count 98 | else: 99 | p, s, im0, frame = path, '', im0s.copy(), getattr(dataset, 'frame', 0) 100 | p, s, im0_, frame = path, '', im0s_.copy(), getattr(dataset2, 'frame', 0) 101 | 102 | p = Path(p) # to Path 103 | save_path = str(save_dir / p.name) # img.jpg 104 | txt_path = str(save_dir / 'labels' / p.stem) + ('' if dataset.mode == 'image' else f'_{frame}') # img.txt 105 | s += '%gx%g ' % img.shape[2:] # print string 106 | gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh 107 | # print(gn) 108 | 109 | # # ---------------------------------------------------------------------------- 110 | # # 画GT，替换det 111 | # # 112 | # # --------------------------------------------------------------------------- 113 | # annoPath = "/home/fqy/proj/paper/test_result/gt/" 114 | # annoName = (path_.split("/")[-1]).split(".")[0] + ".txt" 115 | # annoPath += annoName 116 | # # print(annoPath) 117 | # gt = np.loadtxt(annoPath) 118 | # gt = gt.reshape((-1, 5)) 119 | # ones = np.ones((gt.shape[0], 1)) 120 | # gt = np.hstack((gt, ones)) 121 | # gt[:, [0,1,2,3,4,5]] = gt[:, [1,2,3,4,5,0]] 122 | # gt = torch.from_numpy(gt).to(device) 123 | # # print(gt[:, :4]) 124 | # gt[:, :4] = xywh2xyxy(gt[:, :4]) * 640 125 | # 126 | # det = gt 127 | 128 | # print(det) 129 | if len(det): 130 | # Rescale boxes from img_size to im0 size 131 | det[:, :4] = scale_coords(img.shape[2:], det[:, :4], im0.shape).round() 132 | 133 | # Print results 134 | for c in det[:, -1].unique(): 135 | n = (det[:, -1] == c).sum() # detections per class 136 | s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string 137 | 138 | # Write results 139 | for *xyxy, conf, cls in reversed(det): 140 | if save_txt: # Write to file 141 | xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh 142 | line = (cls, *xywh, conf) if opt.save_conf else (cls, *xywh) # label format 143 | with open(txt_path + '.txt', 'a') as f: 144 | f.write(('%g ' * len(line)).rstrip() % line + '\n') 145 | 146 | if save_img or opt.save_crop or view_img: # Add bbox to image 147 | c = int(cls) # integer class 148 | label = None if opt.hide_labels else (names[c] if opt.hide_conf else f'{names[c]} {conf:.2f}') 149 | 150 | plot_one_box(xyxy, im0, label=label, color=colors(c, True), line_thickness=opt.line_thickness) 151 | plot_one_box(xyxy, im0_, label=label, color=colors(c, True), line_thickness=opt.line_thickness) 152 | if opt.save_crop: 153 | save_one_box(xyxy, im0s, file=save_dir / 'crops' / names[c] / f'{p.stem}.jpg', BGR=True) 154 | 155 | # Print time (inference + NMS) 156 | print(f'{s}Done. ({t2 - t1:.6f}s, {1/(t2 - t1):.6f}Hz)') 157 | # add all the fps 158 | img_num += 1 159 | fps_sum += 1/(t2 - t1) 160 | 161 | # Stream results 162 | if view_img: 163 | cv2.imshow(str(p), im0) 164 | cv2.waitKey(1) # 1 millisecond 165 | 166 | # Save results (image with detections) 167 | if save_img: 168 | if dataset.mode == 'image': 169 | save_path_rgb = save_path.split('.')[0] + '_rgb.' + save_path.split('.')[1] 170 | save_path_ir = save_path.split('.')[0] + '_ir.' + save_path.split('.')[1] 171 | print(save_path_rgb) 172 | cv2.imwrite(save_path_rgb, im0) 173 | cv2.imwrite(save_path_ir, im0_) 174 | else: # 'video' or 'stream' 175 | if vid_path != save_path: # new video 176 | vid_path = save_path 177 | if isinstance(vid_writer, cv2.VideoWriter): 178 | vid_writer.release() # release previous video writer 179 | if vid_cap: # video 180 | fps = vid_cap.get(cv2.CAP_PROP_FPS) 181 | w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH)) 182 | h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) 183 | else: # stream 184 | fps, w, h = 30, im0.shape[1], im0.shape[0] 185 | save_path += '.mp4' 186 | vid_writer = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (w, h)) 187 | vid_writer.write(im0) 188 | 189 | if save_txt or save_img: 190 | s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else '' 191 | print(f"Results saved to {save_dir}{s}") 192 | 193 | print(f'Done. ({time.time() - t0:.3f}s)') 194 | print(f'Average Speed: {fps_sum/img_num:.6f}Hz') 195 | 196 | 197 | if __name__ == '__main__': 198 | parser = argparse.ArgumentParser() 199 | parser.add_argument('--weights', nargs='+', type=str, default='/home/fqy/proj/multispectral-object-detection/best.pt', help='model.pt path(s)') 200 | parser.add_argument('--source1', type=str, default='/home/fqy/DATA/FLIR_ADAS_1_3/align/yolo/test/rgb/', help='source') # file/folder, 0 for webcam 201 | parser.add_argument('--source2', type=str, default='/home/fqy/DATA/FLIR_ADAS_1_3/align/yolo/test/ir', help='source') # file/folder, 0 for webcam 202 | parser.add_argument('--img-size', type=int, default=640, help='inference size (pixels)') 203 | parser.add_argument('--conf-thres', type=float, default=0.4, help='object confidence threshold') 204 | parser.add_argument('--iou-thres', type=float, default=0.45, help='IOU threshold for NMS') 205 | parser.add_argument('--device', default='0', help='cuda device, i.e. 0 or 0,1,2,3 or cpu') 206 | parser.add_argument('--view-img', default=False, action='store_true', help='display results') 207 | parser.add_argument('--save-txt', action='store_true', help='save results to *.txt') 208 | parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels') 209 | parser.add_argument('--save-crop', action='store_true', help='save cropped prediction boxes') 210 | parser.add_argument('--nosave', action='store_true', help='do not save images/videos') 211 | parser.add_argument('--classes', nargs='+', type=int, help='filter by class: --class 0, or --class 0 2 3') 212 | parser.add_argument('--agnostic-nms', action='store_true', help='class-agnostic NMS') 213 | parser.add_argument('--augment', action='store_true', help='augmented inference') 214 | parser.add_argument('--update', action='store_true', help='update all models') 215 | parser.add_argument('--project', default='runs/detect', help='save results to project/name') 216 | parser.add_argument('--name', default='exp', help='save results to project/name') 217 | parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment') 218 | parser.add_argument('--line-thickness', default=2, type=int, help='bounding box thickness (pixels)') 219 | parser.add_argument('--hide-labels', default=False, action='store_true', help='hide labels') 220 | parser.add_argument('--hide-conf', default=True, action='store_true', help='hide confidences') 221 | opt = parser.parse_args() 222 | print(opt) 223 | check_requirements(exclude=('pycocotools', 'thop')) 224 | 225 | with torch.no_grad(): 226 | if opt.update: # update all models (to fix SourceChangeWarning) 227 | for opt.weights in ['yolov5s.pt', 'yolov5m.pt', 'yolov5l.pt', 'yolov5x.pt']: 228 | detect(opt=opt) 229 | strip_optimizer(opt.weights) 230 | else: 231 | print("helloxxxxxxxxxxxxxxxxxxxx") 232 | detect(opt=opt) 233 | -------------------------------------------------------------------------------- /example.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/example.png -------------------------------------------------------------------------------- /global_var.py: -------------------------------------------------------------------------------- 1 | # @Time : 2021/6/22 下午4:46 2 | # @Author : Richard FANG 3 | # @File : global_var.py.py 4 | # @Software: PyCharm 5 | 6 | 7 | def _init(): # 初始化 8 | global _global_dict 9 | _global_dict = {} 10 | 11 | 12 | def set_value(key, value): 13 | # 定义一个全局变量 14 | _global_dict[key] = value 15 | 16 | 17 | def get_value(key): 18 | # 获得一个全局变量，不存在则提示读取对应变量失败 19 | try: 20 | return _global_dict[key] 21 | except: 22 | print('读取'+key+'失败\r\n') -------------------------------------------------------------------------------- /hubconf.py: -------------------------------------------------------------------------------- 1 | """YOLOv5 PyTorch Hub models https://pytorch.org/hub/ultralytics_yolov5/ 2 | 3 | Usage: 4 | import torch 5 | model = torch.hub.load('ultralytics/yolov5', 'yolov5s') 6 | """ 7 | 8 | from pathlib import Path 9 | 10 | import torch 11 | 12 | from models.yolo import Model 13 | from utils.general import check_requirements, set_logging 14 | from utils.google_utils import attempt_download 15 | from utils.torch_utils import select_device 16 | 17 | dependencies = ['torch', 'yaml'] 18 | check_requirements(Path(__file__).parent / 'requirements.txt', exclude=('pycocotools', 'thop')) 19 | 20 | 21 | def create(name, pretrained, channels, classes, autoshape, verbose): 22 | """Creates a specified YOLOv5 model 23 | 24 | Arguments: 25 | name (str): name of model, i.e. 'yolov5s' 26 | pretrained (bool): load pretrained weights into the model 27 | channels (int): number of input channels 28 | classes (int): number of model classes 29 | 30 | Returns: 31 | pytorch model 32 | """ 33 | try: 34 | set_logging(verbose=verbose) 35 | 36 | cfg = list((Path(__file__).parent / 'models').rglob(f'{name}.yaml'))[0] # model.yaml path 37 | model = Model(cfg, channels, classes) 38 | if pretrained: 39 | fname = f'{name}.pt' # checkpoint filename 40 | attempt_download(fname) # download if not found locally 41 | ckpt = torch.load(fname, map_location=torch.device('cpu')) # load 42 | msd = model.state_dict() # model state_dict 43 | csd = ckpt['model'].float().state_dict() # checkpoint state_dict as FP32 44 | csd = {k: v for k, v in csd.items() if msd[k].shape == v.shape} # filter 45 | model.load_state_dict(csd, strict=False) # load 46 | if len(ckpt['model'].names) == classes: 47 | model.names = ckpt['model'].names # set class names attribute 48 | if autoshape: 49 | model = model.autoshape() # for file/URI/PIL/cv2/np inputs and NMS 50 | device = select_device('0' if torch.cuda.is_available() else 'cpu') # default to GPU if available 51 | return model.to(device) 52 | 53 | except Exception as e: 54 | help_url = 'https://github.com/ultralytics/yolov5/issues/36' 55 | s = 'Cache maybe be out of date, try force_reload=True. See %s for help.' % help_url 56 | raise Exception(s) from e 57 | 58 | 59 | def custom(path_or_model='path/to/model.pt', autoshape=True, verbose=True): 60 | """YOLOv5-custom model https://github.com/ultralytics/yolov5 61 | 62 | Arguments (3 options): 63 | path_or_model (str): 'path/to/model.pt' 64 | path_or_model (dict): torch.load('path/to/model.pt') 65 | path_or_model (nn.Module): torch.load('path/to/model.pt')['model'] 66 | 67 | Returns: 68 | pytorch model 69 | """ 70 | set_logging(verbose=verbose) 71 | 72 | model = torch.load(path_or_model) if isinstance(path_or_model, str) else path_or_model # load checkpoint 73 | if isinstance(model, dict): 74 | model = model['ema' if model.get('ema') else 'model'] # load model 75 | 76 | hub_model = Model(model.yaml).to(next(model.parameters()).device) # create 77 | hub_model.load_state_dict(model.float().state_dict()) # load state_dict 78 | hub_model.names = model.names # class names 79 | if autoshape: 80 | hub_model = hub_model.autoshape() # for file/URI/PIL/cv2/np inputs and NMS 81 | device = select_device('0' if torch.cuda.is_available() else 'cpu') # default to GPU if available 82 | return hub_model.to(device) 83 | 84 | 85 | def yolov5s(pretrained=True, channels=3, classes=80, autoshape=True, verbose=True): 86 | # YOLOv5-small model https://github.com/ultralytics/yolov5 87 | return create('yolov5s', pretrained, channels, classes, autoshape, verbose) 88 | 89 | 90 | def yolov5m(pretrained=True, channels=3, classes=80, autoshape=True, verbose=True): 91 | # YOLOv5-medium model https://github.com/ultralytics/yolov5 92 | return create('yolov5m', pretrained, channels, classes, autoshape, verbose) 93 | 94 | 95 | def yolov5l(pretrained=True, channels=3, classes=80, autoshape=True, verbose=True): 96 | # YOLOv5-large model https://github.com/ultralytics/yolov5 97 | return create('yolov5l', pretrained, channels, classes, autoshape, verbose) 98 | 99 | 100 | def yolov5x(pretrained=True, channels=3, classes=80, autoshape=True, verbose=True): 101 | # YOLOv5-xlarge model https://github.com/ultralytics/yolov5 102 | return create('yolov5x', pretrained, channels, classes, autoshape, verbose) 103 | 104 | 105 | def yolov5s6(pretrained=True, channels=3, classes=80, autoshape=True, verbose=True): 106 | # YOLOv5-small-P6 model https://github.com/ultralytics/yolov5 107 | return create('yolov5s6', pretrained, channels, classes, autoshape, verbose) 108 | 109 | 110 | def yolov5m6(pretrained=True, channels=3, classes=80, autoshape=True, verbose=True): 111 | # YOLOv5-medium-P6 model https://github.com/ultralytics/yolov5 112 | return create('yolov5m6', pretrained, channels, classes, autoshape, verbose) 113 | 114 | 115 | def yolov5l6(pretrained=True, channels=3, classes=80, autoshape=True, verbose=True): 116 | # YOLOv5-large-P6 model https://github.com/ultralytics/yolov5 117 | return create('yolov5l6', pretrained, channels, classes, autoshape, verbose) 118 | 119 | 120 | def yolov5x6(pretrained=True, channels=3, classes=80, autoshape=True, verbose=True): 121 | # YOLOv5-xlarge-P6 model https://github.com/ultralytics/yolov5 122 | return create('yolov5x6', pretrained, channels, classes, autoshape, verbose) 123 | 124 | 125 | if __name__ == '__main__': 126 | model = create(name='yolov5s', pretrained=True, channels=3, classes=80, autoshape=True, verbose=True) # pretrained 127 | # model = custom(path_or_model='path/to/model.pt') # custom 128 | 129 | # Verify inference 130 | import cv2 131 | import numpy as np 132 | from PIL import Image 133 | 134 | imgs = ['data/images/zidane.jpg', # filename 135 | 'https://github.com/ultralytics/yolov5/releases/download/v1.0/zidane.jpg', # URI 136 | cv2.imread('data/images/bus.jpg')[:, :, ::-1], # OpenCV 137 | Image.open('data/images/bus.jpg'), # PIL 138 | np.zeros((320, 640, 3))] # numpy 139 | 140 | results = model(imgs) # batched inference 141 | results.print() 142 | results.save() 143 | -------------------------------------------------------------------------------- /models/__init__.py: -------------------------------------------------------------------------------- 1 | 2 | -------------------------------------------------------------------------------- /models/__pycache__/__init__.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/models/__pycache__/__init__.cpython-37.pyc -------------------------------------------------------------------------------- /models/__pycache__/common.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/models/__pycache__/common.cpython-37.pyc -------------------------------------------------------------------------------- /models/__pycache__/experimental.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/models/__pycache__/experimental.cpython-37.pyc -------------------------------------------------------------------------------- /models/__pycache__/yolo.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/models/__pycache__/yolo.cpython-37.pyc -------------------------------------------------------------------------------- /models/__pycache__/yolo_test.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/models/__pycache__/yolo_test.cpython-37.pyc -------------------------------------------------------------------------------- /models/experimental.py: -------------------------------------------------------------------------------- 1 | # YOLOv5 experimental modules 2 | 3 | import numpy as np 4 | import torch 5 | import torch.nn as nn 6 | 7 | from models.common import Conv, DWConv 8 | from utils.google_utils import attempt_download 9 | 10 | 11 | class CrossConv(nn.Module): 12 | # Cross Convolution Downsample 13 | def __init__(self, c1, c2, k=3, s=1, g=1, e=1.0, shortcut=False): 14 | # ch_in, ch_out, kernel, stride, groups, expansion, shortcut 15 | super(CrossConv, self).__init__() 16 | c_ = int(c2 * e) # hidden channels 17 | self.cv1 = Conv(c1, c_, (1, k), (1, s)) 18 | self.cv2 = Conv(c_, c2, (k, 1), (s, 1), g=g) 19 | self.add = shortcut and c1 == c2 20 | 21 | def forward(self, x): 22 | return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x)) 23 | 24 | 25 | class Sum(nn.Module): 26 | # Weighted sum of 2 or more layers https://arxiv.org/abs/1911.09070 27 | def __init__(self, n, weight=False): # n: number of inputs 28 | super(Sum, self).__init__() 29 | self.weight = weight # apply weights boolean 30 | self.iter = range(n - 1) # iter object 31 | if weight: 32 | self.w = nn.Parameter(-torch.arange(1., n) / 2, requires_grad=True) # layer weights 33 | 34 | def forward(self, x): 35 | y = x[0] # no weight 36 | if self.weight: 37 | w = torch.sigmoid(self.w) * 2 38 | for i in self.iter: 39 | y = y + x[i + 1] * w[i] 40 | else: 41 | for i in self.iter: 42 | y = y + x[i + 1] 43 | return y 44 | 45 | 46 | class GhostConv(nn.Module): 47 | # Ghost Convolution https://github.com/huawei-noah/ghostnet 48 | def __init__(self, c1, c2, k=1, s=1, g=1, act=True): # ch_in, ch_out, kernel, stride, groups 49 | super(GhostConv, self).__init__() 50 | c_ = c2 // 2 # hidden channels 51 | self.cv1 = Conv(c1, c_, k, s, None, g, act) 52 | self.cv2 = Conv(c_, c_, 5, 1, None, c_, act) 53 | 54 | def forward(self, x): 55 | y = self.cv1(x) 56 | return torch.cat([y, self.cv2(y)], 1) 57 | 58 | 59 | class GhostBottleneck(nn.Module): 60 | # Ghost Bottleneck https://github.com/huawei-noah/ghostnet 61 | def __init__(self, c1, c2, k=3, s=1): # ch_in, ch_out, kernel, stride 62 | super(GhostBottleneck, self).__init__() 63 | c_ = c2 // 2 64 | self.conv = nn.Sequential(GhostConv(c1, c_, 1, 1), # pw 65 | DWConv(c_, c_, k, s, act=False) if s == 2 else nn.Identity(), # dw 66 | GhostConv(c_, c2, 1, 1, act=False)) # pw-linear 67 | self.shortcut = nn.Sequential(DWConv(c1, c1, k, s, act=False), 68 | Conv(c1, c2, 1, 1, act=False)) if s == 2 else nn.Identity() 69 | 70 | def forward(self, x): 71 | return self.conv(x) + self.shortcut(x) 72 | 73 | 74 | class MixConv2d(nn.Module): 75 | # Mixed Depthwise Conv https://arxiv.org/abs/1907.09595 76 | def __init__(self, c1, c2, k=(1, 3), s=1, equal_ch=True): 77 | super(MixConv2d, self).__init__() 78 | groups = len(k) 79 | if equal_ch: # equal c_ per group 80 | i = torch.linspace(0, groups - 1E-6, c2).floor() # c2 indices 81 | c_ = [(i == g).sum() for g in range(groups)] # intermediate channels 82 | else: # equal weight.numel() per group 83 | b = [c2] + [0] * groups 84 | a = np.eye(groups + 1, groups, k=-1) 85 | a -= np.roll(a, 1, axis=1) 86 | a *= np.array(k) ** 2 87 | a[0] = 1 88 | c_ = np.linalg.lstsq(a, b, rcond=None)[0].round() # solve for equal weight indices, ax = b 89 | 90 | self.m = nn.ModuleList([nn.Conv2d(c1, int(c_[g]), k[g], s, k[g] // 2, bias=False) for g in range(groups)]) 91 | self.bn = nn.BatchNorm2d(c2) 92 | self.act = nn.LeakyReLU(0.1, inplace=True) 93 | 94 | def forward(self, x): 95 | return x + self.act(self.bn(torch.cat([m(x) for m in self.m], 1))) 96 | 97 | 98 | class Ensemble(nn.ModuleList): 99 | # Ensemble of models 100 | def __init__(self): 101 | super(Ensemble, self).__init__() 102 | 103 | def forward(self, x, augment=False): 104 | y = [] 105 | for module in self: 106 | y.append(module(x, augment)[0]) 107 | # y = torch.stack(y).max(0)[0] # max ensemble 108 | # y = torch.stack(y).mean(0) # mean ensemble 109 | y = torch.cat(y, 1) # nms ensemble 110 | return y, None # inference, train output 111 | 112 | 113 | def attempt_load(weights, map_location=None): 114 | # Loads an ensemble of models weights=[a,b,c] or a single model weights=[a] or weights=a 115 | model = Ensemble() 116 | for w in weights if isinstance(weights, list) else [weights]: 117 | attempt_download(w) 118 | ckpt = torch.load(w, map_location=map_location) # load 119 | model.append(ckpt['ema' if ckpt.get('ema') else 'model'].float().fuse().eval()) # FP32 model 120 | 121 | # Compatibility updates 122 | for m in model.modules(): 123 | if type(m) in [nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6, nn.SiLU]: 124 | m.inplace = True # pytorch 1.7.0 compatibility 125 | elif type(m) is Conv: 126 | m._non_persistent_buffers_set = set() # pytorch 1.6.0 compatibility 127 | 128 | if len(model) == 1: 129 | return model[-1] # return model 130 | else: 131 | print('Ensemble created with %s\n' % weights) 132 | for k in ['names', 'stride']: 133 | setattr(model, k, getattr(model[-1], k)) 134 | return model # return ensemble 135 | -------------------------------------------------------------------------------- /models/export.py: -------------------------------------------------------------------------------- 1 | """Exports a YOLOv5 *.pt model to ONNX and TorchScript formats 2 | 3 | Usage: 4 | $ export PYTHONPATH="$PWD" && python models/export.py --weights yolov5s.pt --img 640 --batch 1 5 | """ 6 | 7 | import argparse 8 | import sys 9 | import time 10 | from pathlib import Path 11 | 12 | sys.path.append(Path(__file__).parent.parent.absolute().