├── .gitignore
├── README.md
├── configs
    ├── cityscapes.yaml
    └── pascal.yaml
├── corrmatch.py
├── dataset
    ├── semi.py
    └── transform.py
├── evaluate.py
├── images
    └── cvpr_pipeline.png
├── model
    ├── backbone
    │   ├── resnet.py
    │   └── xception.py
    └── semseg
    │   ├── deeplabv3plus.py
    │   └── deeplabv3plus_vis.py
├── partitions
    ├── cityscapes
    │   ├── 1_16
    │   │   ├── labeled.txt
    │   │   └── unlabeled.txt
    │   ├── 1_30
    │   │   ├── labeled.txt
    │   │   └── unlabeled.txt
    │   ├── 1_4
    │   │   ├── labeled.txt
    │   │   └── unlabeled.txt
    │   ├── 1_8
    │   │   ├── labeled.txt
    │   │   └── unlabeled.txt
    │   └── val.txt
    ├── coco
    │   ├── 1_128
    │   │   ├── labeled.txt
    │   │   └── unlabeled.txt
    │   ├── 1_256
    │   │   ├── labeled.txt
    │   │   └── unlabeled.txt
    │   ├── 1_32
    │   │   ├── labeled.txt
    │   │   └── unlabeled.txt
    │   ├── 1_512
    │   │   ├── labeled.txt
    │   │   └── unlabeled.txt
    │   ├── 1_64
    │   │   ├── labeled.txt
    │   │   └── unlabeled.txt
    │   ├── train.txt
    │   └── val.txt
    └── pascal
    │   ├── 92
    │       ├── labeled.txt
    │       └── unlabeled.txt
    │   ├── 183
    │       ├── labeled.txt
    │       └── unlabeled.txt
    │   ├── 366
    │       ├── labeled.txt
    │       └── unlabeled.txt
    │   ├── 732
    │       ├── labeled.txt
    │       └── unlabeled.txt
    │   ├── 1464
    │       ├── labeled.txt
    │       └── unlabeled.txt
    │   ├── 1_16
    │       ├── labeled.txt
    │       └── unlabeled.txt
    │   ├── 1_4
    │       ├── labeled.txt
    │       └── unlabeled.txt
    │   ├── 1_8
    │       ├── labeled.txt
    │       └── unlabeled.txt
    │   ├── train_aug.txt
    │   └── val.txt
├── tools
    ├── train.sh
    ├── val.sh
    ├── vis.sh
    └── vis_mask.sh
├── util
    ├── dist_helper.py
    ├── mesh_helper.py
    ├── ohem.py
    ├── thresh_helper.py
    └── utils.py
├── vis.py
└── vis_mask.py


/.gitignore:
--------------------------------------------------------------------------------
1 | /exp/
2 | /pretrained/
3 | /.idea/


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
  1 | # CorrMatch (CVPR 2024)
  2 | 
  3 | This repository contains the official implementation of the following paper:
  4 | 
  5 | > **[CorrMatch: Label Propagation via Correlation Matching for Semi-Supervised Semantic Segmentation](https://arxiv.org/abs/2306.04300)**</br>
  6 | > [Boyuan Sun](https://bbbbchan.github.io), [Yuqi Yang](https://github.com/BBBBchan/CorrMatch), [Le Zhang](http://zhangleuestc.cn/), [Ming-Ming Cheng](https://mmcheng.net/cmm/),  [Qibin Hou](https://houqb.github.io/)</br>
  7 | 
  8 | 🔥 The **Jittor vsersion implementation** of CorrMatch is available at [Jittor Version](https://github.com/BBBBchan/CorrMatch-Jittor) !!!
  9 | 
 10 | 🔥 Our paper is accepted by IEEE Computer Vision and Pattern Recognition (CVPR) 2024 !!!
 11 | ## Overview
 12 | CorrMatch provides a solution for mining more high-quality regions from the unlabeled images to leverage the unlabeled data more efficiently for consistency regularization.
 13 | ![avatar](./images/cvpr_pipeline.png "pipeline")
 14 | 
 15 | Previous approaches mostly employ complicated training strategies to leverage unlabeled data but overlook the role of correlation maps in modeling the relationships between pairs of locations. Thus, we introduce two label propagation strategies (Pixel Propagation and Region Propagation) with the help of correlation maps. 
 16 | 
 17 | For technical details, please refer to our full paper on [arXiv](https://arxiv.org/abs/2306.04300).
 18 | ## Getting Started
 19 | 
 20 | ### Installation
 21 | 
 22 | ```bash
 23 | git clone git@github.com:BBBBchan/CorrMatch.git
 24 | cd CorrMatch
 25 | conda create -n corrmatch python=3.9
 26 | conda activate corrmatch
 27 | conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia
 28 | pip install opencv-python tqdm einops pyyaml
 29 | ```
 30 | 
 31 | ### Pretrained Backbone:
 32 | [ResNet-101](https://drive.google.com/file/d/1Rx0legsMolCWENpfvE2jUScT3ogalMO8/view?usp=sharing)
 33 | ```bash
 34 | mkdir pretrained
 35 | ```
 36 | Please put the pretrained model under `pretrained` dictionary.
 37 | 
 38 | 
 39 | ### Dataset:
 40 | 
 41 | - Pascal VOC 2012: [JPEGImages](http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar) | [SegmentationClass](https://drive.google.com/file/d/1ikrDlsai5QSf2GiSUR3f8PZUzyTubcuF/view?usp=sharing)
 42 | - Cityscapes: [leftImg8bit](https://www.cityscapes-dataset.com/file-handling/?packageID=3) | [gtFine](https://drive.google.com/file/d/1E_27g9tuHm6baBqcA7jct_jqcGA89QPm/view?usp=sharing)
 43 | 
 44 | Please modify the dataset path in configuration files.*The groundtruth mask ids have already been pre-processed. You may use them directly.*
 45 | 
 46 | Your dataset path may look like:
 47 | ```
 48 | ├── [Your Pascal Path]
 49 |     ├── JPEGImages
 50 |     └── SegmentationClass
 51 |     
 52 | ├── [Your Cityscapes Path]
 53 |     ├── leftImg8bit
 54 |     └── gtFine
 55 | ```
 56 | 
 57 | ## Usage
 58 | 
 59 | ### Training CorrMatch
 60 | 
 61 | ```bash
 62 | sh tools/train.sh <num_gpu> <port>
 63 | ```
 64 | To run on different labeled data partitions or different datasets, please modify:
 65 | 
 66 | ``config``, ``labeled_id_path``, ``unlabeled_id_path``, and ``save_path`` in [train.sh](https://github.com/BBBBchan/CorrMatch/blob/main/tools/train.sh).
 67 | 
 68 | ### Evaluation
 69 | ```bash
 70 | sh tools/val.sh <num_gpu> <port>
 71 | ```
 72 | To evaluate your checkpoint, please modify ``checkpoint_path`` in [val.sh](https://github.com/BBBBchan/CorrMatch/blob/main/tools/val.sh).
 73 | 
 74 | ## Results
 75 | 
 76 | ### Pascal VOC 2012
 77 | 
 78 | Labeled images are sampled from the **original high-quality** training set. Results are obtained by DeepLabv3+ based on ResNet-101 with training size 321(513).
 79 | 
 80 | |        Method        | 1/16 (92) | 1/8 (183) |   1/4 (366)    | 1/2 (732) | Full (1464) |
 81 | |:--------------------:|:---------:|:---------:|:--------------:|:---------:|:-----------:|
 82 | |       SupOnly        |   45.1    |   55.3    |      64.8      |   69.7    |    73.5     |
 83 | |         ST++         |   65.2    |   71.0    |      74.6      |   77.3    |    79.1     |
 84 | |        PS-MT         |   65.8    |   69.6    |      76.6      |   78.4    |    80.0     |
 85 | |       UniMatch       |   75.2    |   77.2    |      78.8      |   79.9    |    81.2     |
 86 | | **CorrMatch (Ours)** | **76.4**  | **78.5**  |    **79.4**    | **80.6**  |  **81.8**   |
 87 | 
 88 | 
 89 | ### Cityscapes
 90 | 
 91 | Results are obtained by DeepLabv3+ based on ResNet-101.
 92 | 
 93 | |        Method        | 1/16 (186) | 1/8 (372) | 1/4   (744) | 1/2 (1488) |
 94 | |:--------------------:|:----------:|:---------:|:-----------:|:----------:|
 95 | |       SupOnly        |    65.7    |   72.5    |    74.4     |    77.8    |
 96 | |       UniMatch       |    76.6    |   77.9    |   79.2     |    79.5    |
 97 | | **CorrMatch (Ours)** |  **77.3**  | **78.5**  |  **79.4**   |  **80.4**  |
 98 | 
 99 | ## Citation
100 | 
101 | If you find our repo useful for your research, please consider citing our paper:
102 | 
103 | ```bibtex
104 | @article{sun2023corrmatch,
105 |   title={CorrMatch: Label Propagation via Correlation Matching for Semi-Supervised Semantic Segmentation},
106 |   author={Sun, Boyuan and Yang, Yuqi and Zhang, Le and Cheng, Ming-Ming and Hou, Qibin},
107 |   journal={IEEE Computer Vision and Pattern Recognition (CVPR)},
108 |   year={2024}
109 | }
110 | ```
111 | 
112 | ## License
113 | This code is licensed under the [Creative Commons Attribution-NonCommercial 4.0 International](https://creativecommons.org/licenses/by-nc/4.0/) for non-commercial use only.
114 | Please note that any commercial use of this code requires formal permission prior to use.
115 | 
116 | ## Contact
117 | 
118 | For technical questions, please contact `sbysbysby123[AT]gmail.com`.
119 | 
120 | For commercial licensing, please contact `cmm[AT]nankai.edu.cn` or `andrewhoux@gmail.com`.