__str__()) # to run '$ python *.py' files in subdirectories 13 | 14 | import torch 15 | import torch.nn as nn 16 | from torch.utils.mobile_optimizer import optimize_for_mobile 17 | 18 | import models 19 | from models.experimental import attempt_load 20 | from utils.activations import Hardswish, SiLU 21 | from utils.general import colorstr, check_img_size, check_requirements, file_size, set_logging 22 | from utils.torch_utils import select_device 23 | 24 | if __name__ == '__main__': 25 | parser = argparse.ArgumentParser() 26 | parser.add_argument('--weights', type=str, default='./yolov5s.pt', help='weights path') 27 | parser.add_argument('--img-size', nargs='+', type=int, default=[640, 640], help='image size') # height, width 28 | parser.add_argument('--batch-size', type=int, default=1, help='batch size') 29 | parser.add_argument('--grid', action='store_true', help='export Detect() layer grid') 30 | parser.add_argument('--device', default='cpu', help='cuda device, i.e. 0 or 0,1,2,3 or cpu') 31 | parser.add_argument('--dynamic', action='store_true', help='dynamic ONNX axes') # ONNX-only 32 | parser.add_argument('--simplify', action='store_true', help='simplify ONNX model') # ONNX-only 33 | opt = parser.parse_args() 34 | opt.img_size *= 2 if len(opt.img_size) == 1 else 1 # expand 35 | print(opt) 36 | set_logging() 37 | t = time.time() 38 | 39 | # Load PyTorch model 40 | device = select_device(opt.device) 41 | model = attempt_load(opt.weights, map_location=device) # load FP32 model 42 | labels = model.names 43 | 44 | # Checks 45 | gs = int(max(model.stride)) # grid size (max stride) 46 | opt.img_size = [check_img_size(x, gs) for x in opt.img_size] # verify img_size are gs-multiples 47 | 48 | # Input 49 | img = torch.zeros(opt.batch_size, 3, *opt.img_size).to(device) # image size(1,3,320,192) iDetection 50 | 51 | # Update model 52 | for k, m in model.named_modules(): 53 | m._non_persistent_buffers_set = set() # pytorch 1.6.0 compatibility 54 | if isinstance(m, models.common.Conv): # assign export-friendly activations 55 | if isinstance(m.act, nn.Hardswish): 56 | m.act = Hardswish() 57 | elif isinstance(m.act, nn.SiLU): 58 | m.act = SiLU() 59 | # elif isinstance(m, models.yolo.Detect): 60 | # m.forward = m.forward_export # assign forward (optional) 61 | model.model[-1].export = not opt.grid # set Detect() layer grid export 62 | for _ in range(2): 63 | y = model(img) # dry runs 64 | print(f"\n{colorstr('PyTorch:')} starting from {opt.weights} ({file_size(opt.weights):.1f} MB)") 65 | 66 | # TorchScript export ----------------------------------------------------------------------------------------------- 67 | prefix = colorstr('TorchScript:') 68 | try: 69 | print(f'\n{prefix} starting export with torch {torch.__version__}...') 70 | f = opt.weights.replace('.pt', '.torchscript.pt') # filename 71 | ts = torch.jit.trace(model, img, strict=False) 72 | ts = optimize_for_mobile(ts) # https://pytorch.org/tutorials/recipes/script_optimized.html 73 | ts.save(f) 74 | print(f'{prefix} export success, saved as {f} ({file_size(f):.1f} MB)') 75 | except Exception as e: 76 | print(f'{prefix} export failure: {e}') 77 | 78 | # ONNX export ------------------------------------------------------------------------------------------------------ 79 | prefix = colorstr('ONNX:') 80 | try: 81 | import onnx 82 | 83 | print(f'{prefix} starting export with onnx {onnx.__version__}...') 84 | f = opt.weights.replace('.pt', '.onnx') # filename 85 | torch.onnx.export(model, img, f, verbose=False, opset_version=12, input_names=['images'], 86 | dynamic_axes={'images': {0: 'batch', 2: 'height', 3: 'width'}, # size(1,3,640,640) 87 | 'output': {0: 'batch', 2: 'y', 3: 'x'}} if opt.dynamic else None) 88 | 89 | # Checks 90 | model_onnx = onnx.load(f) # load onnx model 91 | onnx.checker.check_model(model_onnx) # check onnx model 92 | # print(onnx.helper.printable_graph(model_onnx.graph)) # print 93 | 94 | # Simplify 95 | if opt.simplify: 96 | try: 97 | check_requirements(['onnx-simplifier']) 98 | import onnxsim 99 | 100 | print(f'{prefix} simplifying with onnx-simplifier {onnxsim.__version__}...') 101 | model_onnx, check = onnxsim.simplify(model_onnx, 102 | dynamic_input_shape=opt.dynamic, 103 | input_shapes={'images': list(img.shape)} if opt.dynamic else None) 104 | assert check, 'assert check failed' 105 | onnx.save(model_onnx, f) 106 | except Exception as e: 107 | print(f'{prefix} simplifier failure: {e}') 108 | print(f'{prefix} export success, saved as {f} ({file_size(f):.1f} MB)') 109 | except Exception as e: 110 | print(f'{prefix} export failure: {e}') 111 | 112 | # CoreML export ---------------------------------------------------------------------------------------------------- 113 | prefix = colorstr('CoreML:') 114 | try: 115 | import coremltools as ct 116 | 117 | print(f'{prefix} starting export with coremltools {ct.__version__}...') 118 | # convert model from torchscript and apply pixel scaling as per detect.py 119 | model = ct.convert(ts, inputs=[ct.ImageType(name='image', shape=img.shape, scale=1 / 255.0, bias=[0, 0, 0])]) 120 | f = opt.weights.replace('.pt', '.mlmodel') # filename 121 | model.save(f) 122 | print(f'{prefix} export success, saved as {f} ({file_size(f):.1f} MB)') 123 | except Exception as e: 124 | print(f'{prefix} export failure: {e}') 125 | 126 | # Finish 127 | print(f'\nExport complete ({time.time() - t:.2f}s). Visualize with https://github.com/lutzroeder/netron.') 128 | -------------------------------------------------------------------------------- /models/hub/anchors.yaml: -------------------------------------------------------------------------------- 1 | # Default YOLOv5 anchors for COCO data 2 | 3 | 4 | # P5 ------------------------------------------------------------------------------------------------------------------- 5 | # P5-640: 6 | anchors_p5_640: 7 | - [ 10,13, 16,30, 33,23 ] # P3/8 8 | - [ 30,61, 62,45, 59,119 ] # P4/16 9 | - [ 116,90, 156,198, 373,326 ] # P5/32 10 | 11 | 12 | # P6 ------------------------------------------------------------------------------------------------------------------- 13 | # P6-640: thr=0.25: 0.9964 BPR, 5.54 anchors past thr, n=12, img_size=640, metric_all=0.281/0.716-mean/best, past_thr=0.469-mean: 9,11, 21,19, 17,41, 43,32, 39,70, 86,64, 65,131, 134,130, 120,265, 282,180, 247,354, 512,387 14 | anchors_p6_640: 15 | - [ 9,11, 21,19, 17,41 ] # P3/8 16 | - [ 43,32, 39,70, 86,64 ] # P4/16 17 | - [ 65,131, 134,130, 120,265 ] # P5/32 18 | - [ 282,180, 247,354, 512,387 ] # P6/64 19 | 20 | # P6-1280: thr=0.25: 0.9950 BPR, 5.55 anchors past thr, n=12, img_size=1280, metric_all=0.281/0.714-mean/best, past_thr=0.468-mean: 19,27, 44,40, 38,94, 96,68, 86,152, 180,137, 140,301, 303,264, 238,542, 436,615, 739,380, 925,792 21 | anchors_p6_1280: 22 | - [ 19,27, 44,40, 38,94 ] # P3/8 23 | - [ 96,68, 86,152, 180,137 ] # P4/16 24 | - [ 140,301, 303,264, 238,542 ] # P5/32 25 | - [ 436,615, 739,380, 925,792 ] # P6/64 26 | 27 | # P6-1920: thr=0.25: 0.9950 BPR, 5.55 anchors past thr, n=12, img_size=1920, metric_all=0.281/0.714-mean/best, past_thr=0.468-mean: 28,41, 67,59, 57,141, 144,103, 129,227, 270,205, 209,452, 455,396, 358,812, 653,922, 1109,570, 1387,1187 28 | anchors_p6_1920: 29 | - [ 28,41, 67,59, 57,141 ] # P3/8 30 | - [ 144,103, 129,227, 270,205 ] # P4/16 31 | - [ 209,452, 455,396, 358,812 ] # P5/32 32 | - [ 653,922, 1109,570, 1387,1187 ] # P6/64 33 | 34 | 35 | # P7 ------------------------------------------------------------------------------------------------------------------- 36 | # P7-640: thr=0.25: 0.9962 BPR, 6.76 anchors past thr, n=15, img_size=640, metric_all=0.275/0.733-mean/best, past_thr=0.466-mean: 11,11, 13,30, 29,20, 30,46, 61,38, 39,92, 78,80, 146,66, 79,163, 149,150, 321,143, 157,303, 257,402, 359,290, 524,372 37 | anchors_p7_640: 38 | - [ 11,11, 13,30, 29,20 ] # P3/8 39 | - [ 30,46, 61,38, 39,92 ] # P4/16 40 | - [ 78,80, 146,66, 79,163 ] # P5/32 41 | - [ 149,150, 321,143, 157,303 ] # P6/64 42 | - [ 257,402, 359,290, 524,372 ] # P7/128 43 | 44 | # P7-1280: thr=0.25: 0.9968 BPR, 6.71 anchors past thr, n=15, img_size=1280, metric_all=0.273/0.732-mean/best, past_thr=0.463-mean: 19,22, 54,36, 32,77, 70,83, 138,71, 75,173, 165,159, 148,334, 375,151, 334,317, 251,626, 499,474, 750,326, 534,814, 1079,818 45 | anchors_p7_1280: 46 | - [ 19,22, 54,36, 32,77 ] # P3/8 47 | - [ 70,83, 138,71, 75,173 ] # P4/16 48 | - [ 165,159, 148,334, 375,151 ] # P5/32 49 | - [ 334,317, 251,626, 499,474 ] # P6/64 50 | - [ 750,326, 534,814, 1079,818 ] # P7/128 51 | 52 | # P7-1920: thr=0.25: 0.9968 BPR, 6.71 anchors past thr, n=15, img_size=1920, metric_all=0.273/0.732-mean/best, past_thr=0.463-mean: 29,34, 81,55, 47,115, 105,124, 207,107, 113,259, 247,238, 222,500, 563,227, 501,476, 376,939, 749,711, 1126,489, 801,1222, 1618,1227 53 | anchors_p7_1920: 54 | - [ 29,34, 81,55, 47,115 ] # P3/8 55 | - [ 105,124, 207,107, 113,259 ] # P4/16 56 | - [ 247,238, 222,500, 563,227 ] # P5/32 57 | - [ 501,476, 376,939, 749,711 ] # P6/64 58 | - [ 1126,489, 801,1222, 1618,1227 ] # P7/128 59 | -------------------------------------------------------------------------------- /models/hub/yolov3-spp.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # darknet53 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [32, 3, 1]], # 0 16 | [-1, 1, Conv, [64, 3, 2]], # 1-P1/2 17 | [-1, 1, Bottleneck, [64]], 18 | [-1, 1, Conv, [128, 3, 2]], # 3-P2/4 19 | [-1, 2, Bottleneck, [128]], 20 | [-1, 1, Conv, [256, 3, 2]], # 5-P3/8 21 | [-1, 8, Bottleneck, [256]], 22 | [-1, 1, Conv, [512, 3, 2]], # 7-P4/16 23 | [-1, 8, Bottleneck, [512]], 24 | [-1, 1, Conv, [1024, 3, 2]], # 9-P5/32 25 | [-1, 4, Bottleneck, [1024]], # 10 26 | ] 27 | 28 | # YOLOv3-SPP head 29 | head: 30 | [[-1, 1, Bottleneck, [1024, False]], 31 | [-1, 1, SPP, [512, [5, 9, 13]]], 32 | [-1, 1, Conv, [1024, 3, 1]], 33 | [-1, 1, Conv, [512, 1, 1]], 34 | [-1, 1, Conv, [1024, 3, 1]], # 15 (P5/32-large) 35 | 36 | [-2, 1, Conv, [256, 1, 1]], 37 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 38 | [[-1, 8], 1, Concat, [1]], # cat backbone P4 39 | [-1, 1, Bottleneck, [512, False]], 40 | [-1, 1, Bottleneck, [512, False]], 41 | [-1, 1, Conv, [256, 1, 1]], 42 | [-1, 1, Conv, [512, 3, 1]], # 22 (P4/16-medium) 43 | 44 | [-2, 1, Conv, [128, 1, 1]], 45 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 46 | [[-1, 6], 1, Concat, [1]], # cat backbone P3 47 | [-1, 1, Bottleneck, [256, False]], 48 | [-1, 2, Bottleneck, [256, False]], # 27 (P3/8-small) 49 | 50 | [[27, 22, 15], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 51 | ] 52 | -------------------------------------------------------------------------------- /models/hub/yolov3-tiny.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,14, 23,27, 37,58] # P4/16 9 | - [81,82, 135,169, 344,319] # P5/32 10 | 11 | # YOLOv3-tiny backbone 12 | backbone: 13 | # [from, number, module, args] 14 | [[-1, 1, Conv, [16, 3, 1]], # 0 15 | [-1, 1, nn.MaxPool2d, [2, 2, 0]], # 1-P1/2 16 | [-1, 1, Conv, [32, 3, 1]], 17 | [-1, 1, nn.MaxPool2d, [2, 2, 0]], # 3-P2/4 18 | [-1, 1, Conv, [64, 3, 1]], 19 | [-1, 1, nn.MaxPool2d, [2, 2, 0]], # 5-P3/8 20 | [-1, 1, Conv, [128, 3, 1]], 21 | [-1, 1, nn.MaxPool2d, [2, 2, 0]], # 7-P4/16 22 | [-1, 1, Conv, [256, 3, 1]], 23 | [-1, 1, nn.MaxPool2d, [2, 2, 0]], # 9-P5/32 24 | [-1, 1, Conv, [512, 3, 1]], 25 | [-1, 1, nn.ZeroPad2d, [[0, 1, 0, 1]]], # 11 26 | [-1, 1, nn.MaxPool2d, [2, 1, 0]], # 12 27 | ] 28 | 29 | # YOLOv3-tiny head 30 | head: 31 | [[-1, 1, Conv, [1024, 3, 1]], 32 | [-1, 1, Conv, [256, 1, 1]], 33 | [-1, 1, Conv, [512, 3, 1]], # 15 (P5/32-large) 34 | 35 | [-2, 1, Conv, [128, 1, 1]], 36 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 37 | [[-1, 8], 1, Concat, [1]], # cat backbone P4 38 | [-1, 1, Conv, [256, 3, 1]], # 19 (P4/16-medium) 39 | 40 | [[19, 15], 1, Detect, [nc, anchors]], # Detect(P4, P5) 41 | ] 42 | -------------------------------------------------------------------------------- /models/hub/yolov3.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # darknet53 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [32, 3, 1]], # 0 16 | [-1, 1, Conv, [64, 3, 2]], # 1-P1/2 17 | [-1, 1, Bottleneck, [64]], 18 | [-1, 1, Conv, [128, 3, 2]], # 3-P2/4 19 | [-1, 2, Bottleneck, [128]], 20 | [-1, 1, Conv, [256, 3, 2]], # 5-P3/8 21 | [-1, 8, Bottleneck, [256]], 22 | [-1, 1, Conv, [512, 3, 2]], # 7-P4/16 23 | [-1, 8, Bottleneck, [512]], 24 | [-1, 1, Conv, [1024, 3, 2]], # 9-P5/32 25 | [-1, 4, Bottleneck, [1024]], # 10 26 | ] 27 | 28 | # YOLOv3 head 29 | head: 30 | [[-1, 1, Bottleneck, [1024, False]], 31 | [-1, 1, Conv, [512, [1, 1]]], 32 | [-1, 1, Conv, [1024, 3, 1]], 33 | [-1, 1, Conv, [512, 1, 1]], 34 | [-1, 1, Conv, [1024, 3, 1]], # 15 (P5/32-large) 35 | 36 | [-2, 1, Conv, [256, 1, 1]], 37 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 38 | [[-1, 8], 1, Concat, [1]], # cat backbone P4 39 | [-1, 1, Bottleneck, [512, False]], 40 | [-1, 1, Bottleneck, [512, False]], 41 | [-1, 1, Conv, [256, 1, 1]], 42 | [-1, 1, Conv, [512, 3, 1]], # 22 (P4/16-medium) 43 | 44 | [-2, 1, Conv, [128, 1, 1]], 45 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 46 | [[-1, 6], 1, Concat, [1]], # cat backbone P3 47 | [-1, 1, Bottleneck, [256, False]], 48 | [-1, 2, Bottleneck, [256, False]], # 27 (P3/8-small) 49 | 50 | [[27, 22, 15], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 51 | ] 52 | -------------------------------------------------------------------------------- /models/hub/yolov5-fpn.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, Bottleneck, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 6, BottleneckCSP, [1024]], # 9 25 | ] 26 | 27 | # YOLOv5 FPN head 28 | head: 29 | [[-1, 3, BottleneckCSP, [1024, False]], # 10 (P5/32-large) 30 | 31 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 32 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 33 | [-1, 1, Conv, [512, 1, 1]], 34 | [-1, 3, BottleneckCSP, [512, False]], # 14 (P4/16-medium) 35 | 36 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 37 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 38 | [-1, 1, Conv, [256, 1, 1]], 39 | [-1, 3, BottleneckCSP, [256, False]], # 18 (P3/8-small) 40 | 41 | [[18, 14, 10], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 42 | ] 43 | -------------------------------------------------------------------------------- /models/hub/yolov5-p2.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 3 8 | 9 | # YOLOv5 backbone 10 | backbone: 11 | # [from, number, module, args] 12 | [ [ -1, 1, Focus, [ 64, 3 ] ], # 0-P1/2 13 | [ -1, 1, Conv, [ 128, 3, 2 ] ], # 1-P2/4 14 | [ -1, 3, C3, [ 128 ] ], 15 | [ -1, 1, Conv, [ 256, 3, 2 ] ], # 3-P3/8 16 | [ -1, 9, C3, [ 256 ] ], 17 | [ -1, 1, Conv, [ 512, 3, 2 ] ], # 5-P4/16 18 | [ -1, 9, C3, [ 512 ] ], 19 | [ -1, 1, Conv, [ 1024, 3, 2 ] ], # 7-P5/32 20 | [ -1, 1, SPP, [ 1024, [ 5, 9, 13 ] ] ], 21 | [ -1, 3, C3, [ 1024, False ] ], # 9 22 | ] 23 | 24 | # YOLOv5 head 25 | head: 26 | [ [ -1, 1, Conv, [ 512, 1, 1 ] ], 27 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 28 | [ [ -1, 6 ], 1, Concat, [ 1 ] ], # cat backbone P4 29 | [ -1, 3, C3, [ 512, False ] ], # 13 30 | 31 | [ -1, 1, Conv, [ 256, 1, 1 ] ], 32 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 33 | [ [ -1, 4 ], 1, Concat, [ 1 ] ], # cat backbone P3 34 | [ -1, 3, C3, [ 256, False ] ], # 17 (P3/8-small) 35 | 36 | [ -1, 1, Conv, [ 128, 1, 1 ] ], 37 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 38 | [ [ -1, 2 ], 1, Concat, [ 1 ] ], # cat backbone P2 39 | [ -1, 1, C3, [ 128, False ] ], # 21 (P2/4-xsmall) 40 | 41 | [ -1, 1, Conv, [ 128, 3, 2 ] ], 42 | [ [ -1, 18 ], 1, Concat, [ 1 ] ], # cat head P3 43 | [ -1, 3, C3, [ 256, False ] ], # 24 (P3/8-small) 44 | 45 | [ -1, 1, Conv, [ 256, 3, 2 ] ], 46 | [ [ -1, 14 ], 1, Concat, [ 1 ] ], # cat head P4 47 | [ -1, 3, C3, [ 512, False ] ], # 27 (P4/16-medium) 48 | 49 | [ -1, 1, Conv, [ 512, 3, 2 ] ], 50 | [ [ -1, 10 ], 1, Concat, [ 1 ] ], # cat head P5 51 | [ -1, 3, C3, [ 1024, False ] ], # 30 (P5/32-large) 52 | 53 | [ [ 24, 27, 30 ], 1, Detect, [ nc, anchors ] ], # Detect(P3, P4, P5) 54 | ] 55 | -------------------------------------------------------------------------------- /models/hub/yolov5-p6.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 3 8 | 9 | # YOLOv5 backbone 10 | backbone: 11 | # [from, number, module, args] 12 | [ [ -1, 1, Focus, [ 64, 3 ] ], # 0-P1/2 13 | [ -1, 1, Conv, [ 128, 3, 2 ] ], # 1-P2/4 14 | [ -1, 3, C3, [ 128 ] ], 15 | [ -1, 1, Conv, [ 256, 3, 2 ] ], # 3-P3/8 16 | [ -1, 9, C3, [ 256 ] ], 17 | [ -1, 1, Conv, [ 512, 3, 2 ] ], # 5-P4/16 18 | [ -1, 9, C3, [ 512 ] ], 19 | [ -1, 1, Conv, [ 768, 3, 2 ] ], # 7-P5/32 20 | [ -1, 3, C3, [ 768 ] ], 21 | [ -1, 1, Conv, [ 1024, 3, 2 ] ], # 9-P6/64 22 | [ -1, 1, SPP, [ 1024, [ 3, 5, 7 ] ] ], 23 | [ -1, 3, C3, [ 1024, False ] ], # 11 24 | ] 25 | 26 | # YOLOv5 head 27 | head: 28 | [ [ -1, 1, Conv, [ 768, 1, 1 ] ], 29 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 30 | [ [ -1, 8 ], 1, Concat, [ 1 ] ], # cat backbone P5 31 | [ -1, 3, C3, [ 768, False ] ], # 15 32 | 33 | [ -1, 1, Conv, [ 512, 1, 1 ] ], 34 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 35 | [ [ -1, 6 ], 1, Concat, [ 1 ] ], # cat backbone P4 36 | [ -1, 3, C3, [ 512, False ] ], # 19 37 | 38 | [ -1, 1, Conv, [ 256, 1, 1 ] ], 39 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 40 | [ [ -1, 4 ], 1, Concat, [ 1 ] ], # cat backbone P3 41 | [ -1, 3, C3, [ 256, False ] ], # 23 (P3/8-small) 42 | 43 | [ -1, 1, Conv, [ 256, 3, 2 ] ], 44 | [ [ -1, 20 ], 1, Concat, [ 1 ] ], # cat head P4 45 | [ -1, 3, C3, [ 512, False ] ], # 26 (P4/16-medium) 46 | 47 | [ -1, 1, Conv, [ 512, 3, 2 ] ], 48 | [ [ -1, 16 ], 1, Concat, [ 1 ] ], # cat head P5 49 | [ -1, 3, C3, [ 768, False ] ], # 29 (P5/32-large) 50 | 51 | [ -1, 1, Conv, [ 768, 3, 2 ] ], 52 | [ [ -1, 12 ], 1, Concat, [ 1 ] ], # cat head P6 53 | [ -1, 3, C3, [ 1024, False ] ], # 32 (P5/64-xlarge) 54 | 55 | [ [ 23, 26, 29, 32 ], 1, Detect, [ nc, anchors ] ], # Detect(P3, P4, P5, P6) 56 | ] 57 | -------------------------------------------------------------------------------- /models/hub/yolov5-p7.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 3 8 | 9 | # YOLOv5 backbone 10 | backbone: 11 | # [from, number, module, args] 12 | [ [ -1, 1, Focus, [ 64, 3 ] ], # 0-P1/2 13 | [ -1, 1, Conv, [ 128, 3, 2 ] ], # 1-P2/4 14 | [ -1, 3, C3, [ 128 ] ], 15 | [ -1, 1, Conv, [ 256, 3, 2 ] ], # 3-P3/8 16 | [ -1, 9, C3, [ 256 ] ], 17 | [ -1, 1, Conv, [ 512, 3, 2 ] ], # 5-P4/16 18 | [ -1, 9, C3, [ 512 ] ], 19 | [ -1, 1, Conv, [ 768, 3, 2 ] ], # 7-P5/32 20 | [ -1, 3, C3, [ 768 ] ], 21 | [ -1, 1, Conv, [ 1024, 3, 2 ] ], # 9-P6/64 22 | [ -1, 3, C3, [ 1024 ] ], 23 | [ -1, 1, Conv, [ 1280, 3, 2 ] ], # 11-P7/128 24 | [ -1, 1, SPP, [ 1280, [ 3, 5 ] ] ], 25 | [ -1, 3, C3, [ 1280, False ] ], # 13 26 | ] 27 | 28 | # YOLOv5 head 29 | head: 30 | [ [ -1, 1, Conv, [ 1024, 1, 1 ] ], 31 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 32 | [ [ -1, 10 ], 1, Concat, [ 1 ] ], # cat backbone P6 33 | [ -1, 3, C3, [ 1024, False ] ], # 17 34 | 35 | [ -1, 1, Conv, [ 768, 1, 1 ] ], 36 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 37 | [ [ -1, 8 ], 1, Concat, [ 1 ] ], # cat backbone P5 38 | [ -1, 3, C3, [ 768, False ] ], # 21 39 | 40 | [ -1, 1, Conv, [ 512, 1, 1 ] ], 41 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 42 | [ [ -1, 6 ], 1, Concat, [ 1 ] ], # cat backbone P4 43 | [ -1, 3, C3, [ 512, False ] ], # 25 44 | 45 | [ -1, 1, Conv, [ 256, 1, 1 ] ], 46 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 47 | [ [ -1, 4 ], 1, Concat, [ 1 ] ], # cat backbone P3 48 | [ -1, 3, C3, [ 256, False ] ], # 29 (P3/8-small) 49 | 50 | [ -1, 1, Conv, [ 256, 3, 2 ] ], 51 | [ [ -1, 26 ], 1, Concat, [ 1 ] ], # cat head P4 52 | [ -1, 3, C3, [ 512, False ] ], # 32 (P4/16-medium) 53 | 54 | [ -1, 1, Conv, [ 512, 3, 2 ] ], 55 | [ [ -1, 22 ], 1, Concat, [ 1 ] ], # cat head P5 56 | [ -1, 3, C3, [ 768, False ] ], # 35 (P5/32-large) 57 | 58 | [ -1, 1, Conv, [ 768, 3, 2 ] ], 59 | [ [ -1, 18 ], 1, Concat, [ 1 ] ], # cat head P6 60 | [ -1, 3, C3, [ 1024, False ] ], # 38 (P6/64-xlarge) 61 | 62 | [ -1, 1, Conv, [ 1024, 3, 2 ] ], 63 | [ [ -1, 14 ], 1, Concat, [ 1 ] ], # cat head P7 64 | [ -1, 3, C3, [ 1280, False ] ], # 41 (P7/128-xxlarge) 65 | 66 | [ [ 29, 32, 35, 38, 41 ], 1, Detect, [ nc, anchors ] ], # Detect(P3, P4, P5, P6, P7) 67 | ] 68 | -------------------------------------------------------------------------------- /models/hub/yolov5-panet.