121 | 
122 | ## Acknowledgement
123 | 
124 | We thank [UniMatch](https://github.com/LiheYoung/UniMatch), [CPS](https://github.com/charlesCXK/TorchSemiSeg), [CutMix-Seg](https://github.com/Britefury/cutmix-semisup-seg), [DeepLabv3Plus](https://github.com/YudeWang/deeplabv3plus-pytorch),  [U<sup>2</sup>PL](https://github.com/Haochen-Wang409/U2PL) and other excellent works (see this [project](https://github.com/BBBBchan/Awesome-Semi-Supervised-Semantic-Segmentation)) for their amazing projects!
125 | 


--------------------------------------------------------------------------------
/configs/cityscapes.yaml:
--------------------------------------------------------------------------------
 1 | # arguments for dataset
 2 | dataset: cityscapes
 3 | nclass: 19
 4 | crop_size: 801
 5 | data_root: your/cityscape/path
 6 | 
 7 | # arguments for training
 8 | epochs: 240
 9 | batch_size: 2
10 | lr: 0.005  # 4GPUs
11 | lr_multi: 1.0
12 | criterion:
13 |   name: OHEM
14 |   kwargs:
15 |     ignore_index: 255
16 |     thresh: 0.7
17 |     min_kept: 100000
18 | conf_thresh: 0.0
19 | 
20 | # arguments for model
21 | backbone: resnet101
22 | multi_grid: False
23 | replace_stride_with_dilation: [False, True, True]
24 | dilations: [12, 24, 36]
25 | 


--------------------------------------------------------------------------------
/configs/pascal.yaml:
--------------------------------------------------------------------------------
 1 | # arguments for dataset
 2 | dataset: pascal
 3 | nclass: 21
 4 | crop_size: 321
 5 | data_root: your/pascal/path
 6 | 
 7 | # arguments for training
 8 | epochs: 80
 9 | batch_size: 4
10 | lr: 0.001  # 4GPUs
11 | lr_multi: 10.0
12 | criterion:
13 |   name: CELoss
14 |   kwargs:
15 |     ignore_index: 255
16 | thresh_init: 0.85
17 | 
18 | # arguments for model
19 | backbone: resnet101
20 | pretrain: True
21 | multi_grid: False
22 | replace_stride_with_dilation: [False, True, True]
23 | #dilations: [6, 12, 18]
24 | dilations: [12, 24, 36]
25 | 


--------------------------------------------------------------------------------
/corrmatch.py:
--------------------------------------------------------------------------------
  1 | import argparse
  2 | import logging
  3 | import os
  4 | import pprint
  5 | 
  6 | from tqdm import tqdm
  7 | import numpy as np
  8 | import torch
  9 | from torch import nn
 10 | import torch.distributed as dist
 11 | import torch.nn.functional as F
 12 | import torch.backends.cudnn as cudnn
 13 | from torch.optim import SGD
 14 | from torch.utils.data import DataLoader
 15 | from PIL import Image
 16 | import matplotlib
 17 | import matplotlib.pyplot as plt
 18 | 
 19 | matplotlib.use('agg')
 20 | import yaml
 21 | 
 22 | from dataset.semi import SemiDataset
 23 | from model.semseg.deeplabv3plus import DeepLabV3Plus
 24 | from evaluate import evaluate
 25 | from util.ohem import ProbOhemCrossEntropy2d
 26 | from util.utils import count_params, init_log
 27 | from util.dist_helper import setup_distributed
 28 | from util.thresh_helper import ThreshController
 29 | from einops import rearrange
 30 | import random
 31 | 
 32 | os.environ["CUDA_VISIBLE_DEVICES"] = "0,1"
 33 | 
 34 | parser = argparse.ArgumentParser(description='Semi-Supervised Semantic Segmentation')
 35 | parser.add_argument('--config', type=str, required=True)
 36 | parser.add_argument('--labeled-id-path', type=str, required=True)
 37 | parser.add_argument('--unlabeled-id-path', type=str, required=True)
 38 | parser.add_argument('--save-path', type=str, required=True)
 39 | parser.add_argument('--local_rank', default=0, type=int)
 40 | parser.add_argument('--port', default=None, type=int)
 41 | 
 42 | 
 43 | def init_seeds(seed=0, cuda_deterministic=False):
 44 |     random.seed(seed)
 45 |     np.random.seed(seed)
 46 |     torch.manual_seed(seed)
 47 |     cudnn.enabled = True
 48 |     # Speed-reproducibility tradeoff https://pytorch.org/docs/stable/notes/randomness.html
 49 |     if cuda_deterministic:  # slower, more reproducible
 50 |         cudnn.deterministic = True
 51 |         cudnn.benchmark = False
 52 |     else:  # faster, less reproducible
 53 |         cudnn.deterministic = False
 54 |         cudnn.benchmark = True
 55 | 
 56 | 
 57 | def main():
 58 |     args = parser.parse_args()
 59 | 