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, BottleneckCSP, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, BottleneckCSP, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 PANet head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, BottleneckCSP, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, BottleneckCSP, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, BottleneckCSP, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, BottleneckCSP, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /models/hub/yolov5l6.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [ 19,27, 44,40, 38,94 ] # P3/8 9 | - [ 96,68, 86,152, 180,137 ] # P4/16 10 | - [ 140,301, 303,264, 238,542 ] # P5/32 11 | - [ 436,615, 739,380, 925,792 ] # P6/64 12 | 13 | # YOLOv5 backbone 14 | backbone: 15 | # [from, number, module, args] 16 | [ [ -1, 1, Focus, [ 64, 3 ] ], # 0-P1/2 17 | [ -1, 1, Conv, [ 128, 3, 2 ] ], # 1-P2/4 18 | [ -1, 3, C3, [ 128 ] ], 19 | [ -1, 1, Conv, [ 256, 3, 2 ] ], # 3-P3/8 20 | [ -1, 9, C3, [ 256 ] ], 21 | [ -1, 1, Conv, [ 512, 3, 2 ] ], # 5-P4/16 22 | [ -1, 9, C3, [ 512 ] ], 23 | [ -1, 1, Conv, [ 768, 3, 2 ] ], # 7-P5/32 24 | [ -1, 3, C3, [ 768 ] ], 25 | [ -1, 1, Conv, [ 1024, 3, 2 ] ], # 9-P6/64 26 | [ -1, 1, SPP, [ 1024, [ 3, 5, 7 ] ] ], 27 | [ -1, 3, C3, [ 1024, False ] ], # 11 28 | ] 29 | 30 | # YOLOv5 head 31 | head: 32 | [ [ -1, 1, Conv, [ 768, 1, 1 ] ], 33 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 34 | [ [ -1, 8 ], 1, Concat, [ 1 ] ], # cat backbone P5 35 | [ -1, 3, C3, [ 768, False ] ], # 15 36 | 37 | [ -1, 1, Conv, [ 512, 1, 1 ] ], 38 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 39 | [ [ -1, 6 ], 1, Concat, [ 1 ] ], # cat backbone P4 40 | [ -1, 3, C3, [ 512, False ] ], # 19 41 | 42 | [ -1, 1, Conv, [ 256, 1, 1 ] ], 43 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 44 | [ [ -1, 4 ], 1, Concat, [ 1 ] ], # cat backbone P3 45 | [ -1, 3, C3, [ 256, False ] ], # 23 (P3/8-small) 46 | 47 | [ -1, 1, Conv, [ 256, 3, 2 ] ], 48 | [ [ -1, 20 ], 1, Concat, [ 1 ] ], # cat head P4 49 | [ -1, 3, C3, [ 512, False ] ], # 26 (P4/16-medium) 50 | 51 | [ -1, 1, Conv, [ 512, 3, 2 ] ], 52 | [ [ -1, 16 ], 1, Concat, [ 1 ] ], # cat head P5 53 | [ -1, 3, C3, [ 768, False ] ], # 29 (P5/32-large) 54 | 55 | [ -1, 1, Conv, [ 768, 3, 2 ] ], 56 | [ [ -1, 12 ], 1, Concat, [ 1 ] ], # cat head P6 57 | [ -1, 3, C3, [ 1024, False ] ], # 32 (P6/64-xlarge) 58 | 59 | [ [ 23, 26, 29, 32 ], 1, Detect, [ nc, anchors ] ], # Detect(P3, P4, P5, P6) 60 | ] 61 | -------------------------------------------------------------------------------- /models/hub/yolov5m6.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.67 # model depth multiple 4 | width_multiple: 0.75 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [ 19,27, 44,40, 38,94 ] # P3/8 9 | - [ 96,68, 86,152, 180,137 ] # P4/16 10 | - [ 140,301, 303,264, 238,542 ] # P5/32 11 | - [ 436,615, 739,380, 925,792 ] # P6/64 12 | 13 | # YOLOv5 backbone 14 | backbone: 15 | # [from, number, module, args] 16 | [ [ -1, 1, Focus, [ 64, 3 ] ], # 0-P1/2 17 | [ -1, 1, Conv, [ 128, 3, 2 ] ], # 1-P2/4 18 | [ -1, 3, C3, [ 128 ] ], 19 | [ -1, 1, Conv, [ 256, 3, 2 ] ], # 3-P3/8 20 | [ -1, 9, C3, [ 256 ] ], 21 | [ -1, 1, Conv, [ 512, 3, 2 ] ], # 5-P4/16 22 | [ -1, 9, C3, [ 512 ] ], 23 | [ -1, 1, Conv, [ 768, 3, 2 ] ], # 7-P5/32 24 | [ -1, 3, C3, [ 768 ] ], 25 | [ -1, 1, Conv, [ 1024, 3, 2 ] ], # 9-P6/64 26 | [ -1, 1, SPP, [ 1024, [ 3, 5, 7 ] ] ], 27 | [ -1, 3, C3, [ 1024, False ] ], # 11 28 | ] 29 | 30 | # YOLOv5 head 31 | head: 32 | [ [ -1, 1, Conv, [ 768, 1, 1 ] ], 33 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 34 | [ [ -1, 8 ], 1, Concat, [ 1 ] ], # cat backbone P5 35 | [ -1, 3, C3, [ 768, False ] ], # 15 36 | 37 | [ -1, 1, Conv, [ 512, 1, 1 ] ], 38 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 39 | [ [ -1, 6 ], 1, Concat, [ 1 ] ], # cat backbone P4 40 | [ -1, 3, C3, [ 512, False ] ], # 19 41 | 42 | [ -1, 1, Conv, [ 256, 1, 1 ] ], 43 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 44 | [ [ -1, 4 ], 1, Concat, [ 1 ] ], # cat backbone P3 45 | [ -1, 3, C3, [ 256, False ] ], # 23 (P3/8-small) 46 | 47 | [ -1, 1, Conv, [ 256, 3, 2 ] ], 48 | [ [ -1, 20 ], 1, Concat, [ 1 ] ], # cat head P4 49 | [ -1, 3, C3, [ 512, False ] ], # 26 (P4/16-medium) 50 | 51 | [ -1, 1, Conv, [ 512, 3, 2 ] ], 52 | [ [ -1, 16 ], 1, Concat, [ 1 ] ], # cat head P5 53 | [ -1, 3, C3, [ 768, False ] ], # 29 (P5/32-large) 54 | 55 | [ -1, 1, Conv, [ 768, 3, 2 ] ], 56 | [ [ -1, 12 ], 1, Concat, [ 1 ] ], # cat head P6 57 | [ -1, 3, C3, [ 1024, False ] ], # 32 (P6/64-xlarge) 58 | 59 | [ [ 23, 26, 29, 32 ], 1, Detect, [ nc, anchors ] ], # Detect(P3, P4, P5, P6) 60 | ] 61 | -------------------------------------------------------------------------------- /models/hub/yolov5s-transformer.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.33 # model depth multiple 4 | width_multiple: 0.50 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, C3TR, [1024, False]], # 9 <-------- C3TR() Transformer module 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /models/hub/yolov5s6.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.33 # model depth multiple 4 | width_multiple: 0.50 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [ 19,27, 44,40, 38,94 ] # P3/8 9 | - [ 96,68, 86,152, 180,137 ] # P4/16 10 | - [ 140,301, 303,264, 238,542 ] # P5/32 11 | - [ 436,615, 739,380, 925,792 ] # P6/64 12 | 13 | # YOLOv5 backbone 14 | backbone: 15 | # [from, number, module, args] 16 | [ [ -1, 1, Focus, [ 64, 3 ] ], # 0-P1/2 17 | [ -1, 1, Conv, [ 128, 3, 2 ] ], # 1-P2/4 18 | [ -1, 3, C3, [ 128 ] ], 19 | [ -1, 1, Conv, [ 256, 3, 2 ] ], # 3-P3/8 20 | [ -1, 9, C3, [ 256 ] ], 21 | [ -1, 1, Conv, [ 512, 3, 2 ] ], # 5-P4/16 22 | [ -1, 9, C3, [ 512 ] ], 23 | [ -1, 1, Conv, [ 768, 3, 2 ] ], # 7-P5/32 24 | [ -1, 3, C3, [ 768 ] ], 25 | [ -1, 1, Conv, [ 1024, 3, 2 ] ], # 9-P6/64 26 | [ -1, 1, SPP, [ 1024, [ 3, 5, 7 ] ] ], 27 | [ -1, 3, C3, [ 1024, False ] ], # 11 28 | ] 29 | 30 | # YOLOv5 head 31 | head: 32 | [ [ -1, 1, Conv, [ 768, 1, 1 ] ], 33 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 34 | [ [ -1, 8 ], 1, Concat, [ 1 ] ], # cat backbone P5 35 | [ -1, 3, C3, [ 768, False ] ], # 15 36 | 37 | [ -1, 1, Conv, [ 512, 1, 1 ] ], 38 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 39 | [ [ -1, 6 ], 1, Concat, [ 1 ] ], # cat backbone P4 40 | [ -1, 3, C3, [ 512, False ] ], # 19 41 | 42 | [ -1, 1, Conv, [ 256, 1, 1 ] ], 43 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 44 | [ [ -1, 4 ], 1, Concat, [ 1 ] ], # cat backbone P3 45 | [ -1, 3, C3, [ 256, False ] ], # 23 (P3/8-small) 46 | 47 | [ -1, 1, Conv, [ 256, 3, 2 ] ], 48 | [ [ -1, 20 ], 1, Concat, [ 1 ] ], # cat head P4 49 | [ -1, 3, C3, [ 512, False ] ], # 26 (P4/16-medium) 50 | 51 | [ -1, 1, Conv, [ 512, 3, 2 ] ], 52 | [ [ -1, 16 ], 1, Concat, [ 1 ] ], # cat head P5 53 | [ -1, 3, C3, [ 768, False ] ], # 29 (P5/32-large) 54 | 55 | [ -1, 1, Conv, [ 768, 3, 2 ] ], 56 | [ [ -1, 12 ], 1, Concat, [ 1 ] ], # cat head P6 57 | [ -1, 3, C3, [ 1024, False ] ], # 32 (P6/64-xlarge) 58 | 59 | [ [ 23, 26, 29, 32 ], 1, Detect, [ nc, anchors ] ], # Detect(P3, P4, P5, P6) 60 | ] 61 | -------------------------------------------------------------------------------- /models/hub/yolov5x6.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.33 # model depth multiple 4 | width_multiple: 1.25 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [ 19,27, 44,40, 38,94 ] # P3/8 9 | - [ 96,68, 86,152, 180,137 ] # P4/16 10 | - [ 140,301, 303,264, 238,542 ] # P5/32 11 | - [ 436,615, 739,380, 925,792 ] # P6/64 12 | 13 | # YOLOv5 backbone 14 | backbone: 15 | # [from, number, module, args] 16 | [ [ -1, 1, Focus, [ 64, 3 ] ], # 0-P1/2 17 | [ -1, 1, Conv, [ 128, 3, 2 ] ], # 1-P2/4 18 | [ -1, 3, C3, [ 128 ] ], 19 | [ -1, 1, Conv, [ 256, 3, 2 ] ], # 3-P3/8 20 | [ -1, 9, C3, [ 256 ] ], 21 | [ -1, 1, Conv, [ 512, 3, 2 ] ], # 5-P4/16 22 | [ -1, 9, C3, [ 512 ] ], 23 | [ -1, 1, Conv, [ 768, 3, 2 ] ], # 7-P5/32 24 | [ -1, 3, C3, [ 768 ] ], 25 | [ -1, 1, Conv, [ 1024, 3, 2 ] ], # 9-P6/64 26 | [ -1, 1, SPP, [ 1024, [ 3, 5, 7 ] ] ], 27 | [ -1, 3, C3, [ 1024, False ] ], # 11 28 | ] 29 | 30 | # YOLOv5 head 31 | head: 32 | [ [ -1, 1, Conv, [ 768, 1, 1 ] ], 33 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 34 | [ [ -1, 8 ], 1, Concat, [ 1 ] ], # cat backbone P5 35 | [ -1, 3, C3, [ 768, False ] ], # 15 36 | 37 | [ -1, 1, Conv, [ 512, 1, 1 ] ], 38 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 39 | [ [ -1, 6 ], 1, Concat, [ 1 ] ], # cat backbone P4 40 | [ -1, 3, C3, [ 512, False ] ], # 19 41 | 42 | [ -1, 1, Conv, [ 256, 1, 1 ] ], 43 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 44 | [ [ -1, 4 ], 1, Concat, [ 1 ] ], # cat backbone P3 45 | [ -1, 3, C3, [ 256, False ] ], # 23 (P3/8-small) 46 | 47 | [ -1, 1, Conv, [ 256, 3, 2 ] ], 48 | [ [ -1, 20 ], 1, Concat, [ 1 ] ], # cat head P4 49 | [ -1, 3, C3, [ 512, False ] ], # 26 (P4/16-medium) 50 | 51 | [ -1, 1, Conv, [ 512, 3, 2 ] ], 52 | [ [ -1, 16 ], 1, Concat, [ 1 ] ], # cat head P5 53 | [ -1, 3, C3, [ 768, False ] ], # 29 (P5/32-large) 54 | 55 | [ -1, 1, Conv, [ 768, 3, 2 ] ], 56 | [ [ -1, 12 ], 1, Concat, [ 1 ] ], # cat head P6 57 | [ -1, 3, C3, [ 1024, False ] ], # 32 (P6/64-xlarge) 58 | 59 | [ [ 23, 26, 29, 32 ], 1, Detect, [ nc, anchors ] ], # Detect(P3, P4, P5, P6) 60 | ] 61 | -------------------------------------------------------------------------------- /models/transformer/yolov5l_fusion_add_FLIR_aligned.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 3 # number of classes 3 | depth_multiple: 1.00 # model depth multiple 4 | width_multiple: 1.00 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | # Two Stream 16 | [ 17 | [-1, 1, Focus, [64, 3]], # 0-P1/2 18 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 19 | [-1, 3, C3, [128]], 20 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 21 | [-1, 9, C3, [256]], 22 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 23 | [-1, 9, C3, [512]], 24 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 25 | [-1, 1, SPP, [1024, [5, 9, 13]]], 26 | [-1, 3, C3, [1024, False]], # 9 27 | 28 | [-4, 1, Focus, [64, 3]], # 10-P1/2 29 | [-1, 1, Conv, [128, 3, 2]], # 11-P2/4 30 | [-1, 3, C3, [128]], 31 | [-1, 1, Conv, [256, 3, 2]], # 13-P3/8 32 | [-1, 9, C3, [256]], 33 | [-1, 1, Conv, [512, 3, 2]], # 15-P4/16 34 | [-1, 9, C3, [512]], 35 | [-1, 1, Conv, [1024, 3, 2]], # 17-P5/32 36 | [-1, 1, SPP, [1024, [5, 9, 13]]], 37 | [-1, 3, C3, [1024, False]], # 19 38 | 39 | ######### Add Block ############# 40 | [[4,14], 1, Add, [1]], # 20 two stream fuse 41 | [[6,16], 1, Add, [1]], # 21 two stream fuse 42 | [[9,19], 1, Add, [1]], # 22 two stream fuse 43 | ] 44 | 45 | 46 | # YOLOv5 head 47 | head: 48 | [[-1, 1, Conv, [512, 1, 1]], # 23 49 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 24 50 | [[-1,21], 1, Concat, [1]], # 25 cat backbone P4 51 | [-1, 3, C3, [512, False]], # 26 52 | 53 | [-1, 1, Conv, [256, 1, 1]], # 27 54 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 28 55 | [[-1,20], 1, Concat, [1]], # 29 cat backbone P3 56 | [-1, 3, C3, [256, False]], # 30 (P3/8-small) 57 | 58 | [-1, 1, Conv, [256, 3, 2]], # 31 59 | [[-1,27], 1, Concat, [1]], # 32 cat head P4 60 | [-1, 3, C3, [512, False]], # 33 (P4/16-medium) 61 | 62 | [-1, 1, Conv, [512, 3, 2]], # 34 63 | [[-1,23], 1, Concat, [1]], # 35 cat head P5 64 | [-1, 3, C3, [1024, False]], # 36 (P5/32-large) 65 | 66 | [[30, 33, 36], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 67 | ] 68 | -------------------------------------------------------------------------------- /models/transformer/yolov5l_fusion_add_llvip.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 1 # number of classes 3 | depth_multiple: 1.00 # model depth multiple 4 | width_multiple: 1.00 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | # Two Stream 16 | [ 17 | [-1, 1, Focus, [64, 3]], # 0-P1/2 18 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 19 | [-1, 3, C3, [128]], 20 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 21 | [-1, 9, C3, [256]], 22 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 23 | [-1, 9, C3, [512]], 24 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 25 | [-1, 1, SPP, [1024, [5, 9, 13]]], 26 | [-1, 3, C3, [1024, False]], # 9 27 | 28 | [-4, 1, Focus, [64, 3]], # 10-P1/2 29 | [-1, 1, Conv, [128, 3, 2]], # 11-P2/4 30 | [-1, 3, C3, [128]], 31 | [-1, 1, Conv, [256, 3, 2]], # 13-P3/8 32 | [-1, 9, C3, [256]], 33 | [-1, 1, Conv, [512, 3, 2]], # 15-P4/16 34 | [-1, 9, C3, [512]], 35 | [-1, 1, Conv, [1024, 3, 2]], # 17-P5/32 36 | [-1, 1, SPP, [1024, [5, 9, 13]]], 37 | [-1, 3, C3, [1024, False]], # 19 38 | 39 | ######### Add Block ############# 40 | [[4,14], 1, Add, [1]], # 20 two stream fuse 41 | [[6,16], 1, Add, [1]], # 21 two stream fuse 42 | [[9,19], 1, Add, [1]], # 22 two stream fuse 43 | ] 44 | 45 | 46 | # YOLOv5 head 47 | head: 48 | [[-1, 1, Conv, [512, 1, 1]], # 23 49 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 24 50 | [[-1,21], 1, Concat, [1]], # 25 cat backbone P4 51 | [-1, 3, C3, [512, False]], # 26 52 | 53 | [-1, 1, Conv, [256, 1, 1]], # 27 54 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 28 55 | [[-1,20], 1, Concat, [1]], # 29 cat backbone P3 56 | [-1, 3, C3, [256, False]], # 30 (P3/8-small) 57 | 58 | [-1, 1, Conv, [256, 3, 2]], # 31 59 | [[-1,27], 1, Concat, [1]], # 32 cat head P4 60 | [-1, 3, C3, [512, False]], # 33 (P4/16-medium) 61 | 62 | [-1, 1, Conv, [512, 3, 2]], # 34 63 | [[-1,23], 1, Concat, [1]], # 35 cat head P5 64 | [-1, 3, C3, [1024, False]], # 36 (P5/32-large) 65 | 66 | [[30, 33, 36], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 67 | ] 68 | -------------------------------------------------------------------------------- /models/transformer/yolov5l_fusion_transformer_FLIR.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 3 # number of classes 3 | depth_multiple: 1.00 # model depth multiple 4 | width_multiple: 1.00 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | # Two Stream 16 | [ 17 | ######### TransformerBlock One ############# 18 | # stream one 19 | [-1, 1, Focus, [64, 3]], # 0-P1/2 20 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 21 | [-1, 3, C3, [128]], # 2-P2/4 22 | # stream two 23 | [-4, 1, Focus, [64, 3]], # 3-P1/2 24 | [-1, 1, Conv, [128, 3, 2]], # 4-P2/4 25 | [-1, 3, C3, [128]], # 5-P2/4 26 | # transformer fusion 27 | [[2,5], 1, GPT, [128]], # 6-P2/4 28 | [[2,6], 1, Add2, [128,0]], # 7-P2/4 stream one:x+trans[0] 29 | [[5,6], 1, Add2, [128,1]], # 8-P2/4 stream two:x+trans[1] 30 | 31 | ######### TransformerBlock Two ############# 32 | # stream one 33 | [7, 1, Conv, [256, 3, 2]], # 9-P3/8 34 | [-1, 9, C3, [256]], # 10-P3/8 35 | # stream two 36 | [8, 1, Conv, [256, 3, 2]], # 11-P3/8 37 | [-1, 9, C3, [256]], # 12-P3/8 38 | # transformer fusion 39 | [[10,12], 1, GPT, [256]], # 13-P3/8 40 | [[10,13], 1, Add2, [256,0]], # 14-P3/8 stream one x+trans[0] 41 | [[12,13], 1, Add2, [256,1]], # 15-P3/8 stream two x+trans[1] 42 | 43 | 44 | ######### TransformerBlock Three ############# 45 | # stream one 46 | [14, 1, Conv, [512, 3, 2]], # 16-P4/16 47 | [-1, 9, C3, [512]], # 17-P4/16 48 | # stream two 49 | [15, 1, Conv, [512, 3, 2]], # 18-P4/16 50 | [-1, 9, C3, [512]], # 19-P4/16 51 | # transformer fusion 52 | [[17,19], 1, GPT, [512]], # 20-P3/8 53 | [[17,20], 1, Add2, [512,0]], # 21-P3/8 stream one x+trans[0] 54 | [[19,20], 1, Add2, [512,1]], # 22-P3/8 stream two x+trans[1] 55 | 56 | 57 | ######### TransformerBlock Four ############# 58 | # stream one 59 | [-2, 1, Conv, [1024, 3, 2]], # 23-P5/32 60 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 24-P5/32 61 | [-1, 3, C3, [1024, False]], # 25-P5/32 62 | # stream two 63 | [22, 1, Conv, [1024, 3, 2]], # 26-P5/32 64 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 27-P5/32 65 | [-1, 3, C3, [1024, False]], # 28-P5/32 66 | # transformer fusion 67 | [[25,28], 1, GPT, [1024]], # 29-P5/32 68 | [[25,29], 1, Add2, [1024,0]], # 30-P5/32 stream one x+trans[0] 69 | [[28,29], 1, Add2, [1024,1]], # 31-P5/32 stream two x+trans[1] 70 | 71 | 72 | ######### Add Block ############# 73 | [[14,15], 1, Add, [1]], # 32-P3/8 fusion backbone P3 74 | [[21,22], 1, Add, [1]], # 33-P4/16 fusion backbone P4 75 | [[30,31], 1, Add, [1]], # 34-P5/32 fusion backbone P5 76 | 77 | ] 78 | 79 | 80 | # YOLOv5 head 81 | head: 82 | [ 83 | [-1, 1, Conv, [512, 1, 1]], # 35 84 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 36 85 | [[-1,33], 1, Concat, [1]], # 37 cat backbone P4 86 | [-1, 3, C3, [512, False]], # 38 87 | 88 | [-1, 1, Conv, [256, 1, 1]], # 39 89 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 40 90 | [[-1,32], 1, Concat, [1]], # 41 cat backbone P3 91 | [-1, 3, C3, [256, False]], # 42 (P3/8-small) 92 | 93 | [-1, 1, Conv, [256, 3, 2]], # 43 94 | [[-1,39], 1, Concat, [1]], # 44 cat head P4 95 | [-1, 3, C3, [512, False]], # 45 (P4/16-medium) 96 | 97 | [-1, 1, Conv, [512, 3, 2]], # 46 98 | [[-1,35], 1, Concat, [1]], # 47 cat head P5 99 | [-1, 3, C3, [1024, False]], # 48 (P5/32-large) 100 | 101 | [[42, 45, 48], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 102 | ] 103 | -------------------------------------------------------------------------------- /models/transformer/yolov5l_fusion_transformer_FLIR_aligned.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 3 # number of classes 3 | depth_multiple: 1.00 # model depth multiple 4 | width_multiple: 1.00 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | # Two Stream 16 | [ 17 | ######### TransformerBlock One ############# 18 | # stream one 19 | [-1, 1, Focus, [64, 3]], # 0-P1/2 20 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 21 | [-1, 3, C3, [128]], # 2-P2/4 22 | # stream two 23 | [-4, 1, Focus, [64, 3]], # 3-P1/2 24 | [-1, 1, Conv, [128, 3, 2]], # 4-P2/4 25 | [-1, 3, C3, [128]], # 5-P2/4 26 | # transformer fusion 27 | [[2,5], 1, GPT, [128]], # 6-P2/4 28 | [[2,6], 1, Add2, [128,0]], # 7-P2/4 stream one:x+trans[0] 29 | [[5,6], 1, Add2, [128,1]], # 8-P2/4 stream two:x+trans[1] 30 | 31 | ######### TransformerBlock Two ############# 32 | # stream one 33 | [7, 1, Conv, [256, 3, 2]], # 9-P3/8 34 | [-1, 9, C3, [256]], # 10-P3/8 35 | # stream two 36 | [8, 1, Conv, [256, 3, 2]], # 11-P3/8 37 | [-1, 9, C3, [256]], # 12-P3/8 38 | # transformer fusion 39 | [[10,12], 1, GPT, [256]], # 13-P3/8 40 | [[10,13], 1, Add2, [256,0]], # 14-P3/8 stream one x+trans[0] 41 | [[12,13], 1, Add2, [256,1]], # 15-P3/8 stream two x+trans[1] 42 | 43 | 44 | ######### TransformerBlock Three ############# 45 | # stream one 46 | [14, 1, Conv, [512, 3, 2]], # 16-P4/16 47 | [-1, 9, C3, [512]], # 17-P4/16 48 | # stream two 49 | [15, 1, Conv, [512, 3, 2]], # 18-P4/16 50 | [-1, 9, C3, [512]], # 19-P4/16 51 | # transformer fusion 52 | [[17,19], 1, GPT, [512]], # 20-P3/8 53 | [[17,20], 1, Add2, [512,0]], # 21-P3/8 stream one x+trans[0] 54 | [[19,20], 1, Add2, [512,1]], # 22-P3/8 stream two x+trans[1] 55 | 56 | 57 | ######### TransformerBlock Four ############# 58 | # stream one 59 | [-2, 1, Conv, [1024, 3, 2]], # 23-P5/32 60 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 24-P5/32 61 | [-1, 3, C3, [1024, False]], # 25-P5/32 62 | # stream two 63 | [22, 1, Conv, [1024, 3, 2]], # 26-P5/32 64 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 27-P5/32 65 | [-1, 3, C3, [1024, False]], # 28-P5/32 66 | # transformer fusion 67 | [[25,28], 1, GPT, [1024]], # 29-P5/32 68 | [[25,29], 1, Add2, [1024,0]], # 30-P5/32 stream one x+trans[0] 69 | [[28,29], 1, Add2, [1024,1]], # 31-P5/32 stream two x+trans[1] 70 | 71 | 72 | ######### Add Block ############# 73 | [[14,15], 1, Add, [1]], # 32-P3/8 fusion backbone P3 74 | [[21,22], 1, Add, [1]], # 33-P4/16 fusion backbone P4 75 | [[30,31], 1, Add, [1]], # 34-P5/32 fusion backbone P5 76 | 77 | ] 78 | 79 | 80 | # YOLOv5 head 81 | head: 82 | [ 83 | [-1, 1, Conv, [512, 1, 1]], # 35 84 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 36 85 | [[-1,33], 1, Concat, [1]], # 37 cat backbone P4 86 | [-1, 3, C3, [512, False]], # 38 87 | 88 | [-1, 1, Conv, [256, 1, 1]], # 39 89 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 40 90 | [[-1,32], 1, Concat, [1]], # 41 cat backbone P3 91 | [-1, 3, C3, [256, False]], # 42 (P3/8-small) 92 | 93 | [-1, 1, Conv, [256, 3, 2]], # 43 94 | [[-1,39], 1, Concat, [1]], # 44 cat head P4 95 | [-1, 3, C3, [512, False]], # 45 (P4/16-medium) 96 | 97 | [-1, 1, Conv, [512, 3, 2]], # 46 98 | [[-1,35], 1, Concat, [1]], # 47 cat head P5 99 | [-1, 3, C3, [1024, False]], # 48 (P5/32-large) 100 | 101 | [[42, 45, 48], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 102 | ] 103 | -------------------------------------------------------------------------------- /models/transformer/yolov5l_fusion_transformer_llvip.