 60 |     cfg = yaml.load(open(args.config, "r"), Loader=yaml.Loader)
 61 | 
 62 |     logger = init_log('global', logging.INFO)
 63 |     logger.propagate = 0
 64 | 
 65 |     rank, word_size = setup_distributed(port=args.port)
 66 | 
 67 |     if rank == 0:
 68 |         logger.info('{}\n'.format(pprint.pformat(cfg)))
 69 | 
 70 |     if rank == 0:
 71 |         os.makedirs(args.save_path, exist_ok=True)
 72 |     init_seeds(0, False)
 73 | 
 74 |     model = DeepLabV3Plus(cfg)
 75 |     # sam = sam_model_registry["vit_b"](checkpoint="sam/checkpoints/sam_vit_b.pth")
 76 |     # sam.cuda()
 77 | 
 78 |     if rank == 0:
 79 |         logger.info('Total params: {:.1f}M\n'.format(count_params(model)))
 80 | 
 81 |     optimizer = SGD([{'params': model.backbone.parameters(), 'lr': cfg['lr']},
 82 |                      {'params': [param for name, param in model.named_parameters() if 'backbone' not in name],
 83 |                       'lr': cfg['lr'] * cfg['lr_multi']}], lr=cfg['lr'], momentum=0.9, weight_decay=1e-4)
 84 | 
 85 |     local_rank = int(os.environ["LOCAL_RANK"])
 86 |     model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model)
 87 |     model.cuda()
 88 | 
 89 |     model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[local_rank],
 90 |                                                       output_device=local_rank, find_unused_parameters=False)
 91 | 
 92 |     if cfg['criterion']['name'] == 'CELoss':
 93 |         criterion_l = nn.CrossEntropyLoss(**cfg['criterion']['kwargs']).cuda(local_rank)
 94 |     elif cfg['criterion']['name'] == 'OHEM':
 95 |         criterion_l = ProbOhemCrossEntropy2d(**cfg['criterion']['kwargs']).cuda(local_rank)
 96 |     else:
 97 |         raise NotImplementedError('%s criterion is not implemented' % cfg['criterion']['name'])
 98 | 
 99 |     criterion_u = nn.CrossEntropyLoss(reduction='none').cuda(local_rank)
100 |     criterion_kl = nn.KLDivLoss(reduction='none').cuda(local_rank)
101 | 
102 |     trainset_u = SemiDataset(cfg['dataset'], cfg['data_root'], 'train_u',
103 |                              cfg['crop_size'], args.unlabeled_id_path)
104 |     trainset_l = SemiDataset(cfg['dataset'], cfg['data_root'], 'train_l',
105 |                              cfg['crop_size'], args.labeled_id_path, nsample=len(trainset_u.ids))
106 |     valset = SemiDataset(cfg['dataset'], cfg['data_root'], 'val')
107 | 
108 |     trainsampler_l = torch.utils.data.distributed.DistributedSampler(trainset_l)
109 |     trainloader_l = DataLoader(trainset_l, batch_size=cfg['batch_size'],
110 |                                pin_memory=False, num_workers=4, drop_last=True, sampler=trainsampler_l)
111 |     trainsampler_u = torch.utils.data.distributed.DistributedSampler(trainset_u)
112 |     trainloader_u = DataLoader(trainset_u, batch_size=cfg['batch_size'],
113 |                                pin_memory=False, num_workers=4, drop_last=True, sampler=trainsampler_u)
114 |     valsampler = torch.utils.data.distributed.DistributedSampler(valset)
115 |     valloader = DataLoader(valset, batch_size=1, pin_memory=True, num_workers=4,
116 |                            drop_last=False, sampler=valsampler)
117 | 
118 |     total_iters = len(trainloader_u) * cfg['epochs']
119 |     previous_best = 0.0
120 |     thresh_controller = ThreshController(nclass=21, momentum=0.999, thresh_init=cfg['thresh_init'])
121 | 
122 |     for epoch in range(cfg['epochs']):
123 |         if rank == 0:
124 |             logger.info('===========> Epoch: {:}, LR: {:.4f}, Previous best: {:.2f}'.format(
125 |                 epoch, optimizer.param_groups[0]['lr'], previous_best))
126 | 
127 |         total_loss, total_loss_x, total_loss_s, total_loss_w_fp = 0.0, 0.0, 0.0, 0.0
128 |         total_loss_kl = 0.0
129 |         total_loss_corr_ce, total_loss_corr_u = 0.0, 0.0
130 |         total_mask_ratio = 0.0
131 | 
132 |         trainloader_l.sampler.set_epoch(epoch)
133 |         trainloader_u.sampler.set_epoch(epoch)
134 | 
135 |         loader = zip(trainloader_l, trainloader_u, trainloader_u)
136 | 
137 |         if rank == 0:
138 |             tbar = tqdm(total=len(trainloader_l))
139 | 
140 |         for i, ((img_x, mask_x),
141 |                 (img_u_w, img_u_s1, _, ignore_mask, cutmix_box1, _),
142 |                 (img_u_w_mix, img_u_s1_mix, _, ignore_mask_mix, _, _)) in enumerate(loader):
143 | 
144 |             img_x, mask_x = img_x.cuda(), mask_x.cuda()
145 |             img_u_w = img_u_w.cuda()
146 |             img_u_s1, ignore_mask = img_u_s1.cuda(), ignore_mask.cuda()
147 |             cutmix_box1 = cutmix_box1.cuda()
148 |             img_u_w_mix = img_u_w_mix.cuda()
149 |             img_u_s1_mix = img_u_s1_mix.cuda()
150 |             ignore_mask_mix = ignore_mask_mix.cuda()
151 |             b, c, h, w = img_x.shape
152 | 
153 |             with torch.no_grad():
154 |                 model.eval()
155 |                 res_u_w_mix = model(img_u_w_mix, need_fp=False, use_corr=False)
156 |                 pred_u_w_mix = res_u_w_mix['out'].detach()
157 |                 conf_u_w_mix = pred_u_w_mix.softmax(dim=1).max(dim=1)[0]
158 |                 mask_u_w_mix = pred_u_w_mix.argmax(dim=1)
159 |                 img_u_s1[cutmix_box1.unsqueeze(1).expand(img_u_s1.shape) == 1] = \
160 |                     img_u_s1_mix[cutmix_box1.unsqueeze(1).expand(img_u_s1.shape) == 1]
161 | 
162 |             model.train()
163 | 
164 |             num_lb, num_ulb = img_x.shape[0], img_u_w.shape[0]
165 | 
166 |             res_w = model(torch.cat((img_x, img_u_w)), need_fp=True, use_corr=True)
167 | 
168 |             preds = res_w['out']
169 |             preds_fp = res_w['out_fp']
170 |             preds_corr = res_w['corr_out']
171 |             preds_corr_map = res_w['corr_map'].detach()
172 |             pred_x_corr, pred_u_w_corr = preds_corr.split([num_lb, num_ulb])
173 |             pred_u_w_corr_map = preds_corr_map[num_lb:]
174 |             pred_x, pred_u_w = preds.split([num_lb, num_ulb])
175 |             pred_u_w_fp = preds_fp[num_lb:]
176 | 
177 |             res_s = model(img_u_s1, need_fp=False, use_corr=True)
178 |             pred_u_s1 = res_s['out']
179 |             pred_u_s1_corr = res_s['corr_out']
180 | 
181 |             pred_u_w = pred_u_w.detach()
182 |             conf_u_w = pred_u_w.detach().softmax(dim=1).max(dim=1)[0]
183 |             mask_u_w = pred_u_w.detach().argmax(dim=1)
184 | 
185 |             mask_u_w_cutmixed1, conf_u_w_cutmixed1, ignore_mask_cutmixed1 = \
186 |                 mask_u_w.clone(), conf_u_w.clone(), ignore_mask.clone()
187 |             corr_map_u_w_cutmixed1 = pred_u_w_corr_map.clone()
188 |             b_sample, c_sample, _, _ = corr_map_u_w_cutmixed1.shape
189 | 
190 |             cutmix_box1_map = (cutmix_box1 == 1)
191 | 
192 |             mask_u_w_cutmixed1[cutmix_box1_map] = mask_u_w_mix[cutmix_box1_map]
193 |             mask_u_w_cutmixed1_copy = mask_u_w_cutmixed1.clone()
194 |             conf_u_w_cutmixed1[cutmix_box1_map] = conf_u_w_mix[cutmix_box1_map]
195 |             ignore_mask_cutmixed1[cutmix_box1_map] = ignore_mask_mix[cutmix_box1_map]
196 |             cutmix_box1_sample = rearrange(cutmix_box1_map, 'n h w -> n 1 h w')
197 |             ignore_mask_cutmixed1_sample = rearrange((ignore_mask_cutmixed1 != 255), 'n h w -> n 1 h w')
198 |             corr_map_u_w_cutmixed1 = (corr_map_u_w_cutmixed1 * ~cutmix_box1_sample * ignore_mask_cutmixed1_sample).bool()
199 | 
200 |             thresh_controller.thresh_update(pred_u_w.detach(), ignore_mask_cutmixed1, update_g=True)
201 |             thresh_global = thresh_controller.get_thresh_global()
202 | 
203 |             conf_fliter_u_w = ((conf_u_w_cutmixed1 >= thresh_global) & (ignore_mask_cutmixed1 != 255))
204 |             conf_fliter_u_w_without_cutmix = conf_fliter_u_w.clone()
205 |             conf_fliter_u_w_sample = rearrange(conf_fliter_u_w_without_cutmix, 'n h w -> n 1 h w')
206 | 
207 |             segments = (corr_map_u_w_cutmixed1 * conf_fliter_u_w_sample).bool()
208 | 
209 |             for img_idx in range(b_sample):
210 |                 for segment_idx in range(c_sample):
211 | 
212 |                     segment = segments[img_idx, segment_idx]
213 |                     segment_ori = corr_map_u_w_cutmixed1[img_idx, segment_idx]
214 |                     high_conf_ratio = torch.sum(segment)/torch.sum(segment_ori)
215 |                     if torch.sum(segment) == 0 or high_conf_ratio < thresh_global:
216 |                         continue
217 |                     unique_cls, count = torch.unique(mask_u_w_cutmixed1[img_idx][segment==1], return_counts=True)