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 1 # number of classes 3 | depth_multiple: 1.00 # model depth multiple 4 | width_multiple: 1.00 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | # Two Stream 16 | [ 17 | ######### TransformerBlock One ############# 18 | # stream one 19 | [-1, 1, Focus, [64, 3]], # 0-P1/2 20 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 21 | [-1, 3, C3, [128]], # 2-P2/4 22 | # stream two 23 | [-4, 1, Focus, [64, 3]], # 3-P1/2 24 | [-1, 1, Conv, [128, 3, 2]], # 4-P2/4 25 | [-1, 3, C3, [128]], # 5-P2/4 26 | # transformer fusion 27 | [[2,5], 1, GPT, [128]], # 6-P2/4 28 | [[2,6], 1, Add2, [128,0]], # 7-P2/4 stream one:x+trans[0] 29 | [[5,6], 1, Add2, [128,1]], # 8-P2/4 stream two:x+trans[1] 30 | 31 | ######### TransformerBlock Two ############# 32 | # stream one 33 | [7, 1, Conv, [256, 3, 2]], # 9-P3/8 34 | [-1, 9, C3, [256]], # 10-P3/8 35 | # stream two 36 | [8, 1, Conv, [256, 3, 2]], # 11-P3/8 37 | [-1, 9, C3, [256]], # 12-P3/8 38 | # transformer fusion 39 | [[10,12], 1, GPT, [256]], # 13-P3/8 40 | [[10,13], 1, Add2, [256,0]], # 14-P3/8 stream one x+trans[0] 41 | [[12,13], 1, Add2, [256,1]], # 15-P3/8 stream two x+trans[1] 42 | 43 | 44 | ######### TransformerBlock Three ############# 45 | # stream one 46 | [14, 1, Conv, [512, 3, 2]], # 16-P4/16 47 | [-1, 9, C3, [512]], # 17-P4/16 48 | # stream two 49 | [15, 1, Conv, [512, 3, 2]], # 18-P4/16 50 | [-1, 9, C3, [512]], # 19-P4/16 51 | # transformer fusion 52 | [[17,19], 1, GPT, [512]], # 20-P3/8 53 | [[17,20], 1, Add2, [512,0]], # 21-P3/8 stream one x+trans[0] 54 | [[19,20], 1, Add2, [512,1]], # 22-P3/8 stream two x+trans[1] 55 | 56 | 57 | ######### TransformerBlock Four ############# 58 | # stream one 59 | [-2, 1, Conv, [1024, 3, 2]], # 23-P5/32 60 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 24-P5/32 61 | [-1, 3, C3, [1024, False]], # 25-P5/32 62 | # stream two 63 | [22, 1, Conv, [1024, 3, 2]], # 26-P5/32 64 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 27-P5/32 65 | [-1, 3, C3, [1024, False]], # 28-P5/32 66 | # transformer fusion 67 | [[25,28], 1, GPT, [1024]], # 29-P5/32 68 | [[25,29], 1, Add2, [1024,0]], # 30-P5/32 stream one x+trans[0] 69 | [[28,29], 1, Add2, [1024,1]], # 31-P5/32 stream two x+trans[1] 70 | 71 | 72 | ######### Add Block ############# 73 | [[14,15], 1, Add, [1]], # 32-P3/8 fusion backbone P3 74 | [[21,22], 1, Add, [1]], # 33-P4/16 fusion backbone P4 75 | [[30,31], 1, Add, [1]], # 34-P5/32 fusion backbone P5 76 | 77 | ] 78 | 79 | 80 | # YOLOv5 head 81 | head: 82 | [ 83 | [-1, 1, Conv, [512, 1, 1]], # 35 84 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 36 85 | [[-1,33], 1, Concat, [1]], # 37 cat backbone P4 86 | [-1, 3, C3, [512, False]], # 38 87 | 88 | [-1, 1, Conv, [256, 1, 1]], # 39 89 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 40 90 | [[-1,32], 1, Concat, [1]], # 41 cat backbone P3 91 | [-1, 3, C3, [256, False]], # 42 (P3/8-small) 92 | 93 | [-1, 1, Conv, [256, 3, 2]], # 43 94 | [[-1,39], 1, Concat, [1]], # 44 cat head P4 95 | [-1, 3, C3, [512, False]], # 45 (P4/16-medium) 96 | 97 | [-1, 1, Conv, [512, 3, 2]], # 46 98 | [[-1,35], 1, Concat, [1]], # 47 cat head P5 99 | [-1, 3, C3, [1024, False]], # 48 (P5/32-large) 100 | 101 | [[42, 45, 48], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 102 | ] 103 | -------------------------------------------------------------------------------- /models/transformer/yolov5l_fusion_transformer_vedai.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 9 # number of classes 3 | depth_multiple: 1.00 # model depth multiple 4 | width_multiple: 1.00 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | # Two Stream 16 | [ 17 | ######### TransformerBlock One ############# 18 | # stream one 19 | [-1, 1, Focus, [64, 3]], # 0-P1/2 20 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 21 | [-1, 3, C3, [128]], # 2-P2/4 22 | # stream two 23 | [-4, 1, Focus, [64, 3]], # 3-P1/2 24 | [-1, 1, Conv, [128, 3, 2]], # 4-P2/4 25 | [-1, 3, C3, [128]], # 5-P2/4 26 | # transformer fusion 27 | [[2,5], 1, GPT, [128]], # 6-P2/4 28 | [[2,6], 1, Add2, [128,0]], # 7-P2/4 stream one:x+trans[0] 29 | [[5,6], 1, Add2, [128,1]], # 8-P2/4 stream two:x+trans[1] 30 | 31 | ######### TransformerBlock Two ############# 32 | # stream one 33 | [7, 1, Conv, [256, 3, 2]], # 9-P3/8 34 | [-1, 9, C3, [256]], # 10-P3/8 35 | # stream two 36 | [8, 1, Conv, [256, 3, 2]], # 11-P3/8 37 | [-1, 9, C3, [256]], # 12-P3/8 38 | # transformer fusion 39 | [[10,12], 1, GPT, [256]], # 13-P3/8 40 | [[10,13], 1, Add2, [256,0]], # 14-P3/8 stream one x+trans[0] 41 | [[12,13], 1, Add2, [256,1]], # 15-P3/8 stream two x+trans[1] 42 | 43 | 44 | ######### TransformerBlock Three ############# 45 | # stream one 46 | [14, 1, Conv, [512, 3, 2]], # 16-P4/16 47 | [-1, 9, C3, [512]], # 17-P4/16 48 | # stream two 49 | [15, 1, Conv, [512, 3, 2]], # 18-P4/16 50 | [-1, 9, C3, [512]], # 19-P4/16 51 | # transformer fusion 52 | [[17,19], 1, GPT, [512]], # 20-P3/8 53 | [[17,20], 1, Add2, [512,0]], # 21-P3/8 stream one x+trans[0] 54 | [[19,20], 1, Add2, [512,1]], # 22-P3/8 stream two x+trans[1] 55 | 56 | 57 | ######### TransformerBlock Four ############# 58 | # stream one 59 | [-2, 1, Conv, [1024, 3, 2]], # 23-P5/32 60 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 24-P5/32 61 | [-1, 3, C3, [1024, False]], # 25-P5/32 62 | # stream two 63 | [22, 1, Conv, [1024, 3, 2]], # 26-P5/32 64 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 27-P5/32 65 | [-1, 3, C3, [1024, False]], # 28-P5/32 66 | # transformer fusion 67 | [[25,28], 1, GPT, [1024]], # 29-P5/32 68 | [[25,29], 1, Add2, [1024,0]], # 30-P5/32 stream one x+trans[0] 69 | [[28,29], 1, Add2, [1024,1]], # 31-P5/32 stream two x+trans[1] 70 | 71 | 72 | ######### Add Block ############# 73 | [[14,15], 1, Add, [1]], # 32-P3/8 fusion backbone P3 74 | [[21,22], 1, Add, [1]], # 33-P4/16 fusion backbone P4 75 | [[30,31], 1, Add, [1]], # 34-P5/32 fusion backbone P5 76 | 77 | ] 78 | 79 | 80 | # YOLOv5 head 81 | head: 82 | [ 83 | [-1, 1, Conv, [512, 1, 1]], # 35 84 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 36 85 | [[-1,33], 1, Concat, [1]], # 37 cat backbone P4 86 | [-1, 3, C3, [512, False]], # 38 87 | 88 | [-1, 1, Conv, [256, 1, 1]], # 39 89 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 40 90 | [[-1,32], 1, Concat, [1]], # 41 cat backbone P3 91 | [-1, 3, C3, [256, False]], # 42 (P3/8-small) 92 | 93 | [-1, 1, Conv, [256, 3, 2]], # 43 94 | [[-1,39], 1, Concat, [1]], # 44 cat head P4 95 | [-1, 3, C3, [512, False]], # 45 (P4/16-medium) 96 | 97 | [-1, 1, Conv, [512, 3, 2]], # 46 98 | [[-1,35], 1, Concat, [1]], # 47 cat head P5 99 | [-1, 3, C3, [1024, False]], # 48 (P5/32-large) 100 | 101 | [[42, 45, 48], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 102 | ] 103 | -------------------------------------------------------------------------------- /models/transformer/yolov5l_fusion_transformerx3_FLIR_aligned.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 3 # number of classes 3 | depth_multiple: 1.00 # model depth multiple 4 | width_multiple: 1.00 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | # Two Stream 16 | [ 17 | ######### TransformerBlock One ############# 18 | # stream one 19 | [-1, 1, Focus, [64, 3]], # 0-P1/2 20 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 21 | [-1, 3, C3, [128]], # 2-P2/4 22 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 23 | [-1, 9, C3, [256]], # 4-P3/8 24 | # stream two 25 | [-4, 1, Focus, [64, 3]], # 5-P1/2 26 | [-1, 1, Conv, [128, 3, 2]], # 6-P2/4 27 | [-1, 3, C3, [128]], # 7-P2/4 28 | [-1, 1, Conv, [256, 3, 2]], # 8-P3/8 29 | [-1, 9, C3, [256]], # 9-P3/8 30 | 31 | 32 | ######### TransformerBlock Two ############# 33 | # transformer fusion 34 | [[4,9], 1, GPT, [256]], # 10-P3/8 35 | [[4,10], 1, Add2, [256,0]], # 11-P3/8 stream one x+trans[0] 36 | [[9,10], 1, Add2, [256,1]], # 12-P3/8 stream two x+trans[1] 37 | 38 | 39 | ######### TransformerBlock Three ############# 40 | # stream one 41 | [11, 1, Conv, [512, 3, 2]], # 13-P4/16 42 | [-1, 9, C3, [512]], # 14-P4/16 43 | # stream two 44 | [12, 1, Conv, [512, 3, 2]], # 15-P4/16 45 | [-1, 9, C3, [512]], # 16-P4/16 46 | # transformer fusion 47 | [[14,16], 1, GPT, [512]], # 17-P3/8 48 | [[14,17], 1, Add2, [512,0]], # 18-P3/8 stream one x+trans[0] 49 | [[16,17], 1, Add2, [512,1]], # 19-P3/8 stream two x+trans[1] 50 | 51 | 52 | ######### TransformerBlock Four ############# 53 | # stream one 54 | [18, 1, Conv, [1024, 3, 2]], # 20-P5/32 55 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 21-P5/32 56 | [-1, 3, C3, [1024, False]], # 22-P5/32 57 | # stream two 58 | [19, 1, Conv, [1024, 3, 2]], # 23-P5/32 59 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 24-P5/32 60 | [-1, 3, C3, [1024, False]], # 25-P5/32 61 | # transformer fusion 62 | [[22,25], 1, GPT, [1024]], # 26-P5/32 63 | [[22,26], 1, Add2, [1024,0]], # 27-P5/32 stream one x+trans[0] 64 | [[25,26], 1, Add2, [1024,1]], # 28-P5/32 stream two x+trans[1] 65 | 66 | 67 | ######### Add Block ############# 68 | [[11,12], 1, Add, [1]], # 29-P3/8 fusion backbone P3 69 | [[18,19], 1, Add, [1]], # 30-P4/16 fusion backbone P4 70 | [[27,28], 1, Add, [1]], # 31-P5/32 fusion backbone P5 71 | 72 | ] 73 | 74 | 75 | # YOLOv5 head 76 | head: 77 | [ 78 | [-1, 1, Conv, [512, 1, 1]], # 32 79 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 33 80 | [[-1,30], 1, Concat, [1]], # 34 cat backbone P4 81 | [-1, 3, C3, [512, False]], # 35 82 | 83 | [-1, 1, Conv, [256, 1, 1]], # 36 84 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 37 85 | [[-1,29], 1, Concat, [1]], # 38 cat backbone P3 86 | [-1, 3, C3, [256, False]], # 39 (P3/8-small) 87 | 88 | [-1, 1, Conv, [256, 3, 2]], # 40 89 | [[-1,36], 1, Concat, [1]], # 41 cat head P4 90 | [-1, 3, C3, [512, False]], # 42 (P4/16-medium) 91 | 92 | [-1, 1, Conv, [512, 3, 2]], # 43 93 | [[-1,32], 1, Concat, [1]], # 44 cat head P5 94 | [-1, 3, C3, [1024, False]], # 45 (P5/32-large) 95 | 96 | [[39, 42, 45], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 97 | ] 98 | -------------------------------------------------------------------------------- /models/transformer/yolov5l_fusion_transformerx3_llvip.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 1 # number of classes 3 | depth_multiple: 1.00 # model depth multiple 4 | width_multiple: 1.00 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | # Two Stream 16 | [ 17 | ######### TransformerBlock One ############# 18 | # stream one 19 | [-1, 1, Focus, [64, 3]], # 0-P1/2 20 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 21 | [-1, 3, C3, [128]], # 2-P2/4 22 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 23 | [-1, 9, C3, [256]], # 4-P3/8 24 | # stream two 25 | [-4, 1, Focus, [64, 3]], # 5-P1/2 26 | [-1, 1, Conv, [128, 3, 2]], # 6-P2/4 27 | [-1, 3, C3, [128]], # 7-P2/4 28 | [-1, 1, Conv, [256, 3, 2]], # 8-P3/8 29 | [-1, 9, C3, [256]], # 9-P3/8 30 | 31 | 32 | ######### TransformerBlock Two ############# 33 | # transformer fusion 34 | [[4,9], 1, GPT, [256]], # 10-P3/8 35 | [[4,10], 1, Add2, [256,0]], # 11-P3/8 stream one x+trans[0] 36 | [[9,10], 1, Add2, [256,1]], # 12-P3/8 stream two x+trans[1] 37 | 38 | 39 | ######### TransformerBlock Three ############# 40 | # stream one 41 | [11, 1, Conv, [512, 3, 2]], # 13-P4/16 42 | [-1, 9, C3, [512]], # 14-P4/16 43 | # stream two 44 | [12, 1, Conv, [512, 3, 2]], # 15-P4/16 45 | [-1, 9, C3, [512]], # 16-P4/16 46 | # transformer fusion 47 | [[14,16], 1, GPT, [512]], # 17-P3/8 48 | [[14,17], 1, Add2, [512,0]], # 18-P3/8 stream one x+trans[0] 49 | [[16,17], 1, Add2, [512,1]], # 19-P3/8 stream two x+trans[1] 50 | 51 | 52 | ######### TransformerBlock Four ############# 53 | # stream one 54 | [18, 1, Conv, [1024, 3, 2]], # 20-P5/32 55 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 21-P5/32 56 | [-1, 3, C3, [1024, False]], # 22-P5/32 57 | # stream two 58 | [19, 1, Conv, [1024, 3, 2]], # 23-P5/32 59 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 24-P5/32 60 | [-1, 3, C3, [1024, False]], # 25-P5/32 61 | # transformer fusion 62 | [[22,25], 1, GPT, [1024]], # 26-P5/32 63 | [[22,26], 1, Add2, [1024,0]], # 27-P5/32 stream one x+trans[0] 64 | [[25,26], 1, Add2, [1024,1]], # 28-P5/32 stream two x+trans[1] 65 | 66 | 67 | ######### Add Block ############# 68 | [[11,12], 1, Add, [1]], # 29-P3/8 fusion backbone P3 69 | [[18,19], 1, Add, [1]], # 30-P4/16 fusion backbone P4 70 | [[27,28], 1, Add, [1]], # 31-P5/32 fusion backbone P5 71 | 72 | ] 73 | 74 | 75 | # YOLOv5 head 76 | head: 77 | [ 78 | [-1, 1, Conv, [512, 1, 1]], # 32 79 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 33 80 | [[-1,30], 1, Concat, [1]], # 34 cat backbone P4 81 | [-1, 3, C3, [512, False]], # 35 82 | 83 | [-1, 1, Conv, [256, 1, 1]], # 36 84 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 37 85 | [[-1,29], 1, Concat, [1]], # 38 cat backbone P3 86 | [-1, 3, C3, [256, False]], # 39 (P3/8-small) 87 | 88 | [-1, 1, Conv, [256, 3, 2]], # 40 89 | [[-1,36], 1, Concat, [1]], # 41 cat head P4 90 | [-1, 3, C3, [512, False]], # 42 (P4/16-medium) 91 | 92 | [-1, 1, Conv, [512, 3, 2]], # 43 93 | [[-1,32], 1, Concat, [1]], # 44 cat head P5 94 | [-1, 3, C3, [1024, False]], # 45 (P5/32-large) 95 | 96 | [[39, 42, 45], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 97 | ] 98 | -------------------------------------------------------------------------------- /models/transformer/yolov5s_fusion_add_vedai.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 9 # number of classes 3 | depth_multiple: 0.33 # model depth multiple 4 | width_multiple: 0.50 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | # Two Stream 16 | [ 17 | [-1, 1, Focus, [64, 3]], # 0-P1/2 18 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 19 | [-1, 3, C3, [128]], 20 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 21 | [-1, 9, C3, [256]], 22 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 23 | [-1, 9, C3, [512]], 24 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 25 | [-1, 1, SPP, [1024, [5, 9, 13]]], 26 | [-1, 3, C3, [1024, False]], # 9 27 | 28 | [-4, 1, Focus, [64, 3]], # 10-P1/2 29 | [-1, 1, Conv, [128, 3, 2]], # 11-P2/4 30 | [-1, 3, C3, [128]], 31 | [-1, 1, Conv, [256, 3, 2]], # 13-P3/8 32 | [-1, 9, C3, [256]], 33 | [-1, 1, Conv, [512, 3, 2]], # 15-P4/16 34 | [-1, 9, C3, [512]], 35 | [-1, 1, Conv, [1024, 3, 2]], # 17-P5/32 36 | [-1, 1, SPP, [1024, [5, 9, 13]]], 37 | [-1, 3, C3, [1024, False]], # 19 38 | 39 | ######### Add Block ############# 40 | [[4,14], 1, Add, [1]], # 20 two stream fuse 41 | [[6,16], 1, Add, [1]], # 21 two stream fuse 42 | [[9,19], 1, Add, [1]], # 22 two stream fuse 43 | ] 44 | 45 | 46 | # YOLOv5 head 47 | head: 48 | [[-1, 1, Conv, [512, 1, 1]], # 23 49 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 24 50 | [[-1,21], 1, Concat, [1]], # 25 cat backbone P4 51 | [-1, 3, C3, [512, False]], # 26 52 | 53 | [-1, 1, Conv, [256, 1, 1]], # 27 54 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 28 55 | [[-1,20], 1, Concat, [1]], # 29 cat backbone P3 56 | [-1, 3, C3, [256, False]], # 30 (P3/8-small) 57 | 58 | [-1, 1, Conv, [256, 3, 2]], # 31 59 | [[-1,27], 1, Concat, [1]], # 32 cat head P4 60 | [-1, 3, C3, [512, False]], # 33 (P4/16-medium) 61 | 62 | [-1, 1, Conv, [512, 3, 2]], # 34 63 | [[-1,23], 1, Concat, [1]], # 35 cat head P5 64 | [-1, 3, C3, [1024, False]], # 36 (P5/32-large) 65 | 66 | [[30, 33, 36], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 67 | ] 68 | -------------------------------------------------------------------------------- /models/transformer/yolov5s_fusion_transformer_vedai.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 9 # number of classes 3 | depth_multiple: 0.33 # model depth multiple 4 | width_multiple: 0.50 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | # Two Stream 16 | [ 17 | ######### TransformerBlock One ############# 18 | # stream one 19 | [-1, 1, Focus, [64, 3]], # 0-P1/2 20 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 21 | [-1, 3, C3, [128]], # 2-P2/4 22 | # stream two 23 | [-4, 1, Focus, [64, 3]], # 3-P1/2 24 | [-1, 1, Conv, [128, 3, 2]], # 4-P2/4 25 | [-1, 3, C3, [128]], # 5-P2/4 26 | # transformer fusion 27 | [[2,5], 1, GPT, [128]], # 6-P2/4 28 | [[2,6], 1, Add2, [128,0]], # 7-P2/4 stream one:x+trans[0] 29 | [[5,6], 1, Add2, [128,1]], # 8-P2/4 stream two:x+trans[1] 30 | 31 | ######### TransformerBlock Two ############# 32 | # stream one 33 | [7, 1, Conv, [256, 3, 2]], # 9-P3/8 34 | [-1, 9, C3, [256]], # 10-P3/8 35 | # stream two 36 | [8, 1, Conv, [256, 3, 2]], # 11-P3/8 37 | [-1, 9, C3, [256]], # 12-P3/8 38 | # transformer fusion 39 | [[10,12], 1, GPT, [256]], # 13-P3/8 40 | [[10,13], 1, Add2, [256,0]], # 14-P3/8 stream one x+trans[0] 41 | [[12,13], 1, Add2, [256,1]], # 15-P3/8 stream two x+trans[1] 42 | 43 | 44 | ######### TransformerBlock Three ############# 45 | # stream one 46 | [14, 1, Conv, [512, 3, 2]], # 16-P4/16 47 | [-1, 9, C3, [512]], # 17-P4/16 48 | # stream two 49 | [15, 1, Conv, [512, 3, 2]], # 18-P4/16 50 | [-1, 9, C3, [512]], # 19-P4/16 51 | # transformer fusion 52 | [[17,19], 1, GPT, [512]], # 20-P3/8 53 | [[17,20], 1, Add2, [512,0]], # 21-P3/8 stream one x+trans[0] 54 | [[19,20], 1, Add2, [512,1]], # 22-P3/8 stream two x+trans[1] 55 | 56 | 57 | ######### TransformerBlock Four ############# 58 | # stream one 59 | [-2, 1, Conv, [1024, 3, 2]], # 23-P5/32 60 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 24-P5/32 61 | [-1, 3, C3, [1024, False]], # 25-P5/32 62 | # stream two 63 | [22, 1, Conv, [1024, 3, 2]], # 26-P5/32 64 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 27-P5/32 65 | [-1, 3, C3, [1024, False]], # 28-P5/32 66 | # transformer fusion 67 | [[25,28], 1, GPT, [1024]], # 29-P5/32 68 | [[25,29], 1, Add2, [1024,0]], # 30-P5/32 stream one x+trans[0] 69 | [[28,29], 1, Add2, [1024,1]], # 31-P5/32 stream two x+trans[1] 70 | 71 | 72 | ######### Add Block ############# 73 | [[14,15], 1, Add, [1]], # 32-P3/8 fusion backbone P3 74 | [[21,22], 1, Add, [1]], # 33-P4/16 fusion backbone P4 75 | [[30,31], 1, Add, [1]], # 34-P5/32 fusion backbone P5 76 | 77 | ] 78 | 79 | 80 | # YOLOv5 head 81 | head: 82 | [ 83 | [-1, 1, Conv, [512, 1, 1]], # 35 84 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 36 85 | [[-1,33], 1, Concat, [1]], # 37 cat backbone P4 86 | [-1, 3, C3, [512, False]], # 38 87 | 88 | [-1, 1, Conv, [256, 1, 1]], # 39 89 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 40 90 | [[-1,32], 1, Concat, [1]], # 41 cat backbone P3 91 | [-1, 3, C3, [256, False]], # 42 (P3/8-small) 92 | 93 | [-1, 1, Conv, [256, 3, 2]], # 43 94 | [[-1,39], 1, Concat, [1]], # 44 cat head P4 95 | [-1, 3, C3, [512, False]], # 45 (P4/16-medium) 96 | 97 | [-1, 1, Conv, [512, 3, 2]], # 46 98 | [[-1,35], 1, Concat, [1]], # 47 cat head P5 99 | [-1, 3, C3, [1024, False]], # 48 (P5/32-large) 100 | 101 | [[42, 45, 48], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 102 | ] 103 | -------------------------------------------------------------------------------- /models/transformer/yolov5s_fusion_transformerx3_vedai.