218 | 
219 |                     if torch.max(count) / torch.sum(count) > thresh_global:
220 |                         top_class = unique_cls[torch.argmax(count)]
221 |                         mask_u_w_cutmixed1[img_idx][segment_ori==1] = top_class
222 |                         conf_fliter_u_w_without_cutmix[img_idx] = conf_fliter_u_w_without_cutmix[img_idx] | segment_ori
223 |             conf_fliter_u_w_without_cutmix = conf_fliter_u_w_without_cutmix | conf_fliter_u_w
224 | 
225 | 
226 |             loss_x = criterion_l(pred_x, mask_x)
227 |             loss_x_corr = criterion_l(pred_x_corr, mask_x)
228 | 
229 |             loss_u_s1 = criterion_u(pred_u_s1, mask_u_w_cutmixed1)
230 |             loss_u_s1 = loss_u_s1 * conf_fliter_u_w_without_cutmix
231 |             loss_u_s1 = torch.sum(loss_u_s1) / torch.sum(ignore_mask_cutmixed1 != 255).item()
232 | 
233 |             loss_u_corr_s1 = criterion_u(pred_u_s1_corr, mask_u_w_cutmixed1)
234 |             loss_u_corr_s1 = loss_u_corr_s1 * conf_fliter_u_w_without_cutmix
235 |             loss_u_corr_s1 = torch.sum(loss_u_corr_s1) / torch.sum(ignore_mask_cutmixed1 != 255).item()
236 |             loss_u_corr_s = loss_u_corr_s1
237 | 
238 |             loss_u_corr_w = criterion_u(pred_u_w_corr, mask_u_w)
239 |             loss_u_corr_w = loss_u_corr_w * ((conf_u_w >= thresh_global) & (ignore_mask != 255))
240 |             loss_u_corr_w = torch.sum(loss_u_corr_w) / torch.sum(ignore_mask != 255).item()
241 |             loss_u_corr = 0.5 * (loss_u_corr_s + loss_u_corr_w)
242 | 
243 |             softmax_pred_u_w = F.softmax(pred_u_w.detach(), dim=1)
244 |             logsoftmax_pred_u_s1 = F.log_softmax(pred_u_s1, dim=1)
245 | 
246 |             loss_u_kl_sa2wa = criterion_kl(logsoftmax_pred_u_s1, softmax_pred_u_w)
247 |             loss_u_kl_sa2wa = torch.sum(loss_u_kl_sa2wa, dim=1) * conf_fliter_u_w
248 |             loss_u_kl_sa2wa = torch.sum(loss_u_kl_sa2wa) / torch.sum(ignore_mask_cutmixed1 != 255).item()
249 |             loss_u_kl = loss_u_kl_sa2wa
250 | 
251 |             loss_u_w_fp = criterion_u(pred_u_w_fp, mask_u_w)
252 |             loss_u_w_fp = loss_u_w_fp * ((conf_u_w >= thresh_global) & (ignore_mask != 255))
253 |             loss_u_w_fp = torch.sum(loss_u_w_fp) / torch.sum(ignore_mask != 255).item()
254 | 
255 |             loss = ( 0.5 * loss_x + 0.5 * loss_x_corr + loss_u_s1 * 0.25 + loss_u_kl * 0.25 + loss_u_w_fp * 0.25 + 0.25 * loss_u_corr) / 2.0
256 | 
257 |             optimizer.zero_grad()
258 |             loss.backward()
259 |             optimizer.step()
260 | 
261 |             total_loss += loss.item()
262 |             total_loss_x += loss_x.item()
263 |             total_loss_s += loss_u_s1.item()
264 |             total_loss_kl += loss_u_kl.item()
265 |             total_loss_w_fp += loss_u_w_fp.item()
266 |             total_loss_corr_ce += loss_x_corr.item()
267 |             total_loss_corr_u += loss_u_corr.item()
268 |             total_mask_ratio += ((conf_u_w >= thresh_global) & (ignore_mask != 255)).sum().item() / \
269 |                                 (ignore_mask != 255).sum().item()
270 | 
271 |             iters = epoch * len(trainloader_u) + i
272 |             lr = cfg['lr'] * (1 - iters / total_iters) ** 0.9
273 |             optimizer.param_groups[0]["lr"] = lr
274 |             optimizer.param_groups[1]["lr"] = lr * cfg['lr_multi']
275 | 
276 |             if rank == 0:
277 |                 tbar.set_description(' Total loss: {:.3f}, Loss x: {:.3f}, loss_corr_ce: {:.3f} '
278 |                                      'Loss s: {:.3f}, Loss w_fp: {:.3f},  Mask: {:.3f}, loss_corr_u: {:.3f}'.format(
279 |                     total_loss / (i + 1), total_loss_x / (i + 1), total_loss_corr_ce / (i + 1), total_loss_s / (i + 1),
280 |                     total_loss_w_fp / (i + 1), total_mask_ratio / (i + 1), total_loss_corr_u / (i + 1)))
281 |                 tbar.update(1)
282 | 
283 |         if rank == 0:
284 |             tbar.close()
285 | 
286 |         if cfg['dataset'] == 'cityscapes':
287 |             eval_mode = 'center_crop' if epoch < cfg['epochs'] - 20 else 'sliding_window'
288 |         else:
289 |             eval_mode = 'original'
290 |         torch.cuda.empty_cache()
291 |         res_val = evaluate(model, valloader, eval_mode, cfg)
292 |         mIOU = res_val['mIOU']
293 |         class_IOU = res_val['iou_class']
294 |         torch.distributed.barrier()
295 | 
296 |         if rank == 0:
297 |             logger.info('***** Evaluation {} ***** >>>> meanIOU: {:.4f} \n'.format(eval_mode, mIOU))
298 |             logger.info('***** ClassIOU ***** >>>> \n{}\n'.format(class_IOU))
299 | 
300 |         if mIOU > previous_best and rank == 0:
301 |             if previous_best != 0:
302 |                 os.remove(os.path.join(args.save_path, '%s_%.3f.pth' % (cfg['backbone'], previous_best)))
303 |             previous_best = mIOU
304 |             torch.save(model.module.state_dict(), os.path.join(args.save_path, '%s_%.3f.pth' % (cfg['backbone'], mIOU)))
305 |         torch.distributed.barrier()
306 |         torch.cuda.empty_cache()
307 | 
308 | 
309 | if __name__ == '__main__':
310 |     main()
311 | 


--------------------------------------------------------------------------------
/dataset/semi.py:
--------------------------------------------------------------------------------
 1 | from dataset.transform import *
 2 | 
 3 | from copy import deepcopy
 4 | import math
 5 | import numpy as np
 6 | import os
 7 | import random
 8 | 
 9 | from PIL import Image
10 | import torch
11 | from torch.utils.data import Dataset
12 | from torchvision import transforms
13 | 
14 | 
15 | class SemiDataset(Dataset):
16 |     def __init__(self, name, root, mode, size=None, id_path=None, nsample=None):
17 |         self.name = name
18 |         self.root = root
19 |         self.mode = mode
20 |         self.size = size
21 | 
22 |         if mode == 'train_l' or mode == 'train_u':
23 |             with open(id_path, 'r') as f:
24 |                 self.ids = f.read().splitlines()
25 |             if mode == 'train_l' and nsample is not None:
26 |                 self.ids *= math.ceil(nsample / len(self.ids))
27 |                 random.shuffle(self.ids)
28 |                 self.ids = self.ids[:nsample]
29 |         else:
30 |             with open('partitions/%s/val.txt' % name, 'r') as f:
31 |                 self.ids = f.read().splitlines()
32 | 
33 |     def __getitem__(self, item):
34 |         id = self.ids[item]
35 |         img = Image.open(os.path.join(self.root, id.split(' ')[0])).convert('RGB')
36 |         mask = Image.fromarray(np.array(Image.open(os.path.join(self.root, id.split(' ')[1]))))
37 | 
38 |         if self.mode == 'val':
39 |             img_ori = np.array(img)
40 |             img, mask = normalize(img, mask)
41 |             return img, mask, id, img_ori
42 | 
43 |         img, mask = resize(img, mask, (0.5, 2.0))
44 |         ignore_value = 254 if self.mode == 'train_u' else 255
45 |         img, mask = crop(img, mask, self.size, ignore_value)
46 |         img, mask = hflip(img, mask, p=0.5)
47 | 
48 |         if self.mode == 'train_l':
49 |             return normalize(img, mask)
50 | 
51 |         img_w, img_s1, img_ori = deepcopy(img), deepcopy(img), deepcopy(img)
52 |         img_ori = np.array(img_ori)
53 | 
54 |         if random.random() < 0.8:
55 |             img_s1 = transforms.ColorJitter(0.5, 0.5, 0.5, 0.25)(img_s1)
56 |         img_s1 = transforms.RandomGrayscale(p=0.2)(img_s1)
57 |         img_s1 = blur(img_s1, p=0.5)
58 |         cutmix_box1 = obtain_cutmix_box(img_s1.size[0], p=0.5)
59 | 
60 | 
61 |         ignore_mask = Image.fromarray(np.zeros((mask.size[1], mask.size[0])))
62 | 
63 |         img_s1, ignore_mask = normalize(img_s1, ignore_mask)
64 | 
65 |         mask = torch.from_numpy(np.array(mask)).long()
66 |         ignore_mask[mask == 254] = 255
67 | 
68 |         return normalize(img_w), img_s1, img_ori, ignore_mask, cutmix_box1, id
69 | 
70 |     def __len__(self):
71 |         return len(self.ids)
72 | 


--------------------------------------------------------------------------------
/dataset/transform.py:
--------------------------------------------------------------------------------
  1 | import random
  2 | import math
  3 | 
  4 | import numpy as np
  5 | from PIL import Image, ImageOps, ImageFilter, ImageEnhance
  6 | import torch
  7 | from torchvision import transforms
  8 | 
  9 | 
 10 | def crop(img, mask, size, ignore_value=255):