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 9 # number of classes 3 | depth_multiple: 0.33 # model depth multiple 4 | width_multiple: 0.50 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | # Two Stream 16 | [ 17 | ######### TransformerBlock One ############# 18 | # stream one 19 | [-1, 1, Focus, [64, 3]], # 0-P1/2 20 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 21 | [-1, 3, C3, [128]], # 2-P2/4 22 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 23 | [-1, 9, C3, [256]], # 4-P3/8 24 | # stream two 25 | [-4, 1, Focus, [64, 3]], # 5-P1/2 26 | [-1, 1, Conv, [128, 3, 2]], # 6-P2/4 27 | [-1, 3, C3, [128]], # 7-P2/4 28 | [-1, 1, Conv, [256, 3, 2]], # 8-P3/8 29 | [-1, 9, C3, [256]], # 9-P3/8 30 | 31 | 32 | ######### TransformerBlock Two ############# 33 | # transformer fusion 34 | [[4,9], 1, GPT, [256]], # 10-P3/8 35 | [[4,10], 1, Add2, [256,0]], # 11-P3/8 stream one x+trans[0] 36 | [[9,10], 1, Add2, [256,1]], # 12-P3/8 stream two x+trans[1] 37 | 38 | 39 | ######### TransformerBlock Three ############# 40 | # stream one 41 | [11, 1, Conv, [512, 3, 2]], # 13-P4/16 42 | [-1, 9, C3, [512]], # 14-P4/16 43 | # stream two 44 | [12, 1, Conv, [512, 3, 2]], # 15-P4/16 45 | [-1, 9, C3, [512]], # 16-P4/16 46 | # transformer fusion 47 | [[14,16], 1, GPT, [512]], # 17-P3/8 48 | [[14,17], 1, Add2, [512,0]], # 18-P3/8 stream one x+trans[0] 49 | [[16,17], 1, Add2, [512,1]], # 19-P3/8 stream two x+trans[1] 50 | 51 | 52 | ######### TransformerBlock Four ############# 53 | # stream one 54 | [18, 1, Conv, [1024, 3, 2]], # 20-P5/32 55 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 21-P5/32 56 | [-1, 3, C3, [1024, False]], # 22-P5/32 57 | # stream two 58 | [19, 1, Conv, [1024, 3, 2]], # 23-P5/32 59 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 24-P5/32 60 | [-1, 3, C3, [1024, False]], # 25-P5/32 61 | # transformer fusion 62 | [[22,25], 1, GPT, [1024]], # 26-P5/32 63 | [[22,26], 1, Add2, [1024,0]], # 27-P5/32 stream one x+trans[0] 64 | [[25,26], 1, Add2, [1024,1]], # 28-P5/32 stream two x+trans[1] 65 | 66 | 67 | ######### Add Block ############# 68 | [[11,12], 1, Add, [1]], # 29-P3/8 fusion backbone P3 69 | [[18,19], 1, Add, [1]], # 30-P4/16 fusion backbone P4 70 | [[27,28], 1, Add, [1]], # 31-P5/32 fusion backbone P5 71 | 72 | ] 73 | 74 | 75 | # YOLOv5 head 76 | head: 77 | [ 78 | [-1, 1, Conv, [512, 1, 1]], # 32 79 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 33 80 | [[-1,30], 1, Concat, [1]], # 34 cat backbone P4 81 | [-1, 3, C3, [512, False]], # 35 82 | 83 | [-1, 1, Conv, [256, 1, 1]], # 36 84 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 37 85 | [[-1,29], 1, Concat, [1]], # 38 cat backbone P3 86 | [-1, 3, C3, [256, False]], # 39 (P3/8-small) 87 | 88 | [-1, 1, Conv, [256, 3, 2]], # 40 89 | [[-1,36], 1, Concat, [1]], # 41 cat head P4 90 | [-1, 3, C3, [512, False]], # 42 (P4/16-medium) 91 | 92 | [-1, 1, Conv, [512, 3, 2]], # 43 93 | [[-1,32], 1, Concat, [1]], # 44 cat head P5 94 | [-1, 3, C3, [1024, False]], # 45 (P5/32-large) 95 | 96 | [[39, 42, 45], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 97 | ] 98 | -------------------------------------------------------------------------------- /models/transformer/yolov5x_fusion_transformer_FLIR.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 3 # number of classes 3 | depth_multiple: 1.33 # model depth multiple 4 | width_multiple: 1.25 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | # Two Stream 16 | [ 17 | ######### TransformerBlock One ############# 18 | # stream one 19 | [-1, 1, Focus, [64, 3]], # 0-P1/2 20 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 21 | [-1, 3, C3, [128]], # 2-P2/4 22 | # stream two 23 | [-4, 1, Focus, [64, 3]], # 3-P1/2 24 | [-1, 1, Conv, [128, 3, 2]], # 4-P2/4 25 | [-1, 3, C3, [128]], # 5-P2/4 26 | # transformer fusion 27 | [[2,5], 1, GPT, [128]], # 6-P2/4 28 | [[2,6], 1, Add2, [128,0]], # 7-P2/4 stream one:x+trans[0] 29 | [[5,6], 1, Add2, [128,1]], # 8-P2/4 stream two:x+trans[1] 30 | 31 | ######### TransformerBlock Two ############# 32 | # stream one 33 | [7, 1, Conv, [256, 3, 2]], # 9-P3/8 34 | [-1, 9, C3, [256]], # 10-P3/8 35 | # stream two 36 | [8, 1, Conv, [256, 3, 2]], # 11-P3/8 37 | [-1, 9, C3, [256]], # 12-P3/8 38 | # transformer fusion 39 | [[10,12], 1, GPT, [256]], # 13-P3/8 40 | [[10,13], 1, Add2, [256,0]], # 14-P3/8 stream one x+trans[0] 41 | [[12,13], 1, Add2, [256,1]], # 15-P3/8 stream two x+trans[1] 42 | 43 | 44 | ######### TransformerBlock Three ############# 45 | # stream one 46 | [14, 1, Conv, [512, 3, 2]], # 16-P4/16 47 | [-1, 9, C3, [512]], # 17-P4/16 48 | # stream two 49 | [15, 1, Conv, [512, 3, 2]], # 18-P4/16 50 | [-1, 9, C3, [512]], # 19-P4/16 51 | # transformer fusion 52 | [[17,19], 1, GPT, [512]], # 20-P3/8 53 | [[17,20], 1, Add2, [512,0]], # 21-P3/8 stream one x+trans[0] 54 | [[19,20], 1, Add2, [512,1]], # 22-P3/8 stream two x+trans[1] 55 | 56 | 57 | ######### TransformerBlock Four ############# 58 | # stream one 59 | [-2, 1, Conv, [1024, 3, 2]], # 23-P5/32 60 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 24-P5/32 61 | [-1, 3, C3, [1024, False]], # 25-P5/32 62 | # stream two 63 | [22, 1, Conv, [1024, 3, 2]], # 26-P5/32 64 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 27-P5/32 65 | [-1, 3, C3, [1024, False]], # 28-P5/32 66 | # transformer fusion 67 | [[25,28], 1, GPT, [1024]], # 29-P5/32 68 | [[25,29], 1, Add2, [1024,0]], # 30-P5/32 stream one x+trans[0] 69 | [[28,29], 1, Add2, [1024,1]], # 31-P5/32 stream two x+trans[1] 70 | 71 | 72 | ######### Add Block ############# 73 | [[14,15], 1, Add, [1]], # 32-P3/8 fusion backbone P3 74 | [[21,22], 1, Add, [1]], # 33-P4/16 fusion backbone P4 75 | [[30,31], 1, Add, [1]], # 34-P5/32 fusion backbone P5 76 | 77 | ] 78 | 79 | 80 | # YOLOv5 head 81 | head: 82 | [ 83 | [-1, 1, Conv, [512, 1, 1]], # 35 84 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 36 85 | [[-1,33], 1, Concat, [1]], # 37 cat backbone P4 86 | [-1, 3, C3, [512, False]], # 38 87 | 88 | [-1, 1, Conv, [256, 1, 1]], # 39 89 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 40 90 | [[-1,32], 1, Concat, [1]], # 41 cat backbone P3 91 | [-1, 3, C3, [256, False]], # 42 (P3/8-small) 92 | 93 | [-1, 1, Conv, [256, 3, 2]], # 43 94 | [[-1,39], 1, Concat, [1]], # 44 cat head P4 95 | [-1, 3, C3, [512, False]], # 45 (P4/16-medium) 96 | 97 | [-1, 1, Conv, [512, 3, 2]], # 46 98 | [[-1,35], 1, Concat, [1]], # 47 cat head P5 99 | [-1, 3, C3, [1024, False]], # 48 (P5/32-large) 100 | 101 | [[42, 45, 48], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 102 | ] 103 | -------------------------------------------------------------------------------- /models/transformer/yolov5x_fusion_transformer_FLIR_aligned.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 3 # number of classes 3 | depth_multiple: 1.33 # model depth multiple 4 | width_multiple: 1.25 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | # Two Stream 16 | [ 17 | ######### TransformerBlock One ############# 18 | # stream one 19 | [-1, 1, Focus, [64, 3]], # 0-P1/2 20 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 21 | [-1, 3, C3, [128]], # 2-P2/4 22 | # stream two 23 | [-4, 1, Focus, [64, 3]], # 3-P1/2 24 | [-1, 1, Conv, [128, 3, 2]], # 4-P2/4 25 | [-1, 3, C3, [128]], # 5-P2/4 26 | # transformer fusion 27 | [[2,5], 1, GPT, [128]], # 6-P2/4 28 | [[2,6], 1, Add2, [128,0]], # 7-P2/4 stream one:x+trans[0] 29 | [[5,6], 1, Add2, [128,1]], # 8-P2/4 stream two:x+trans[1] 30 | 31 | ######### TransformerBlock Two ############# 32 | # stream one 33 | [7, 1, Conv, [256, 3, 2]], # 9-P3/8 34 | [-1, 9, C3, [256]], # 10-P3/8 35 | # stream two 36 | [8, 1, Conv, [256, 3, 2]], # 11-P3/8 37 | [-1, 9, C3, [256]], # 12-P3/8 38 | # transformer fusion 39 | [[10,12], 1, GPT, [256]], # 13-P3/8 40 | [[10,13], 1, Add2, [256,0]], # 14-P3/8 stream one x+trans[0] 41 | [[12,13], 1, Add2, [256,1]], # 15-P3/8 stream two x+trans[1] 42 | 43 | 44 | ######### TransformerBlock Three ############# 45 | # stream one 46 | [14, 1, Conv, [512, 3, 2]], # 16-P4/16 47 | [-1, 9, C3, [512]], # 17-P4/16 48 | # stream two 49 | [15, 1, Conv, [512, 3, 2]], # 18-P4/16 50 | [-1, 9, C3, [512]], # 19-P4/16 51 | # transformer fusion 52 | [[17,19], 1, GPT, [512]], # 20-P3/8 53 | [[17,20], 1, Add2, [512,0]], # 21-P3/8 stream one x+trans[0] 54 | [[19,20], 1, Add2, [512,1]], # 22-P3/8 stream two x+trans[1] 55 | 56 | 57 | ######### TransformerBlock Four ############# 58 | # stream one 59 | [-2, 1, Conv, [1024, 3, 2]], # 23-P5/32 60 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 24-P5/32 61 | [-1, 3, C3, [1024, False]], # 25-P5/32 62 | # stream two 63 | [22, 1, Conv, [1024, 3, 2]], # 26-P5/32 64 | [-1, 1, SPP, [1024, [5, 9, 13]]], # 27-P5/32 65 | [-1, 3, C3, [1024, False]], # 28-P5/32 66 | # transformer fusion 67 | [[25,28], 1, GPT, [1024]], # 29-P5/32 68 | [[25,29], 1, Add2, [1024,0]], # 30-P5/32 stream one x+trans[0] 69 | [[28,29], 1, Add2, [1024,1]], # 31-P5/32 stream two x+trans[1] 70 | 71 | 72 | ######### Add Block ############# 73 | [[14,15], 1, Add, [1]], # 32-P3/8 fusion backbone P3 74 | [[21,22], 1, Add, [1]], # 33-P4/16 fusion backbone P4 75 | [[30,31], 1, Add, [1]], # 34-P5/32 fusion backbone P5 76 | 77 | ] 78 | 79 | 80 | # YOLOv5 head 81 | head: 82 | [ 83 | [-1, 1, Conv, [512, 1, 1]], # 35 84 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 36 85 | [[-1,33], 1, Concat, [1]], # 37 cat backbone P4 86 | [-1, 3, C3, [512, False]], # 38 87 | 88 | [-1, 1, Conv, [256, 1, 1]], # 39 89 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], # 40 90 | [[-1,32], 1, Concat, [1]], # 41 cat backbone P3 91 | [-1, 3, C3, [256, False]], # 42 (P3/8-small) 92 | 93 | [-1, 1, Conv, [256, 3, 2]], # 43 94 | [[-1,39], 1, Concat, [1]], # 44 cat head P4 95 | [-1, 3, C3, [512, False]], # 45 (P4/16-medium) 96 | 97 | [-1, 1, Conv, [512, 3, 2]], # 46 98 | [[-1,35], 1, Concat, [1]], # 47 cat head P5 99 | [-1, 3, C3, [1024, False]], # 48 (P5/32-large) 100 | 101 | [[42, 45, 48], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 102 | ] 103 | -------------------------------------------------------------------------------- /models/yolov5l.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, C3, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /models/yolov5m.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.67 # model depth multiple 4 | width_multiple: 0.75 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, C3, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /models/yolov5s.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.33 # model depth multiple 4 | width_multiple: 0.50 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, C3, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /models/yolov5x.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.33 # model depth multiple 4 | width_multiple: 1.25 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, C3, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- 1 | # pip install -r requirements.txt 2 | 3 | # base ---------------------------------------- 4 | matplotlib>=3.2.2 5 | numpy>=1.18.5 6 | opencv-python>=4.1.2 7 | Pillow 8 | PyYAML>=5.3.1 9 | scipy>=1.4.1 10 | torch>=1.7.0 11 | torchvision>=0.8.1 12 | tqdm>=4.41.0 13 | 14 | # logging ------------------------------------- 15 | tensorboard>=2.4.1 16 | # wandb 17 | 18 | # plotting ------------------------------------ 19 | seaborn>=0.11.0 20 | pandas 21 | 22 | # export -------------------------------------- 23 | # coremltools>=4.1 24 | # onnx>=1.8.1 25 | # scikit-learn==0.19.2 # for coreml quantization 26 | 27 | # extras -------------------------------------- 28 | thop # FLOPS computation 29 | pycocotools>=2.0 # COCO mAP 30 | -------------------------------------------------------------------------------- /utils/__init__.py: -------------------------------------------------------------------------------- 1 | 2 | -------------------------------------------------------------------------------- /utils/__pycache__/__init__.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/utils/__pycache__/__init__.cpython-37.pyc -------------------------------------------------------------------------------- /utils/__pycache__/autoanchor.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/utils/__pycache__/autoanchor.cpython-37.pyc -------------------------------------------------------------------------------- /utils/__pycache__/datasets.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/utils/__pycache__/datasets.cpython-37.pyc -------------------------------------------------------------------------------- /utils/__pycache__/ds_fusion.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/utils/__pycache__/ds_fusion.cpython-37.pyc -------------------------------------------------------------------------------- /utils/__pycache__/general.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/utils/__pycache__/general.cpython-37.pyc -------------------------------------------------------------------------------- /utils/__pycache__/google_utils.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/utils/__pycache__/google_utils.cpython-37.pyc -------------------------------------------------------------------------------- /utils/__pycache__/gradcam.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/utils/__pycache__/gradcam.cpython-37.pyc -------------------------------------------------------------------------------- /utils/__pycache__/loss.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/utils/__pycache__/loss.cpython-37.pyc -------------------------------------------------------------------------------- /utils/__pycache__/metrics.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/utils/__pycache__/metrics.cpython-37.pyc -------------------------------------------------------------------------------- /utils/__pycache__/plots.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/utils/__pycache__/plots.cpython-37.pyc -------------------------------------------------------------------------------- /utils/__pycache__/torch_utils.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/utils/__pycache__/torch_utils.cpython-37.pyc -------------------------------------------------------------------------------- /utils/activations.py: -------------------------------------------------------------------------------- 1 | # Activation functions 2 | 3 | import torch 4 | import torch.nn as nn 5 | import torch.nn.functional as F 6 | 7 | 8 | # SiLU https://arxiv.org/pdf/1606.08415.pdf ---------------------------------------------------------------------------- 9 | class SiLU(nn.Module): # export-friendly version of nn.SiLU() 10 | @staticmethod 11 | def forward(x): 12 | return x * torch.sigmoid(x) 13 | 14 | 15 | class Hardswish(nn.Module): # export-friendly version of nn.Hardswish() 16 | @staticmethod 17 | def forward(x): 18 | # return x * F.hardsigmoid(x) # for torchscript and CoreML 19 | return x * F.hardtanh(x + 3, 0., 6.) / 6. # for torchscript, CoreML and ONNX 20 | 21 | 22 | # Mish https://github.com/digantamisra98/Mish -------------------------------------------------------------------------- 23 | class Mish(nn.Module): 24 | @staticmethod 25 | def forward(x): 26 | return x * F.softplus(x).tanh() 27 | 28 | 29 | class MemoryEfficientMish(nn.Module): 30 | class F(torch.autograd.Function): 31 | @staticmethod 32 | def forward(ctx, x): 33 | ctx.save_for_backward(x) 34 | return x.mul(torch.tanh(F.softplus(x))) # x * tanh(ln(1 + exp(x))) 35 | 36 | @staticmethod 37 | def backward(ctx, grad_output): 38 | x = ctx.saved_tensors[0] 39 | sx = torch.sigmoid(x) 40 | fx = F.softplus(x).tanh() 41 | return grad_output * (fx + x * sx * (1 - fx * fx)) 42 | 43 | def forward(self, x): 44 | return self.F.apply(x) 45 | 46 | 47 | # FReLU https://arxiv.org/abs/2007.11824 ------------------------------------------------------------------------------- 48 | class FReLU(nn.Module): 49 | def __init__(self, c1, k=3): # ch_in, kernel 50 | super().__init__() 51 | self.conv = nn.Conv2d(c1, c1, k, 1, 1, groups=c1, bias=False) 52 | self.bn = nn.BatchNorm2d(c1) 53 | 54 | def forward(self, x): 55 | return torch.max(x, self.bn(self.conv(x))) 56 | 57 | 58 | # ACON https://arxiv.org/pdf/2009.04759.pdf ---------------------------------------------------------------------------- 59 | class AconC(nn.Module): 60 | r""" ACON activation (activate or not). 61 | AconC: (p1*x-p2*x) * sigmoid(beta*(p1*x-p2*x)) + p2*x, beta is a learnable parameter 62 | according to "Activate or Not: Learning Customized Activation" . 63 | """ 64 | 65 | def __init__(self, c1): 66 | super().__init__() 67 | self.p1 = nn.Parameter(torch.randn(1, c1, 1, 1)) 68 | self.p2 = nn.Parameter(torch.randn(1, c1, 1, 1)) 69 | self.beta = nn.Parameter(torch.ones(1, c1, 1, 1)) 70 | 71 | def forward(self, x): 72 | dpx = (self.p1 - self.p2) * x 73 | return dpx * torch.sigmoid(self.beta * dpx) + self.p2 * x 74 | 75 | 76 | class MetaAconC(nn.Module): 77 | r""" ACON activation (activate or not). 78 | MetaAconC: (p1*x-p2*x) * sigmoid(beta*(p1*x-p2*x)) + p2*x, beta is generated by a small network 79 | according to "Activate or Not: Learning Customized Activation" . 80 | """ 81 | 82 | def __init__(self, c1, k=1, s=1, r=16): # ch_in, kernel, stride, r 83 | super().__init__() 84 | c2 = max(r, c1 // r) 85 | self.p1 = nn.Parameter(torch.randn(1, c1, 1, 1)) 86 | self.p2 = nn.Parameter(torch.randn(1, c1, 1, 1)) 87 | self.fc1 = nn.Conv2d(c1, c2, k, s, bias=True) 88 | self.fc2 = nn.Conv2d(c2, c1, k, s, bias=True) 89 | # self.bn1 = nn.BatchNorm2d(c2) 90 | # self.bn2 = nn.BatchNorm2d(c1) 91 | 92 | def forward(self, x): 93 | y = x.mean(dim=2, keepdims=True).mean(dim=3, keepdims=True) 94 | # batch-size 1 bug/instabilities https://github.com/ultralytics/yolov5/issues/2891 95 | # beta = torch.sigmoid(self.bn2(self.fc2(self.bn1(self.fc1(y))))) # bug/unstable 96 | beta = torch.sigmoid(self.fc2(self.fc1(y))) # bug patch BN layers removed 97 | dpx = (self.p1 - self.p2) * x 98 | return dpx * torch.sigmoid(beta * dpx) + self.p2 * x 99 | -------------------------------------------------------------------------------- /utils/autoanchor.py: -------------------------------------------------------------------------------- 1 | # Auto-anchor utils 2 | 3 | import numpy as np 4 | import torch 5 | import yaml 6 | from scipy.cluster.vq import kmeans 7 | from tqdm import tqdm 8 | 9 | from utils.general import colorstr 10 | 11 | 12 | def check_anchor_order(m): 13 | # Check anchor order against stride order for YOLOv5 Detect() module m, and correct if necessary 14 | a = m.anchor_grid.prod(-1).view(-1) # anchor area 15 | da = a[-1] - a[0] # delta a 16 | ds = m.stride[-1] - m.stride[0] # delta s 17 | if da.sign() != ds.sign(): # same order 18 | print('Reversing anchor order') 19 | m.anchors[:] = m.anchors.flip(0) 20 | m.anchor_grid[:] = m.anchor_grid.flip(0) 21 | 22 | 23 | def check_anchors(dataset, model, thr=4.0, imgsz=640): 24 | # Check anchor fit to data, recompute if necessary 25 | prefix = colorstr('autoanchor: ') 26 | print(f'\n{prefix}Analyzing anchors... ', end='') 27 | m = model.module.model[-1] if hasattr(model, 'module') else model.model[-1] # Detect() 28 | shapes = imgsz * dataset.shapes / dataset.shapes.max(1, keepdims=True) 29 | scale = np.random.uniform(0.9, 1.1, size=(shapes.shape[0], 1)) # augment scale 30 | wh = torch.tensor(np.concatenate([l[:, 3:5] * s for s, l in zip(shapes * scale, dataset.labels)])).float() # wh 31 | 32 | def metric(k): # compute metric 33 | r = wh[:, None] / k[None] 34 | x = torch.min(r, 1. / r).min(2)[0] # ratio metric 35 | best = x.max(1)[0] # best_x 36 | aat = (x > 1. / thr).float().sum(1).mean() # anchors above threshold 37 | bpr = (best > 1. / thr).float().mean() # best possible recall 38 | return bpr, aat 39 | 40 | anchors = m.anchor_grid.clone().cpu().view(-1, 2) # current anchors 41 | bpr, aat = metric(anchors) 42 | print(f'anchors/target = {aat:.2f}, Best Possible Recall (BPR) = {bpr:.4f}', end='') 43 | if bpr < 0.98: # threshold to recompute 44 | print('. Attempting to improve anchors, please wait...') 45 | na = m.anchor_grid.numel() // 2 # number of anchors 46 | try: 47 | anchors = kmean_anchors(dataset, n=na, img_size=imgsz, thr=thr, gen=1000, verbose=False) 48 | except Exception as e: 49 | print(f'{prefix}ERROR: {e}') 50 | new_bpr = metric(anchors)[0] 51 | if new_bpr > bpr: # replace anchors 52 | anchors = torch.tensor(anchors, device=m.anchors.device).type_as(m.anchors) 53 | m.anchor_grid[:] = anchors.clone().view_as(m.anchor_grid) # for inference 54 | m.anchors[:] = anchors.clone().view_as(m.anchors) / m.stride.to(m.anchors.device).view(-1, 1, 1) # loss 55 | check_anchor_order(m) 56 | print(f'{prefix}New anchors saved to model. Update model *.yaml to use these anchors in the future.') 57 | else: 58 | print(f'{prefix}Original anchors better than new anchors. Proceeding with original anchors.') 59 | print('') # newline 60 | 61 | 62 | def check_anchors_rgb_ir(dataset, model, thr=4.0, imgsz=640): 63 | # Check anchor fit to data, recompute if necessary 64 | prefix = colorstr('autoanchor: ') 65 | print(f'\n{prefix}Analyzing anchors... ', end='') 66 | # m = model.module.model[-1] if hasattr(model, 'module') else model.model[-1] # Detect() 67 | m = list(model.model.children())[-1] 68 | print(m) 69 | shapes = imgsz * dataset.shapes / dataset.shapes.max(1, keepdims=True) 70 | scale = np.random.uniform(0.9, 1.1, size=(shapes.shape[0], 1)) # augment scale 71 | wh = torch.tensor(np.concatenate([l[:, 3:5] * s for s, l in zip(shapes * scale, dataset.labels)])).float() # wh 72 | 73 | def metric(k): # compute metric 74 | r = wh[:, None] / k[None] 75 | x = torch.min(r, 1. / r).min(2)[0] # ratio metric 76 | best = x.max(1)[0] # best_x 77 | aat = (x > 1. / thr).float().sum(1).mean() # anchors above threshold 78 | bpr = (best > 1. / thr).float().mean() # best possible recall 79 | return bpr, aat 80 | 81 | anchors = m.anchor_grid.clone().cpu().view(-1, 2) # current anchors 82 | bpr, aat = metric(anchors) 83 | print(f'anchors/target = {aat:.2f}, Best Possible Recall (BPR) = {bpr:.4f}', end='') 84 | if bpr < 0.98: # threshold to recompute 85 | print('. Attempting to improve anchors, please wait...') 86 | na = m.anchor_grid.numel() // 2 # number of anchors 87 | try: 88 | anchors = kmean_anchors(dataset, n=na, img_size=imgsz, thr=thr, gen=1000, verbose=False) 89 | except Exception as e: 90 | print(f'{prefix}ERROR: {e}') 91 | new_bpr = metric(anchors)[0] 92 | if new_bpr > bpr: # replace anchors 93 | anchors = torch.tensor(anchors, device=m.anchors.device).type_as(m.anchors) 94 | m.anchor_grid[:] = anchors.clone().view_as(m.anchor_grid) # for inference 95 | m.anchors[:] = anchors.clone().view_as(m.anchors) / m.stride.to(m.anchors.device).view(-1, 1, 1) # loss 96 | check_anchor_order(m) 97 | print(f'{prefix}New anchors saved to model. Update model *.yaml to use these anchors in the future.') 98 | else: 99 | print(f'{prefix}Original anchors better than new anchors. Proceeding with original anchors.') 100 | print('') # newline 101 | 102 | 103 | def kmean_anchors(path='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen=1000, verbose=True): 104 | """ Creates kmeans-evolved anchors from training dataset 105 | 106 | Arguments: 107 | path: path to dataset *.yaml, or a loaded dataset 108 | n: number of anchors 109 | img_size: image size used for training 110 | thr: anchor-label wh ratio threshold hyperparameter hyp['anchor_t'] used for training, default=4.0 111 | gen: generations to evolve anchors using genetic algorithm 112 | verbose: print all results 113 | 114 | Return: 115 | k: kmeans evolved anchors 116 | 117 | Usage: 118 | from utils.autoanchor import *; _ = kmean_anchors() 119 | """ 120 | thr = 1. / thr 121 | prefix = colorstr('autoanchor: ') 122 | 123 | def metric(k, wh): # compute metrics 124 | r = wh[:, None] / k[None] 125 | x = torch.min(r, 1. / r).min(2)[0] # ratio metric 126 | # x = wh_iou(wh, torch.tensor(k)) # iou metric 127 | return x, x.max(1)[0] # x, best_x 128 | 129 | def anchor_fitness(k): # mutation fitness 130 | _, best = metric(torch.tensor(k, dtype=torch.float32), wh) 131 | return (best * (best > thr).float()).mean() # fitness 132 | 133 | def print_results(k): 134 | k = k[np.argsort(k.prod(1))] # sort small to large 135 | x, best = metric(k, wh0) 136 | bpr, aat = (best > thr).float().mean(), (x > thr).float().mean() * n # best possible recall, anch > thr 137 | print(f'{prefix}thr={thr:.2f}: {bpr:.4f} best possible recall, {aat:.2f} anchors past thr') 138 | print(f'{prefix}n={n}, img_size={img_size}, metric_all={x.mean():.3f}/{best.mean():.3f}-mean/best, ' 139 | f'past_thr={x[x > thr].mean():.3f}-mean: ', end='') 140 | for i, x in enumerate(k): 141 | print('%i,%i' % (round(x[0]), round(x[1])), end=', ' if i < len(k) - 1 else '\n') # use in *.cfg 142 | return k 143 | 144 | if isinstance(path, str): # *.yaml file 145 | with open(path) as f: 146 | data_dict = yaml.safe_load(f) # model dict 147 | from utils.datasets import LoadImagesAndLabels 148 | dataset = LoadImagesAndLabels(data_dict['train'], augment=True, rect=True) 149 | else: 150 | dataset = path # dataset 151 | 152 | # Get label wh 153 | shapes = img_size * dataset.shapes / dataset.shapes.max(1, keepdims=True) 154 | wh0 = np.concatenate([l[:, 3:5] * s for s, l in zip(shapes, dataset.labels)]) # wh 155 | 156 | # Filter 157 | i = (wh0 < 3.0).any(1).sum() 158 | if i: 159 | print(f'{prefix}WARNING: Extremely small objects found. {i} of {len(wh0)} labels are < 3 pixels in size.') 160 | wh = wh0[(wh0 >= 2.0).any(1)] # filter > 2 pixels 161 | # wh = wh * (np.random.rand(wh.shape[0], 1) * 0.9 + 0.1) # multiply by random scale 0-1 162 | 163 | # Kmeans calculation 164 | print(f'{prefix}Running kmeans for {n} anchors on {len(wh)} points...') 165 | s = wh.std(0) # sigmas for whitening 166 | k, dist = kmeans(wh / s, n, iter=30) # points, mean distance 167 | assert len(k) == n, print(f'{prefix}ERROR: scipy.cluster.vq.kmeans requested {n} points but returned only {len(k)}') 168 | k *= s 169 | wh = torch.tensor(wh, dtype=torch.float32) # filtered 170 | wh0 = torch.tensor(wh0, dtype=torch.float32) # unfiltered 171 | k = print_results(k) 172 | 173 | # Plot 174 | # k, d = [None] * 20, [None] * 20 175 | # for i in tqdm(range(1, 21)): 176 | # k[i-1], d[i-1] = kmeans(wh / s, i) # points, mean distance 177 | # fig, ax = plt.subplots(1, 2, figsize=(14, 7), tight_layout=True) 178 | # ax = ax.ravel() 179 | # ax[0].plot(np.arange(1, 21), np.array(d) ** 2, marker='.') 180 | # fig, ax = plt.subplots(1, 2, figsize=(14, 7)) # plot wh 181 | # ax[0].hist(wh[wh[:, 0]<100, 0],400) 182 | # ax[1].hist(wh[wh[:, 1]<100, 1],400) 183 | # fig.savefig('wh.png', dpi=200) 184 | 185 | # Evolve 186 | npr = np.random 187 | f, sh, mp, s = anchor_fitness(k), k.shape, 0.9, 0.1 # fitness, generations, mutation prob, sigma 188 | pbar = tqdm(range(gen), desc=f'{prefix}Evolving anchors with Genetic Algorithm:') # progress bar 189 | for _ in pbar: 190 | v = np.ones(sh) 191 | while (v == 1).all(): # mutate until a change occurs (prevent duplicates) 192 | v = ((npr.random(sh) < mp) * npr.random() * npr.randn(*sh) * s + 1).clip(0.3, 3.0) 193 | kg = (k.copy() * v).clip(min=2.0) 194 | fg = anchor_fitness(kg) 195 | if fg > f: 196 | f, k = fg, kg.copy() 197 | pbar.desc = f'{prefix}Evolving anchors with Genetic Algorithm: fitness = {f:.4f}' 198 | if verbose: 199 | print_results(k) 200 | 201 | return print_results(k) 202 | -------------------------------------------------------------------------------- /utils/aws/__init__.py: -------------------------------------------------------------------------------- 1 | 2 | -------------------------------------------------------------------------------- /utils/aws/mime.sh: -------------------------------------------------------------------------------- 1 | # AWS EC2 instance startup 'MIME' script https://aws.amazon.com/premiumsupport/knowledge-center/execute-user-data-ec2/ 2 | # This script will run on every instance restart, not only on first start 3 | # --- DO NOT COPY ABOVE COMMENTS WHEN PASTING INTO USERDATA --- 4 | 5 | Content-Type: multipart/mixed; boundary="//" 6 | MIME-Version: 1.0 7 | 8 | --// 9 | Content-Type: text/cloud-config; charset="us-ascii" 10 | MIME-Version: 1.0 11 | Content-Transfer-Encoding: 7bit 12 | Content-Disposition: attachment; filename="cloud-config.txt" 13 | 14 | #cloud-config 15 | cloud_final_modules: 16 | - [scripts-user, always] 17 | 18 | --// 19 | Content-Type: text/x-shellscript; charset="us-ascii" 20 | MIME-Version: 1.0 21 | Content-Transfer-Encoding: 7bit 22 | Content-Disposition: attachment; filename="userdata.txt" 23 | 24 | #!/bin/bash 25 | # --- paste contents of userdata.sh here --- 26 | --// 27 | -------------------------------------------------------------------------------- /utils/aws/resume.py: -------------------------------------------------------------------------------- 1 | # Resume all interrupted trainings in yolov5/ dir including DDP trainings 2 | # Usage: $ python utils/aws/resume.py 3 | 4 | import os 5 | import sys 6 | from pathlib import Path 7 | 8 | import torch 9 | import yaml 10 | 11 | sys.path.append('./') # to run '$ python *.py' files in subdirectories 12 | 13 | port = 0 # --master_port 14 | path = Path('').resolve() 15 | for last in path.rglob('*/**/last.pt'): 16 | ckpt = torch.load(last) 17 | if ckpt['optimizer'] is None: 18 | continue 19 | 20 | # Load opt.yaml 21 | with open(last.parent.parent / 'opt.yaml') as f: 22 | opt = yaml.safe_load(f) 23 | 24 | # Get device count 25 | d = opt['device'].split(',') # devices 26 | nd = len(d) # number of devices 27 | ddp = nd > 1 or (nd == 0 and torch.cuda.device_count() > 1) # distributed data parallel 28 | 29 | if ddp: # multi-GPU 30 | port += 1 31 | cmd = f'python -m torch.distributed.launch --nproc_per_node {nd} --master_port {port} train.py --resume {last}' 32 | else: # single-GPU 33 | cmd = f'python train.py --resume {last}' 34 | 35 | cmd += ' > /dev/null 2>&1 &' # redirect output to dev/null and run in daemon thread 36 | print(cmd) 37 | os.system(cmd) 38 | -------------------------------------------------------------------------------- /utils/aws/userdata.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | # AWS EC2 instance startup script https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/user-data.html 3 | # This script will run only once on first instance start (for a re-start script see mime.sh) 4 | # /home/ubuntu (ubuntu) or /home/ec2-user (amazon-linux) is working dir 5 | # Use >300 GB SSD 6 | 7 | cd home/ubuntu 8 | if [ ! -d yolov5 ]; then 9 | echo "Running first-time script." # install dependencies, download COCO, pull Docker 10 | git clone https://github.com/ultralytics/yolov5 && sudo chmod -R 777 yolov5 11 | cd yolov5 12 | bash data/scripts/get_coco.sh && echo "Data done." & 13 | sudo docker pull ultralytics/yolov5:latest && echo "Docker done." & 14 | python -m pip install --upgrade pip && pip install -r requirements.txt && python detect.py && echo "Requirements done." & 15 | wait && echo "All tasks done." # finish background tasks 16 | else 17 | echo "Running re-start script." # resume interrupted runs 18 | i=0 19 | list=$(sudo docker ps -qa) # container list i.e. $'one\ntwo\nthree\nfour' 20 | while IFS= read -r id; do 21 | ((i++)) 22 | echo "restarting container $i: $id" 23 | sudo docker start $id 24 | # sudo docker exec -it $id python train.py --resume # single-GPU 25 | sudo docker exec -d $id python utils/aws/resume.py # multi-scenario 26 | done <<<"$list" 27 | fi 28 | -------------------------------------------------------------------------------- /utils/ds_fusion.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import pandas as pd 3 | import os 4 | import full_arrange 5 | 6 | 7 | def IIM_of_Li(data): 8 | 9 | [evidence_number, data_frame_number] = data.shape 10 | for i in range(evidence_number): 11 | print(str(i) + 'th evidence is ' + str(data[i, :])) 12 | 13 | # 对于识别框架内每一个命题 14 | new_data = np.zeros((evidence_number, data_frame_number + 1)) 15 | for k in range(data_frame_number): 16 | # 计算相容矩阵 17 | R = np.zeros((evidence_number, evidence_number)) 18 | for i in range(evidence_number): 19 | for j in range(evidence_number): 20 | R[i,j] = (data[i,k] * data[j,k]) / (np.square(data[i,k]) + np.square(data[j,k])) * 2 21 | if np.isnan(R[i,j]): 22 | R[i,j] = 0 23 | print(R) 24 | # 计算绝对相容度 25 | D = np.zeros((evidence_number,1)) 26 | for i in range(evidence_number): 27 | D[i,0] = sum(R[i,:]) - 1 28 | # print("+++++++++") 29 | # print(D) 30 | # 计算可信度 31 | W = np.zeros((evidence_number,1)) 32 | for i in range(evidence_number): 33 | W[i,0] = D[i,0] / (evidence_number - 1) 34 | # print("-------------") 35 | # print(W) 36 | 37 | # 根据可行度重新确定该证据对应命题的权重，对基本概率指派函数进行处理，得到新的mass函数 38 | for i in range(evidence_number): 39 | new_data[i, k] = data[i, k] * W[i, 0] 40 | 41 | # 提取不确定事件 \Theta 42 | for i in range(evidence_number): 43 | new_data[i, data_frame_number] = 1 - sum(new_data[i, 0:-1]) 44 | print(str(i) + 'th new_evidence is ' + str(new_data[i, :])) 45 | 46 | return new_data 47 | 48 | 49 | def IIM_of_sun(data): 50 | [evidence_number, data_frame_number] = data.shape 51 | for i in range(evidence_number): 52 | print(str(i) + 'th evidence is ' + str(data[i, :])) 53 | 54 | #冲突矩阵k_m 55 | k_m = np.zeros((evidence_number,evidence_number)) 56 | for i in range(evidence_number): 57 | for j in range(evidence_number): 58 | sum3 = 0 59 | for k in range(data_frame_number): 60 | sum3 = sum3 + data[i,k] * (sum(data[j,:]) - data[j,k]) 61 | k_m[i,j] = sum3 62 | 63 | k_sun = 0 64 | for i in range(evidence_number): 65 | for j in range(evidence_number): 66 | if i < j: 67 | k_sun = k_sun + k_m[i,j] 68 | #全部证据可信度 69 | epsilon = k_sun / (evidence_number * (evidence_number - 1) / 2) 70 | #平均概率 71 | q = np.zeros((1, data_frame_number)) 72 | for i in range(data_frame_number): 73 | q[0,i] = sum(data[:,i]) / evidence_number 74 | 75 | # print('epsilon为 ' + str(epsilon)) 76 | # print('q为 ' + str(q)) 77 | return epsilon, q 78 | 79 | 80 | 81 | def DS_fusion_method(data): 82 | new_data = data 83 | [evidence_number, data_frame_number] = new_data.shape 84 | if evidence_number > 3: 85 | print('can not do') 86 | return 87 | 88 | #算出全部组合 89 | combination = full_arrange.full_arrange(range(data_frame_number), evidence_number) 90 | count = 0 91 | for i in combination: 92 | # print(i) 93 | count = count + 1 94 | print('组合一共 ' + str(count) + ' 个') 95 | 96 | #先算归一化因子K 97 | sum1 = 0 98 | for k in combination: 99 | count = 0 100 | small_set = set(k) 101 | for i in small_set: 102 | if i in set(range(data_frame_number - 1)): 103 | count = count + 1 104 | if count >= 2: 105 | multi = 1 106 | for i in range(evidence_number): 107 | multi = multi * new_data[i,k[i]] 108 | sum1 = sum1 + multi 109 | K = 1 - sum1 110 | print('归一化因子K为 ' + str(K)) 111 | 112 | #计算每一个特征融合概率 113 | fusion = np.zeros((1,data_frame_number)) 114 | for i in range(data_frame_number - 1): 115 | list1 = [i, data_frame_number - 1] 116 | small_combination = full_arrange.full_arrange(list1, evidence_number) 117 | full_info_list = (np.ones((1,evidence_number)) * (data_frame_number - 1)).tolist()[0] 118 | full_info_list = [int(i) for i in full_info_list] 119 | small_combination.remove(full_info_list) 120 | sum2 = 0 121 | for j in small_combination: 122 | multi = 1 123 | for k in range(evidence_number): 124 | multi = multi * new_data[k,j[k]] 125 | sum2 = sum2 + multi 126 | fusion[0,i] = sum2 / K 127 | 128 | multi = 1 129 | for i in range(evidence_number): 130 | multi = multi * new_data[i,data_frame_number - 1] 131 | fusion[0,data_frame_number - 1] = multi / K 132 | 133 | print('DS Fusion ' + str(fusion)) 134 | return fusion, K 135 | 136 | def use_DS_method_of_sun(data): 137 | [epsilon, q] = IIM_of_sun(data) 138 | add_line = np.zeros((data.shape[0])) 139 | data_with_all = np.c_[data, add_line] 140 | [fusion, K] = DS_fusion_method(data_with_all) 141 | num = fusion.shape[1] 142 | for i in range(num - 1): 143 | fusion[0, i] = K * fusion[0, i] + (1 - K) * epsilon * q[0, i] 144 | 145 | fusion[0, num - 1] = (1 - K) * (1 - epsilon) 146 | # 147 | # label.append('ALL') 148 | fusion_all = np.c_['0,2', data_with_all, fusion] 149 | fusion_all = pd.DataFrame(data=fusion_all, index=['ProcessData', 'Alert','c', 'Fusion']) 150 | print(fusion_all) 151 | return fusion_all 152 | 153 | 154 | 155 | if __name__ == '__main__': 156 | #测试data 157 | data = [[0.9, 0.1]] 158 | data.append([0.8, 0.2]) 159 | 160 | data = np.array(data) 161 | 162 | 163 | # new_data = [[0.2459, 0.1224, 0.24, 0.3917]] 164 | # new_data.append([0, 0.2894, 0.06, 0.6506]) 165 | # # new_data.append([0.2951, 0.056, 0.24, 0.4139]) 166 | # new_data = np.array(new_data) 167 | 168 | # data_with_all = [[0.5, 0.2, 0.3, 0]] 169 | # data_with_all.append([0, 0.9, 0.1, 0]) 170 | # data_with_all.append([0.6, 0.1, 0.3, 0]) 171 | # data_with_all = np.array(data_with_all) 172 | 173 | # 测试Li的方法 174 | new_data = IIM_of_Li(data) 175 | DS_fusion_method(new_data) 176 | 177 | # # 测试sun方法 178 | use_DS_method_of_sun(data) 179 | -------------------------------------------------------------------------------- /utils/flask_rest_api/README.md: -------------------------------------------------------------------------------- 1 | # Flask REST API 2 | [REST](https://en.wikipedia.org/wiki/Representational_state_transfer) [API](https://en.wikipedia.org/wiki/API)s are commonly used to expose Machine Learning (ML) models to other services. This folder contains an example REST API created using Flask to expose the `yolov5s` model from [PyTorch Hub](https://pytorch.org/hub/ultralytics_yolov5/). 3 | 4 | ## Requirements 5 | 6 | [Flask](https://palletsprojects.com/p/flask/) is required. Install with: 7 | ```shell 8 | $ pip install Flask 9 | ``` 10 | 11 | ## Run 12 | 13 | After Flask installation run: 14 | 15 | ```shell 16 | $ python3 restapi.py --port 5000 17 | ``` 18 | 19 | Then use [curl](https://curl.se/) to perform a request: 20 | 21 | ```shell 22 | $ curl -X POST -F image=@zidane.jpg 'http://localhost:5000/v1/object-detection/yolov5s'` 23 | ``` 24 | 25 | The model inference results are returned: 26 | 27 | ```shell 28 | [{'class': 0, 29 | 'confidence': 0.8197850585, 30 | 'name': 'person', 31 | 'xmax': 1159.1403808594, 32 | 'xmin': 750.912902832, 33 | 'ymax': 711.2583007812, 34 | 'ymin': 44.0350036621}, 35 | {'class': 0, 36 | 'confidence': 0.5667674541, 37 | 'name': 'person', 38 | 'xmax': 1065.5523681641, 39 | 'xmin': 116.0448303223, 40 | 'ymax': 713.8904418945, 41 | 'ymin': 198.4603881836}, 42 | {'class': 27, 43 | 'confidence': 0.5661227107, 44 | 'name': 'tie', 45 | 'xmax': 516.7975463867, 46 | 'xmin': 416.6880187988, 47 | 'ymax': 717.0524902344, 48 | 'ymin': 429.2020568848}] 49 | ``` 50 | 51 | An example python script to perform inference using [requests](https://docs.python-requests.org/en/master/) is given in `example_request.py` 52 | -------------------------------------------------------------------------------- /utils/flask_rest_api/example_request.py: -------------------------------------------------------------------------------- 1 | """Perform test request""" 2 | import pprint 3 | 4 | import requests 5 | 6 | DETECTION_URL = "http://localhost:5000/v1/object-detection/yolov5s" 7 | TEST_IMAGE = "zidane.jpg" 8 | 9 | image_data = open(TEST_IMAGE, "rb").read() 10 | 11 | response = requests.post(DETECTION_URL, files={"image": image_data}).json() 12 | 13 | pprint.pprint(response) 14 | -------------------------------------------------------------------------------- /utils/flask_rest_api/restapi.py: -------------------------------------------------------------------------------- 1 | """ 2 | Run a rest API exposing the yolov5s object detection model 3 | """ 4 | import argparse 5 | import io 6 | 7 | import torch 8 | from PIL import Image 9 | from flask import Flask, request 10 | 11 | app = Flask(__name__) 12 | 13 | DETECTION_URL = "/v1/object-detection/yolov5s" 14 | 15 | 16 | @app.route(DETECTION_URL, methods=["POST"]) 17 | def predict(): 18 | if not request.method == "POST": 19 | return 20 | 21 | if request.files.get("image"): 22 | image_file = request.files["image"] 23 | image_bytes = image_file.read() 24 | 25 | img = Image.open(io.BytesIO(image_bytes)) 26 | 27 | results = model(img, size=640) # reduce size=320 for faster inference 28 | return results.pandas().xyxy[0].to_json(orient="records") 29 | 30 | 31 | if __name__ == "__main__": 32 | parser = argparse.ArgumentParser(description="Flask API exposing YOLOv5 model") 33 | parser.add_argument("--port", default=5000, type=int, help="port number") 34 | args = parser.parse_args() 35 | 36 | model = torch.hub.load("ultralytics/yolov5", "yolov5s", force_reload=True) # force_reload to recache 37 | app.run(host="0.0.0.0", port=args.port) # debug=True causes Restarting with stat 38 | -------------------------------------------------------------------------------- /utils/google_app_engine/Dockerfile: -------------------------------------------------------------------------------- 1 | FROM gcr.io/google-appengine/python 2 | 3 | # Create a virtualenv for dependencies. This isolates these packages from 4 | # system-level packages. 5 | # Use -p python3 or -p python3.7 to select python version. Default is version 2. 6 | RUN virtualenv /env -p python3 7 | 8 | # Setting these environment variables are the same as running 9 | # source /env/bin/activate. 10 | ENV VIRTUAL_ENV /env 11 | ENV PATH /env/bin:$PATH 12 | 13 | RUN apt-get update && apt-get install -y python-opencv 14 | 15 | # Copy the application's requirements.txt and run pip to install all 16 | # dependencies into the virtualenv. 17 | ADD requirements.txt /app/requirements.txt 18 | RUN pip install -r /app/requirements.txt 19 | 20 | # Add the application source code. 21 | ADD . /app 22 | 23 | # Run a WSGI server to serve the application. gunicorn must be declared as 24 | # a dependency in requirements.txt. 25 | CMD gunicorn -b :$PORT main:app 26 | -------------------------------------------------------------------------------- /utils/google_app_engine/additional_requirements.txt: -------------------------------------------------------------------------------- 1 | # add these requirements in your app on top of the existing ones 2 | pip==18.1 3 | Flask==1.0.2 4 | gunicorn==19.9.0 5 | -------------------------------------------------------------------------------- /utils/google_app_engine/app.yaml: -------------------------------------------------------------------------------- 1 | runtime: custom 2 | env: flex 3 | 4 | service: yolov5app 5 | 6 | liveness_check: 7 | initial_delay_sec: 600 8 | 9 | manual_scaling: 10 | instances: 1 11 | resources: 12 | cpu: 1 13 | memory_gb: 4 14 | disk_size_gb: 20 -------------------------------------------------------------------------------- /utils/google_utils.py: -------------------------------------------------------------------------------- 1 | # Google utils: https://cloud.google.com/storage/docs/reference/libraries 2 | 3 | import os 4 | import platform 5 | import subprocess 6 | import time 7 | from pathlib import Path 8 | 9 | import requests 10 | import torch 11 | 12 | 13 | def gsutil_getsize(url=''): 14 | # gs://bucket/file size https://cloud.google.com/storage/docs/gsutil/commands/du 15 | s = subprocess.check_output(f'gsutil du {url}', shell=True).decode('utf-8') 16 | return eval(s.split(' ')[0]) if len(s) else 0 # bytes 17 | 18 | 19 | def attempt_download(file, repo='ultralytics/yolov5'): 20 | # Attempt file download if does not exist 21 | file = Path(str(file).strip().replace("'", '')) 22 | 23 | if not file.exists(): 24 | try: 25 | response = requests.get(f'https://api.github.com/repos/{repo}/releases/latest').json() # github api 26 | assets = [x['name'] for x in response['assets']] # release assets, i.e. ['yolov5s.pt', 'yolov5m.pt', ...] 27 | tag = response['tag_name'] # i.e. 'v1.0' 28 | except: # fallback plan 29 | assets = ['yolov5s.pt', 'yolov5m.pt', 'yolov5l.pt', 'yolov5x.pt', 30 | 'yolov5s6.pt', 'yolov5m6.pt', 'yolov5l6.pt', 'yolov5x6.pt'] 31 | try: 32 | tag = subprocess.check_output('git tag', shell=True, stderr=subprocess.STDOUT).decode().split()[-1] 33 | except: 34 | tag = 'v5.0' # current release 35 | 36 | name = file.name 37 | if name in assets: 38 | msg = f'{file} missing, try downloading from https://github.com/{repo}/releases/' 39 | redundant = False # second download option 40 | try: # GitHub 41 | url = f'https://github.com/{repo}/releases/download/{tag}/{name}' 42 | print(f'Downloading {url} to {file}...') 43 | torch.hub.download_url_to_file(url, file) 44 | assert file.exists() and file.stat().st_size > 1E6 # check 45 | except Exception as e: # GCP 46 | print(f'Download error: {e}') 47 | assert redundant, 'No secondary mirror' 48 | url = f'https://storage.googleapis.com/{repo}/ckpt/{name}' 49 | print(f'Downloading {url} to {file}...') 50 | os.system(f'curl -L {url} -o {file}') # torch.hub.download_url_to_file(url, weights) 51 | finally: 52 | if not file.exists() or file.stat().st_size < 1E6: # check 53 | file.unlink(missing_ok=True) # remove partial downloads 54 | print(f'ERROR: Download failure: {msg}') 55 | print('') 56 | return 57 | 58 | 59 | def gdrive_download(id='16TiPfZj7htmTyhntwcZyEEAejOUxuT6m', file='tmp.zip'): 60 | # Downloads a file from Google Drive. from yolov5.utils.google_utils import *; gdrive_download() 61 | t = time.time() 62 | file = Path(file) 63 | cookie = Path('cookie') # gdrive cookie 64 | print(f'Downloading https://drive.google.com/uc?export=download&id={id} as {file}... ', end='') 65 | file.unlink(missing_ok=True) # remove existing file 66 | cookie.unlink(missing_ok=True) # remove existing cookie 67 | 68 | # Attempt file download 69 | out = "NUL" if platform.system() == "Windows" else "/dev/null" 70 | os.system(f'curl -c ./cookie -s -L "drive.google.com/uc?export=download&id={id}" > {out}') 71 | if os.path.exists('cookie'): # large file 72 | s = f'curl -Lb ./cookie "drive.google.com/uc?export=download&confirm={get_token()}&id={id}" -o {file}' 73 | else: # small file 74 | s = f'curl -s -L -o {file} "drive.google.com/uc?export=download&id={id}"' 75 | r = os.system(s) # execute, capture return 76 | cookie.unlink(missing_ok=True) # remove existing cookie 77 | 78 | # Error check 79 | if r != 0: 80 | file.unlink(missing_ok=True) # remove partial 81 | print('Download error ') # raise Exception('Download error') 82 | return r 83 | 84 | # Unzip if archive 85 | if file.suffix == '.zip': 86 | print('unzipping... ', end='') 87 | os.system(f'unzip -q {file}') # unzip 88 | file.unlink() # remove zip to free space 89 | 90 | print(f'Done ({time.time() - t:.1f}s)') 91 | return r 92 | 93 | 94 | def get_token(cookie="./cookie"): 95 | with open(cookie) as f: 96 | for line in f: 97 | if "download" in line: 98 | return line.split()[-1] 99 | return "" 100 | 101 | # def upload_blob(bucket_name, source_file_name, destination_blob_name): 102 | # # Uploads a file to a bucket 103 | # # https://cloud.google.com/storage/docs/uploading-objects#storage-upload-object-python 104 | # 105 | # storage_client = storage.Client() 106 | # bucket = storage_client.get_bucket(bucket_name) 107 | # blob = bucket.blob(destination_blob_name) 108 | # 109 | # blob.upload_from_filename(source_file_name) 110 | # 111 | # print('File {} uploaded to {}.'.format( 112 | # source_file_name, 113 | # destination_blob_name)) 114 | # 115 | # 116 | # def download_blob(bucket_name, source_blob_name, destination_file_name): 117 | # # Uploads a blob from a bucket 118 | # storage_client = storage.Client() 119 | # bucket = storage_client.get_bucket(bucket_name) 120 | # blob = bucket.blob(source_blob_name) 121 | # 122 | # blob.download_to_filename(destination_file_name) 123 | # 124 | # print('Blob {} downloaded to {}.'.format( 125 | # source_blob_name, 126 | # destination_file_name)) 127 | -------------------------------------------------------------------------------- /utils/gradcam.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python 2 | # -*- coding:utf-8 -*- 3 | # Author: Richard Fang 4 | import torch 5 | from torch.autograd import Variable 6 | from torch.autograd import Function 7 | from torchvision import models 8 | from torchvision import utils 9 | import cv2 10 | import sys 11 | import numpy as np 12 | import argparse 13 | 14 | import warnings 15 | warnings.filterwarnings("ignore") 16 | 17 | 18 | def preprocess_image(img): 19 | means=[0.485, 0.456, 0.406] 20 | stds=[0.229, 0.224, 0.225] 21 | 22 | preprocessed_img = img.copy()[:, :, ::-1] 23 | for i in range(3): 24 | preprocessed_img[:, :, i] = preprocessed_img[:, :, i] - means[i] 25 | preprocessed_img[:, :, i] = preprocessed_img[:, :, i] / stds[i] 26 | preprocessed_img = \ 27 | np.ascontiguousarray(np.transpose(preprocessed_img, (2, 0, 1))) 28 | preprocessed_img = torch.from_numpy(preprocessed_img) 29 | preprocessed_img.unsqueeze_(0) 30 | input = Variable(preprocessed_img, requires_grad = True) 31 | return input 32 | 33 | 34 | def show_cam_on_image(img, mask, epoch, layer): 35 | heatmap = cv2.applyColorMap(np.uint8(255*mask), cv2.COLORMAP_JET) 36 | heatmap = np.float32(heatmap) / 255 37 | # img = cv2.flip(img, 1) 38 | cam = heatmap + np.float32(img) 39 | cam = cam / np.max(cam) 40 | print(type(cam)) 41 | cv2.imwrite("./cam%d_%d.jpg" %(epoch, layer), np.uint8(255 * cam)) 42 | 43 | 44 | def calcGradCam(imgpath, feature, epoch, layer): 45 | """ 46 | 47 | :param feature: 特征图 48 | :param grad_val: 对应的特征图的梯度 49 | :return 50 | """ 51 | # ------------ Image Preprocess --------------- 52 | image_path = imgpath 53 | img = cv2.imread(image_path, 1) 54 | # print(np.shape(img)) 55 | img = np.float32(cv2.resize(img, (640, 640))) / 255 56 | input = preprocess_image(img) 57 | 58 | # ------------ GradCam ------------------------- 59 | feature = feature.cpu().data.numpy() 60 | print("feature shape", feature.shape) 61 | cam = np.zeros(feature.shape[1:], dtype=np.float32) 62 | 63 | for i in range(feature.shape[0]): 64 | cam += feature[i, :, :] 65 | 66 | cam = np.maximum(cam, 0) # 比较cam的元素与0的大小，<0的都置0 67 | # min = np.min(cam) 68 | # cam -= min 69 | cam = cv2.resize(cam, (640, 640)) 70 | 71 | # 归一化 cam 72 | cam = cam - np.min(cam) 73 | cam = cam / np.max(cam) 74 | 75 | print(cam.shape) 76 | # ---------- Show ------------- 77 | show_cam_on_image(img, cam, epoch, layer) 78 | 79 | return cam 80 | 81 | -------------------------------------------------------------------------------- /utils/loss.py: -------------------------------------------------------------------------------- 1 | # Loss functions 2 | 3 | import torch 4 | import torch.nn as nn 5 | 6 | from utils.general import bbox_iou 7 | from utils.torch_utils import is_parallel 8 | 9 | 10 | def smooth_BCE(eps=0.1): # https://github.com/ultralytics/yolov3/issues/238#issuecomment-598028441 11 | # return positive, negative label smoothing BCE targets 12 | return 1.0 - 0.5 * eps, 0.5 * eps 13 | 14 | 15 | class BCEBlurWithLogitsLoss(nn.Module): 16 | # BCEwithLogitLoss() with reduced missing label effects. 17 | def __init__(self, alpha=0.05): 18 | super(BCEBlurWithLogitsLoss, self).__init__() 19 | self.loss_fcn = nn.BCEWithLogitsLoss(reduction='none') # must be nn.BCEWithLogitsLoss() 20 | self.alpha = alpha 21 | 22 | def forward(self, pred, true): 23 | loss = self.loss_fcn(pred, true) 24 | pred = torch.sigmoid(pred) # prob from logits 25 | dx = pred - true # reduce only missing label effects 26 | # dx = (pred - true).abs() # reduce missing label and false label effects 27 | alpha_factor = 1 - torch.exp((dx - 1) / (self.alpha + 1e-4)) 28 | loss *= alpha_factor 29 | return loss.mean() 30 | 31 | 32 | class FocalLoss(nn.Module): 33 | # Wraps focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5) 34 | def __init__(self, loss_fcn, gamma=1.5, alpha=0.25): 35 | super(FocalLoss, self).__init__() 36 | self.loss_fcn = loss_fcn # must be nn.BCEWithLogitsLoss() 37 | self.gamma = gamma 38 | self.alpha = alpha 39 | self.reduction = loss_fcn.reduction 40 | self.loss_fcn.reduction = 'none' # required to apply FL to each element 41 | 42 | def forward(self, pred, true): 43 | loss = self.loss_fcn(pred, true) 44 | # p_t = torch.exp(-loss) 45 | # loss *= self.alpha * (1.000001 - p_t) ** self.gamma # non-zero power for gradient stability 46 | 47 | # TF implementation https://github.com/tensorflow/addons/blob/v0.7.1/tensorflow_addons/losses/focal_loss.py 48 | pred_prob = torch.sigmoid(pred) # prob from logits 49 | p_t = true * pred_prob + (1 - true) * (1 - pred_prob) 50 | alpha_factor = true * self.alpha + (1 - true) * (1 - self.alpha) 51 | modulating_factor = (1.0 - p_t) ** self.gamma 52 | loss *= alpha_factor * modulating_factor 53 | 54 | if self.reduction == 'mean': 55 | return loss.mean() 56 | elif self.reduction == 'sum': 57 | return loss.sum() 58 | else: # 'none' 59 | return loss 60 | 61 | 62 | class QFocalLoss(nn.Module): 63 | # Wraps Quality focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5) 64 | def __init__(self, loss_fcn, gamma=1.5, alpha=0.25): 65 | super(QFocalLoss, self).__init__() 66 | self.loss_fcn = loss_fcn # must be nn.BCEWithLogitsLoss() 67 | self.gamma = gamma 68 | self.alpha = alpha 69 | self.reduction = loss_fcn.reduction 70 | self.loss_fcn.reduction = 'none' # required to apply FL to each element 71 | 72 | def forward(self, pred, true): 73 | loss = self.loss_fcn(pred, true) 74 | 75 | pred_prob = torch.sigmoid(pred) # prob from logits 76 | alpha_factor = true * self.alpha + (1 - true) * (1 - self.alpha) 77 | modulating_factor = torch.abs(true - pred_prob) ** self.gamma 78 | loss *= alpha_factor * modulating_factor 79 | 80 | if self.reduction == 'mean': 81 | return loss.mean() 82 | elif self.reduction == 'sum': 83 | return loss.sum() 84 | else: # 'none' 85 | return loss 86 | 87 | 88 | class ComputeLoss: 89 | # Compute losses 90 | def __init__(self, model, autobalance=False): 91 | super(ComputeLoss, self).__init__() 92 | device = next(model.parameters()).device # get model device 93 | h = model.hyp # hyperparameters 94 | 95 | # Define criteria 96 | BCEcls = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['cls_pw']], device=device)) 97 | BCEobj = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['obj_pw']], device=device)) 98 | 99 | # Class label smoothing https://arxiv.org/pdf/1902.04103.pdf eqn 3 100 | self.cp, self.cn = smooth_BCE(eps=h.get('label_smoothing', 0.0)) # positive, negative BCE targets 101 | 102 | # Focal loss 103 | g = h['fl_gamma'] # focal loss gamma 104 | if g > 0: 105 | BCEcls, BCEobj = FocalLoss(BCEcls, g), FocalLoss(BCEobj, g) 106 | 107 | det = model.module.model[-1] if is_parallel(model) else model.model[-1] # Detect() module 108 | self.balance = {3: [4.0, 1.0, 0.4]}.get(det.nl, [4.0, 1.0, 0.25, 0.06, .02]) # P3-P7 109 | self.ssi = list(det.stride).index(16) if autobalance else 0 # stride 16 index 110 | self.BCEcls, self.BCEobj, self.gr, self.hyp, self.autobalance = BCEcls, BCEobj, model.gr, h, autobalance 111 | for k in 'na', 'nc', 'nl', 'anchors': 112 | setattr(self, k, getattr(det, k)) 113 | 114 | def __call__(self, p, targets): # predictions, targets, model 115 | device = targets.device 116 | lcls, lbox, lobj = torch.zeros(1, device=device), torch.zeros(1, device=device), torch.zeros(1, device=device) 117 | tcls, tbox, indices, anchors = self.build_targets(p, targets) # targets 118 | 119 | # Losses 120 | for i, pi in enumerate(p): # layer index, layer predictions 121 | b, a, gj, gi = indices[i] # image, anchor, gridy, gridx 122 | tobj = torch.zeros_like(pi[..., 0], device=device) # target obj 123 | 124 | n = b.shape[0] # number of targets 125 | if n: 126 | ps = pi[b, a, gj, gi] # prediction subset corresponding to targets 127 | 128 | # Regression 129 | pxy = ps[:, :2].sigmoid() * 2. - 0.5 130 | pwh = (ps[:, 2:4].sigmoid() * 2) ** 2 * anchors[i] 131 | pbox = torch.cat((pxy, pwh), 1) # predicted box 132 | iou = bbox_iou(pbox.T, tbox[i], x1y1x2y2=False, CIoU=True) # iou(prediction, target) 133 | lbox += (1.0 - iou).mean() # iou loss 134 | 135 | # Objectness 136 | tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * iou.detach().clamp(0).type(tobj.dtype) # iou ratio 137 | 138 | # Classification 139 | if self.nc > 1: # cls loss (only if multiple classes) 140 | t = torch.full_like(ps[:, 5:], self.cn, device=device) # targets 141 | t[range(n), tcls[i]] = self.cp 142 | lcls += self.BCEcls(ps[:, 5:], t) # BCE 143 | 144 | # Append targets to text file 145 | # with open('targets.txt', 'a') as file: 146 | # [file.write('%11.5g ' * 4 % tuple(x) + '\n') for x in torch.cat((txy[i], twh[i]), 1)] 147 | 148 | obji = self.BCEobj(pi[..., 4], tobj) 149 | lobj += obji * self.balance[i] # obj loss 150 | if self.autobalance: 151 | self.balance[i] = self.balance[i] * 0.9999 + 0.0001 / obji.detach().item() 152 | 153 | if self.autobalance: 154 | self.balance = [x / self.balance[self.ssi] for x in self.balance] 155 | lbox *= self.hyp['box'] 156 | lobj *= self.hyp['obj'] 157 | lcls *= self.hyp['cls'] 158 | bs = tobj.shape[0] # batch size 159 | 160 | loss = lbox + lobj + lcls 161 | return loss * bs, torch.cat((lbox, lobj, lcls, loss)).detach() 162 | 163 | def build_targets(self, p, targets): 164 | # Build targets for compute_loss(), input targets(image,class,x,y,w,h) 165 | na, nt = self.na, targets.shape[0] # number of anchors, targets 166 | tcls, tbox, indices, anch = [], [], [], [] 167 | gain = torch.ones(7, device=targets.device) # normalized to gridspace gain 168 | ai = torch.arange(na, device=targets.device).float().view(na, 1).repeat(1, nt) # same as .repeat_interleave(nt) 169 | targets = torch.cat((targets.repeat(na, 1, 1), ai[:, :, None]), 2) # append anchor indices 170 | 171 | g = 0.5 # bias 172 | off = torch.tensor([[0, 0], 173 | [1, 0], [0, 1], [-1, 0], [0, -1], # j,k,l,m 174 | # [1, 1], [1, -1], [-1, 1], [-1, -1], # jk,jm,lk,lm 175 | ], device=targets.device).float() * g # offsets 176 | 177 | for i in range(self.nl): 178 | anchors = self.anchors[i] 179 | gain[2:6] = torch.tensor(p[i].shape)[[3, 2, 3, 2]] # xyxy gain 180 | 181 | # Match targets to anchors 182 | t = targets * gain 183 | if nt: 184 | # Matches 185 | r = t[:, :, 4:6] / anchors[:, None] # wh ratio 186 | j = torch.max(r, 1. / r).max(2)[0] < self.hyp['anchor_t'] # compare 187 | # j = wh_iou(anchors, t[:, 4:6]) > model.hyp['iou_t'] # iou(3,n)=wh_iou(anchors(3,2), gwh(n,2)) 188 | t = t[j] # filter 189 | 190 | # Offsets 191 | gxy = t[:, 2:4] # grid xy 192 | gxi = gain[[2, 3]] - gxy # inverse 193 | j, k = ((gxy % 1. < g) & (gxy > 1.)).T 194 | l, m = ((gxi % 1. < g) & (gxi > 1.)).T 195 | j = torch.stack((torch.ones_like(j), j, k, l, m)) 196 | t = t.repeat((5, 1, 1))[j] 197 | offsets = (torch.zeros_like(gxy)[None] + off[:, None])[j] 198 | else: 199 | t = targets[0] 200 | offsets = 0 201 | 202 | # Define 203 | b, c = t[:, :2].long().T # image, class 204 | gxy = t[:, 2:4] # grid xy 205 | gwh = t[:, 4:6] # grid wh 206 | gij = (gxy - offsets).long() 207 | gi, gj = gij.T # grid xy indices 208 | 209 | # Append 210 | a = t[:, 6].long() # anchor indices 211 | indices.append((b, a, gj.clamp_(0, gain[3] - 1), gi.clamp_(0, gain[2] - 1))) # image, anchor, grid indices 212 | tbox.append(torch.cat((gxy - gij, gwh), 1)) # box 213 | anch.append(anchors[a]) # anchors 214 | tcls.append(c) # class 215 | 216 | return tcls, tbox, indices, anch 217 | -------------------------------------------------------------------------------- /utils/metrics.py: -------------------------------------------------------------------------------- 1 | # Model validation metrics 2 | 3 | from pathlib import Path 4 | 5 | import matplotlib.pyplot as plt 6 | import numpy as np 7 | import torch 8 | 9 | from . import general 10 | 11 | 12 | def fitness(x): 13 | # Model fitness as a weighted combination of metrics 14 | w = [0.0, 0.0, 0.1, 0.9] # weights for [P, R, mAP@0.5, mAP@0.5:0.95] 15 | return (x[:, :4] * w).sum(1) 16 | 17 | 18 | def ap_per_class(tp, conf, pred_cls, target_cls, plot=False, save_dir='.', names=()): 19 | """ Compute the average precision, given the recall and precision curves. 20 | Source: https://github.com/rafaelpadilla/Object-Detection-Metrics. 21 | # Arguments 22 | tp: True positives (nparray, nx1 or nx10). 23 | conf: Objectness value from 0-1 (nparray). 24 | pred_cls: Predicted object classes (nparray). 25 | target_cls: True object classes (nparray). 26 | plot: Plot precision-recall curve at mAP@0.5 27 | save_dir: Plot save directory 28 | # Returns 29 | The average precision as computed in py-faster-rcnn. 30 | """ 31 | 32 | # Sort by objectness 33 | i = np.argsort(-conf) 34 | tp, conf, pred_cls = tp[i], conf[i], pred_cls[i] 35 | 36 | # Find unique classes 37 | unique_classes = np.unique(target_cls) 38 | nc = unique_classes.shape[0] # number of classes, number of detections 39 | 40 | # Create Precision-Recall curve and compute AP for each class 41 | px, py = np.linspace(0, 1, 1000), [] # for plotting 42 | ap, p, r = np.zeros((nc, tp.shape[1])), np.zeros((nc, 1000)), np.zeros((nc, 1000)) 43 | for ci, c in enumerate(unique_classes): 44 | i = pred_cls == c 45 | n_l = (target_cls == c).sum() # number of labels 46 | n_p = i.sum() # number of predictions 47 | 48 | if n_p == 0 or n_l == 0: 49 | continue 50 | else: 51 | # Accumulate FPs and TPs 52 | fpc = (1 - tp[i]).cumsum(0) 53 | tpc = tp[i].cumsum(0) 54 | 55 | # Recall 56 | recall = tpc / (n_l + 1e-16) # recall curve 57 | r[ci] = np.interp(-px, -conf[i], recall[:, 0], left=0) # negative x, xp because xp decreases 58 | 59 | # Precision 60 | precision = tpc / (tpc + fpc) # precision curve 61 | p[ci] = np.interp(-px, -conf[i], precision[:, 0], left=1) # p at pr_score 62 | 63 | # AP from recall-precision curve 64 | for j in range(tp.shape[1]): 65 | ap[ci, j], mpre, mrec = compute_ap(recall[:, j], precision[:, j]) 66 | if plot and j == 0: 67 | py.append(np.interp(px, mrec, mpre)) # precision at mAP@0.5 68 | 69 | # Compute F1 (harmonic mean of precision and recall) 70 | f1 = 2 * p * r / (p + r + 1e-16) 71 | if plot: 72 | plot_pr_curve(px, py, ap, Path(save_dir) / 'PR_curve.png', names) 73 | plot_mc_curve(px, f1, Path(save_dir) / 'F1_curve.png', names, ylabel='F1') 74 | plot_mc_curve(px, p, Path(save_dir) / 'P_curve.png', names, ylabel='Precision') 75 | plot_mc_curve(px, r, Path(save_dir) / 'R_curve.png', names, ylabel='Recall') 76 | 77 | i = f1.mean(0).argmax() # max F1 index 78 | # print("p[:, i]", p[:, i]) 79 | # print("p[:, i]", p[:, i].shape()) 80 | return p[:, i], r[:, i], ap, f1[:, i], unique_classes.astype('int32') 81 | 82 | 83 | def compute_ap(recall, precision): 84 | """ Compute the average precision, given the recall and precision curves 85 | # Arguments 86 | recall: The recall curve (list) 87 | precision: The precision curve (list) 88 | # Returns 89 | Average precision, precision curve, recall curve 90 | """ 91 | 92 | # Append sentinel values to beginning and end 93 | mrec = np.concatenate(([0.], recall, [recall[-1] + 0.01])) 94 | mpre = np.concatenate(([1.], precision, [0.])) 95 | 96 | # Compute the precision envelope 97 | mpre = np.flip(np.maximum.accumulate(np.flip(mpre))) 98 | 99 | # Integrate area under curve 100 | method = 'interp' # methods: 'continuous', 'interp' 101 | if method == 'interp': 102 | x = np.linspace(0, 1, 101) # 101-point interp (COCO) 103 | ap = np.trapz(np.interp(x, mrec, mpre), x) # integrate 104 | else: # 'continuous' 105 | i = np.where(mrec[1:] != mrec[:-1])[0] # points where x axis (recall) changes 106 | ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1]) # area under curve 107 | 108 | return ap, mpre, mrec 109 | 110 | 111 | class ConfusionMatrix: 112 | # Updated version of https://github.com/kaanakan/object_detection_confusion_matrix 113 | def __init__(self, nc, conf=0.25, iou_thres=0.45): 114 | self.matrix = np.zeros((nc + 1, nc + 1)) 115 | self.nc = nc # number of classes 116 | self.conf = conf 117 | self.iou_thres = iou_thres 118 | 119 | def process_batch(self, detections, labels): 120 | """ 121 | Return intersection-over-union (Jaccard index) of boxes. 122 | Both sets of boxes are expected to be in (x1, y1, x2, y2) format. 123 | Arguments: 124 | detections (Array[N, 6]), x1, y1, x2, y2, conf, class 125 | labels (Array[M, 5]), class, x1, y1, x2, y2 126 | Returns: 127 | None, updates confusion matrix accordingly 128 | """ 129 | detections = detections[detections[:, 4] > self.conf] 130 | gt_classes = labels[:, 0].int() 131 | detection_classes = detections[:, 5].int() 132 | iou = general.box_iou(labels[:, 1:], detections[:, :4]) 133 | 134 | x = torch.where(iou > self.iou_thres) 135 | if x[0].shape[0]: 136 | matches = torch.cat((torch.stack(x, 1), iou[x[0], x[1]][:, None]), 1).cpu().numpy() 137 | if x[0].shape[0] > 1: 138 | matches = matches[matches[:, 2].argsort()[::-1]] 139 | matches = matches[np.unique(matches[:, 1], return_index=True)[1]] 140 | matches = matches[matches[:, 2].argsort()[::-1]] 141 | matches = matches[np.unique(matches[:, 0], return_index=True)[1]] 142 | else: 143 | matches = np.zeros((0, 3)) 144 | 145 | n = matches.shape[0] > 0 146 | m0, m1, _ = matches.transpose().astype(np.int16) 147 | for i, gc in enumerate(gt_classes): 148 | j = m0 == i 149 | if n and sum(j) == 1: 150 | self.matrix[detection_classes[m1[j]], gc] += 1 # correct 151 | else: 152 | self.matrix[self.nc, gc] += 1 # background FP 153 | 154 | if n: 155 | for i, dc in enumerate(detection_classes): 156 | if not any(m1 == i): 157 | self.matrix[dc, self.nc] += 1 # background FN 158 | 159 | def matrix(self): 160 | return self.matrix 161 | 162 | def plot(self, save_dir='', names=()): 163 | try: 164 | import seaborn as sn 165 | 166 | array = self.matrix / (self.matrix.sum(0).reshape(1, self.nc + 1) + 1E-6) # normalize 167 | array[array < 0.005] = np.nan # don't annotate (would appear as 0.00) 168 | 169 | fig = plt.figure(figsize=(12, 9), tight_layout=True) 170 | sn.set(font_scale=1.0 if self.nc < 50 else 0.8) # for label size 171 | labels = (0 < len(names) < 99) and len(names) == self.nc # apply names to ticklabels 172 | sn.heatmap(array, annot=self.nc < 30, annot_kws={"size": 8}, cmap='Blues', fmt='.2f', square=True, 173 | xticklabels=names + ['background FP'] if labels else "auto", 174 | yticklabels=names + ['background FN'] if labels else "auto").set_facecolor((1, 1, 1)) 175 | fig.axes[0].set_xlabel('True') 176 | fig.axes[0].set_ylabel('Predicted') 177 | fig.savefig(Path(save_dir) / 'confusion_matrix.png', dpi=250) 178 | except Exception as e: 179 | pass 180 | 181 | def print(self): 182 | for i in range(self.nc + 1): 183 | print(' '.join(map(str, self.matrix[i]))) 184 | 185 | 186 | # Plots ---------------------------------------------------------------------------------------------------------------- 187 | 188 | def plot_pr_curve(px, py, ap, save_dir='pr_curve.png', names=()): 189 | # Precision-recall curve 190 | fig, ax = plt.subplots(1, 1, figsize=(9, 6), tight_layout=True) 191 | py = np.stack(py, axis=1) 192 | 193 | if 0 < len(names) < 21: # display per-class legend if < 21 classes 194 | for i, y in enumerate(py.T): 195 | ax.plot(px, y, linewidth=1, label=f'{names[i]} {ap[i, 0]:.3f}') # plot(recall, precision) 196 | else: 197 | ax.plot(px, py, linewidth=1, color='grey') # plot(recall, precision) 198 | 199 | ax.plot(px, py.mean(1), linewidth=3, color='blue', label='all classes %.3f mAP@0.5' % ap[:, 0].mean()) 200 | ax.set_xlabel('Recall') 201 | ax.set_ylabel('Precision') 202 | ax.set_xlim(0, 1) 203 | ax.set_ylim(0, 1) 204 | plt.legend(bbox_to_anchor=(1.04, 1), loc="upper left") 205 | fig.savefig(Path(save_dir), dpi=250) 206 | 207 | 208 | def plot_mc_curve(px, py, save_dir='mc_curve.png', names=(), xlabel='Confidence', ylabel='Metric'): 209 | # Metric-confidence curve 210 | fig, ax = plt.subplots(1, 1, figsize=(9, 6), tight_layout=True) 211 | 212 | if 0 < len(names) < 21: # display per-class legend if < 21 classes 213 | for i, y in enumerate(py): 214 | ax.plot(px, y, linewidth=1, label=f'{names[i]}') # plot(confidence, metric) 215 | else: 216 | ax.plot(px, py.T, linewidth=1, color='grey') # plot(confidence, metric) 217 | 218 | y = py.mean(0) 219 | ax.plot(px, y, linewidth=3, color='blue', label=f'all classes {y.max():.2f} at {px[y.argmax()]:.3f}') 220 | ax.set_xlabel(xlabel) 221 | ax.set_ylabel(ylabel) 222 | ax.set_xlim(0, 1) 223 | ax.set_ylim(0, 1) 224 | plt.legend(bbox_to_anchor=(1.04, 1), loc="upper left") 225 | fig.savefig(Path(save_dir), dpi=250) 226 | -------------------------------------------------------------------------------- /utils/torch_utils.py: -------------------------------------------------------------------------------- 1 | # YOLOv5 PyTorch utils 2 | 3 | import datetime 4 | import logging 5 | import math 6 | import os 7 | import platform 8 | import subprocess 9 | import time 10 | from contextlib import contextmanager 11 | from copy import deepcopy 12 | from pathlib import Path 13 | 14 | import torch 15 | import torch.backends.cudnn as cudnn 16 | import torch.nn as nn 17 | import torch.nn.functional as F 18 | import torchvision 19 | 20 | try: 21 | import thop # for FLOPS computation 22 | except ImportError: 23 | thop = None 24 | logger = logging.getLogger(__name__) 25 | 26 | 27 | @contextmanager 28 | def torch_distributed_zero_first(local_rank: int): 29 | """ 30 | Decorator to make all processes in distributed training wait for each local_master to do something. 31 | """ 32 | if local_rank not in [-1, 0]: 33 | torch.distributed.barrier() 34 | yield 35 | if local_rank == 0: 36 | torch.distributed.barrier() 37 | 38 | 39 | def init_torch_seeds(seed=0): 40 | # Speed-reproducibility tradeoff https://pytorch.org/docs/stable/notes/randomness.html 41 | torch.manual_seed(seed) 42 | if seed == 0: # slower, more reproducible 43 | cudnn.benchmark, cudnn.deterministic = False, True 44 | else: # faster, less reproducible 45 | cudnn.benchmark, cudnn.deterministic = True, False 46 | 47 | 48 | def date_modified(path=__file__): 49 | # return human-readable file modification date, i.e. '2021-3-26' 50 | t = datetime.datetime.fromtimestamp(Path(path).stat().st_mtime) 51 | return f'{t.year}-{t.month}-{t.day}' 52 | 53 | 54 | def git_describe(path=Path(__file__).parent): # path must be a directory 55 | # return human-readable git description, i.e. v5.0-5-g3e25f1e https://git-scm.com/docs/git-describe 56 | s = f'git -C {path} describe --tags --long --always' 57 | try: 58 | return subprocess.check_output(s, shell=True, stderr=subprocess.STDOUT).decode()[:-1] 59 | except subprocess.CalledProcessError as e: 60 | return '' # not a git repository 61 | 62 | 63 | def select_device(device='', batch_size=None): 64 | # device = 'cpu' or '0' or '0,1,2,3' 65 | s = f'YOLOv5 🚀 {git_describe() or date_modified()} torch {torch.__version__} ' # string 66 | cpu = device.lower() == 'cpu' 67 | if cpu: 68 | os.environ['CUDA_VISIBLE_DEVICES'] = '-1' # force torch.cuda.is_available() = False 69 | elif device: # non-cpu device requested 70 | os.environ['CUDA_VISIBLE_DEVICES'] = device # set environment variable 71 | assert torch.cuda.is_available(), f'CUDA unavailable, invalid device {device} requested' # check availability 72 | 73 | cuda = not cpu and torch.cuda.is_available() 74 | if cuda: 75 | n = torch.cuda.device_count() 76 | if n > 1 and batch_size: # check that batch_size is compatible with device_count 77 | assert batch_size % n == 0, f'batch-size {batch_size} not multiple of GPU count {n}' 78 | space = ' ' * len(s) 79 | for i, d in enumerate(device.split(',') if device else range(n)): 80 | p = torch.cuda.get_device_properties(i) 81 | s += f"{'' if i == 0 else space}CUDA:{d} ({p.name}, {p.total_memory / 1024 ** 2}MB)\n" # bytes to MB 82 | else: 83 | s += 'CPU\n' 84 | 85 | logger.info(s.encode().decode('ascii', 'ignore') if platform.system() == 'Windows' else s) # emoji-safe 86 | return torch.device('cuda:0' if cuda else 'cpu') 87 | 88 | 89 | def time_synchronized(): 90 | # pytorch-accurate time 91 | if torch.cuda.is_available(): 92 | torch.cuda.synchronize() 93 | return time.time() 94 | 95 | 96 | def profile(x, ops, n=100, device=None): 97 | # profile a pytorch module or list of modules. Example usage: 98 | # x = torch.randn(16, 3, 640, 640) # input 99 | # m1 = lambda x: x * torch.sigmoid(x) 100 | # m2 = nn.SiLU() 101 | # profile(x, [m1, m2], n=100) # profile speed over 100 iterations 102 | 103 | device = device or torch.device('cuda:0' if torch.cuda.is_available() else 'cpu') 104 | x = x.to(device) 105 | x.requires_grad = True 106 | print(torch.__version__, device.type, torch.cuda.get_device_properties(0) if device.type == 'cuda' else '') 107 | print(f"\n{'Params':>12s}{'GFLOPS':>12s}{'forward (ms)':>16s}{'backward (ms)':>16s}{'input':>24s}{'output':>24s}") 108 | for m in ops if isinstance(ops, list) else [ops]: 109 | m = m.to(device) if hasattr(m, 'to') else m # device 110 | m = m.half() if hasattr(m, 'half') and isinstance(x, torch.Tensor) and x.dtype is torch.float16 else m # type 111 | dtf, dtb, t = 0., 0., [0., 0., 0.] # dt forward, backward 112 | try: 113 | flops = thop.profile(m, inputs=(x,), verbose=False)[0] / 1E9 * 2 # GFLOPS 114 | except: 115 | flops = 0 116 | 117 | for _ in range(n): 118 | t[0] = time_synchronized() 119 | y = m(x) 120 | t[1] = time_synchronized() 121 | try: 122 | _ = y.sum().backward() 123 | t[2] = time_synchronized() 124 | except: # no backward method 125 | t[2] = float('nan') 126 | dtf += (t[1] - t[0]) * 1000 / n # ms per op forward 127 | dtb += (t[2] - t[1]) * 1000 / n # ms per op backward 128 | 129 | s_in = tuple(x.shape) if isinstance(x, torch.Tensor) else 'list' 130 | s_out = tuple(y.shape) if isinstance(y, torch.Tensor) else 'list' 131 | p = sum(list(x.numel() for x in m.parameters())) if isinstance(m, nn.Module) else 0 # parameters 132 | print(f'{p:12}{flops:12.4g}{dtf:16.4g}{dtb:16.4g}{str(s_in):>24s}{str(s_out):>24s}') 133 | 134 | 135 | def is_parallel(model): 136 | return type(model) in (nn.parallel.DataParallel, nn.parallel.DistributedDataParallel) 137 | 138 | 139 | def intersect_dicts(da, db, exclude=()): 140 | # Dictionary intersection of matching keys and shapes, omitting 'exclude' keys, using da values 141 | return {k: v for k, v in da.items() if k in db and not any(x in k for x in exclude) and v.shape == db[k].shape} 142 | 143 | 144 | def initialize_weights(model): 145 | for m in model.modules(): 146 | t = type(m) 147 | if t is nn.Conv2d: 148 | pass # nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu') 149 | elif t is nn.BatchNorm2d: 150 | m.eps = 1e-3 151 | m.momentum = 0.03 152 | elif t in [nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6]: 153 | m.inplace = True 154 | 155 | 156 | def find_modules(model, mclass=nn.Conv2d): 157 | # Finds layer indices matching module class 'mclass' 158 | return [i for i, m in enumerate(model.module_list) if isinstance(m, mclass)] 159 | 160 | 161 | def sparsity(model): 162 | # Return global model sparsity 163 | a, b = 0., 0. 164 | for p in model.parameters(): 165 | a += p.numel() 166 | b += (p == 0).sum() 167 | return b / a 168 | 169 | 170 | def prune(model, amount=0.3): 171 | # Prune model to requested global sparsity 172 | import torch.nn.utils.prune as prune 173 | print('Pruning model... ', end='') 174 | for name, m in model.named_modules(): 175 | if isinstance(m, nn.Conv2d): 176 | prune.l1_unstructured(m, name='weight', amount=amount) # prune 177 | prune.remove(m, 'weight') # make permanent 178 | print(' %.3g global sparsity' % sparsity(model)) 179 | 180 | 181 | def fuse_conv_and_bn(conv, bn): 182 | # Fuse convolution and batchnorm layers https://tehnokv.com/posts/fusing-batchnorm-and-conv/ 183 | fusedconv = nn.Conv2d(conv.in_channels, 184 | conv.out_channels, 185 | kernel_size=conv.kernel_size, 186 | stride=conv.stride, 187 | padding=conv.padding, 188 | groups=conv.groups, 189 | bias=True).requires_grad_(False).to(conv.weight.device) 190 | 191 | # prepare filters 192 | w_conv = conv.weight.clone().view(conv.out_channels, -1) 193 | w_bn = torch.diag(bn.weight.div(torch.sqrt(bn.eps + bn.running_var))) 194 | fusedconv.weight.copy_(torch.mm(w_bn, w_conv).view(fusedconv.weight.shape)) 195 | 196 | # prepare spatial bias 197 | b_conv = torch.zeros(conv.weight.size(0), device=conv.weight.device) if conv.bias is None else conv.bias 198 | b_bn = bn.bias - bn.weight.mul(bn.running_mean).div(torch.sqrt(bn.running_var + bn.eps)) 199 | fusedconv.bias.copy_(torch.mm(w_bn, b_conv.reshape(-1, 1)).reshape(-1) + b_bn) 200 | 201 | return fusedconv 202 | 203 | 204 | def model_info(model, verbose=False, img_size=640): 205 | # Model information. img_size may be int or list, i.e. img_size=640 or img_size=[640, 320] 206 | n_p = sum(x.numel() for x in model.parameters()) # number parameters 207 | n_g = sum(x.numel() for x in model.parameters() if x.requires_grad) # number gradients 208 | if verbose: 209 | print('%5s %40s %9s %12s %20s %10s %10s' % ('layer', 'name', 'gradient', 'parameters', 'shape', 'mu', 'sigma')) 210 | for i, (name, p) in enumerate(model.named_parameters()): 211 | name = name.replace('module_list.', '') 212 | print('%5g %40s %9s %12g %20s %10.3g %10.3g' % 213 | (i, name, p.requires_grad, p.numel(), list(p.shape), p.mean(), p.std())) 214 | 215 | try: # FLOPS 216 | from thop import profile 217 | stride = max(int(model.stride.max()), 32) if hasattr(model, 'stride') else 32 218 | img = torch.zeros((1, model.yaml.get('ch', 3), stride, stride), device=next(model.parameters()).device) # input 219 | flops = profile(deepcopy(model), inputs=(img,), verbose=False)[0] / 1E9 * 2 # stride GFLOPS 220 | img_size = img_size if isinstance(img_size, list) else [img_size, img_size] # expand if int/float 221 | fs = ', %.1f GFLOPS' % (flops * img_size[0] / stride * img_size[1] / stride) # 640x640 GFLOPS 222 | except (ImportError, Exception): 223 | fs = '' 224 | 225 | logger.info(f"Model Summary: {len(list(model.modules()))} layers, {n_p} parameters, {n_g} gradients{fs}") 226 | 227 | 228 | def load_classifier(name='resnet101', n=2): 229 | # Loads a pretrained model reshaped to n-class output 230 | model = torchvision.models.__dict__[name](pretrained=True) 231 | 232 | # ResNet model properties 233 | # input_size = [3, 224, 224] 234 | # input_space = 'RGB' 235 | # input_range = [0, 1] 236 | # mean = [0.485, 0.456, 0.406] 237 | # std = [0.229, 0.224, 0.225] 238 | 239 | # Reshape output to n classes 240 | filters = model.fc.weight.shape[1] 241 | model.fc.bias = nn.Parameter(torch.zeros(n), requires_grad=True) 242 | model.fc.weight = nn.Parameter(torch.zeros(n, filters), requires_grad=True) 243 | model.fc.out_features = n 244 | return model 245 | 246 | 247 | def scale_img(img, ratio=1.0, same_shape=False, gs=32): # img(16,3,256,416) 248 | # scales img(bs,3,y,x) by ratio constrained to gs-multiple 249 | if ratio == 1.0: 250 | return img 251 | else: 252 | h, w = img.shape[2:] 253 | s = (int(h * ratio), int(w * ratio)) # new size 254 | img = F.interpolate(img, size=s, mode='bilinear', align_corners=False) # resize 255 | if not same_shape: # pad/crop img 256 | h, w = [math.ceil(x * ratio / gs) * gs for x in (h, w)] 257 | return F.pad(img, [0, w - s[1], 0, h - s[0]], value=0.447) # value = imagenet mean 258 | 259 | 260 | def copy_attr(a, b, include=(), exclude=()): 261 | # Copy attributes from b to a, options to only include [...] and to exclude [...] 262 | for k, v in b.__dict__.items(): 263 | if (len(include) and k not in include) or k.startswith('_') or k in exclude: 264 | continue 265 | else: 266 | setattr(a, k, v) 267 | 268 | 269 | class ModelEMA: 270 | """ Model Exponential Moving Average from https://github.com/rwightman/pytorch-image-models 271 | Keep a moving average of everything in the model state_dict (parameters and buffers). 272 | This is intended to allow functionality like 273 | https://www.tensorflow.org/api_docs/python/tf/train/ExponentialMovingAverage 274 | A smoothed version of the weights is necessary for some training schemes to perform well. 275 | This class is sensitive where it is initialized in the sequence of model init, 276 | GPU assignment and distributed training wrappers. 277 | """ 278 | 279 | def __init__(self, model, decay=0.9999, updates=0): 280 | # Create EMA 281 | self.ema = deepcopy(model.module if is_parallel(model) else model).eval() # FP32 EMA 282 | # if next(model.parameters()).device.type != 'cpu': 283 | # self.ema.half() # FP16 EMA 284 | self.updates = updates # number of EMA updates 285 | self.decay = lambda x: decay * (1 - math.exp(-x / 2000)) # decay exponential ramp (to help early epochs) 286 | for p in self.ema.parameters(): 287 | p.requires_grad_(False) 288 | 289 | def update(self, model): 290 | # Update EMA parameters 291 | with torch.no_grad(): 292 | self.updates += 1 293 | d = self.decay(self.updates) 294 | 295 | msd = model.module.state_dict() if is_parallel(model) else model.state_dict() # model state_dict 296 | for k, v in self.ema.state_dict().items(): 297 | if v.dtype.is_floating_point: 298 | v *= d 299 | v += (1. - d) * msd[k].detach() 300 | 301 | def update_attr(self, model, include=(), exclude=('process_group', 'reducer')): 302 | # Update EMA attributes 303 | copy_attr(self.ema, model, include, exclude) 304 | -------------------------------------------------------------------------------- /utils/wandb_logging/__init__.py: -------------------------------------------------------------------------------- 1 | 2 | -------------------------------------------------------------------------------- /utils/wandb_logging/__pycache__/__init__.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/utils/wandb_logging/__pycache__/__init__.cpython-37.pyc -------------------------------------------------------------------------------- /utils/wandb_logging/__pycache__/wandb_utils.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/utils/wandb_logging/__pycache__/wandb_utils.cpython-37.pyc -------------------------------------------------------------------------------- /utils/wandb_logging/log_dataset.py: -------------------------------------------------------------------------------- 1 | import argparse 2 | 3 | import yaml 4 | 5 | from wandb_utils import WandbLogger 6 | 7 | WANDB_ARTIFACT_PREFIX = 'wandb-artifact://' 8 | 9 | 10 | def create_dataset_artifact(opt): 11 | with open(opt.data) as f: 12 | data = yaml.safe_load(f) # data dict 13 | logger = WandbLogger(opt, '', None, data, job_type='Dataset Creation') 14 | 15 | 16 | if __name__ == '__main__': 17 | parser = argparse.ArgumentParser() 18 | parser.add_argument('--data', type=str, default='data/coco128.yaml', help='data.yaml path') 19 | parser.add_argument('--single-cls', action='store_true', help='train as single-class dataset') 20 | parser.add_argument('--project', type=str, default='YOLOv5', help='name of W&B Project') 21 | opt = parser.parse_args() 22 | opt.resume = False # Explicitly disallow resume check for dataset upload job 23 | 24 | create_dataset_artifact(opt) 25 | -------------------------------------------------------------------------------- /video/demo.gif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/video/demo.gif -------------------------------------------------------------------------------- /video/demo1.gif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/DocF/multispectral-object-detection/fb591c9b163177c0e950db08e213e24ddc912d41/video/demo1.gif --------------------------------------------------------------------------------