 11 |     w, h = img.size
 12 |     padw = size - w if w < size else 0
 13 |     padh = size - h if h < size else 0
 14 |     img = ImageOps.expand(img, border=(0, 0, padw, padh), fill=0)
 15 |     mask = ImageOps.expand(mask, border=(0, 0, padw, padh), fill=ignore_value)
 16 | 
 17 |     w, h = img.size
 18 |     x = random.randint(0, w - size)
 19 |     y = random.randint(0, h - size)
 20 |     img = img.crop((x, y, x + size, y + size))
 21 |     mask = mask.crop((x, y, x + size, y + size))
 22 | 
 23 |     return img, mask
 24 | 
 25 | 
 26 | def hflip(img, mask, p=0.5):
 27 |     if random.random() < p:
 28 |         img = img.transpose(Image.FLIP_LEFT_RIGHT)
 29 |         mask = mask.transpose(Image.FLIP_LEFT_RIGHT)
 30 |     return img, mask
 31 | 
 32 | 
 33 | def normalize(img, mask=None):
 34 |     img = transforms.Compose([
 35 |         transforms.ToTensor(),
 36 |         transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
 37 |     ])(img)
 38 |     if mask is not None:
 39 |         mask = torch.from_numpy(np.array(mask)).long()
 40 |         return img, mask
 41 |     return img
 42 | 
 43 | 
 44 | def resize_certain(img, ratio_range):
 45 |     w, h = img.size
 46 |     ow = int(w * ratio_range + 0.5)
 47 |     oh = int(h * ratio_range + 0.5)
 48 |     img = img.resize((ow, oh), Image.BILINEAR)
 49 |     return img
 50 | 
 51 | def resize(img, mask, ratio_range):
 52 |     w, h = img.size
 53 |     long_side = random.randint(int(max(h, w) * ratio_range[0]), int(max(h, w) * ratio_range[1]))
 54 | 
 55 |     if h > w:
 56 |         oh = long_side
 57 |         ow = int(1.0 * w * long_side / h + 0.5)
 58 |     else:
 59 |         ow = long_side
 60 |         oh = int(1.0 * h * long_side / w + 0.5)
 61 | 
 62 |     img = img.resize((ow, oh), Image.BILINEAR)
 63 |     mask = mask.resize((ow, oh), Image.NEAREST)
 64 |     return img, mask
 65 | 
 66 | 
 67 | def blur(img, p=0.5):
 68 |     if random.random() < p:
 69 |         sigma = np.random.uniform(0.1, 2.0)
 70 |         img = img.filter(ImageFilter.GaussianBlur(radius=sigma))
 71 |     return img
 72 | 
 73 | 
 74 | def obtain_cutmix_box(img_size, p=0.5, size_min=0.02, size_max=0.4, ratio_1=0.3, ratio_2=1/0.3):
 75 |     mask = torch.zeros(img_size, img_size)
 76 |     if random.random() > p:
 77 |         return mask
 78 | 
 79 |     size = np.random.uniform(size_min, size_max) * img_size * img_size
 80 |     while True:
 81 |         ratio = np.random.uniform(ratio_1, ratio_2)
 82 |         cutmix_w = int(np.sqrt(size / ratio))
 83 |         cutmix_h = int(np.sqrt(size * ratio))
 84 |         x = np.random.randint(0, img_size)
 85 |         y = np.random.randint(0, img_size)
 86 | 
 87 |         if x + cutmix_w <= img_size and y + cutmix_h <= img_size:
 88 |             break
 89 | 
 90 |     mask[y:y + cutmix_h, x:x + cutmix_w] = 1
 91 | 
 92 |     return mask
 93 | 
 94 | def img_aug_autocontrast(img, scale=None):
 95 |     return ImageOps.autocontrast(img)
 96 | 
 97 | 
 98 | def img_aug_equalize(img, scale=None):
 99 |     return ImageOps.equalize(img)
100 | 
101 | 
102 | def img_aug_invert(img, scale=None):
103 |     return ImageOps.invert(img)
104 | 
105 | 
106 | def img_aug_identity(img, scale=None):
107 |     return img
108 | 
109 | 
110 | def img_aug_blur(img, scale=[0.1, 2.0]):
111 |     assert scale[0] < scale[1]
112 |     sigma = np.random.uniform(scale[0], scale[1])
113 |     return img.filter(ImageFilter.GaussianBlur(radius=sigma))
114 | 
115 | 
116 | def img_aug_contrast(img, scale=[0.05, 0.95], p=0.2):
117 |     if random.random() < p:
118 |         min_v, max_v = min(scale), max(scale)
119 |         v = float(max_v - min_v) * random.random()
120 |         v = max_v - v
121 |         return ImageEnhance.Contrast(img).enhance(v)
122 |     else:
123 |         return img
124 | 
125 | 
126 | def img_aug_brightness(img, scale=[0.05, 0.95]):
127 |     min_v, max_v = min(scale), max(scale)
128 |     v = float(max_v - min_v) * random.random()
129 |     v = max_v - v
130 |     # print(f"final:{v}")
131 |     return ImageEnhance.Brightness(img).enhance(v)
132 | 
133 | 
134 | def img_aug_color(img, scale=[0.05, 0.95]):
135 |     min_v, max_v = min(scale), max(scale)
136 |     v = float(max_v - min_v) * random.random()
137 |     v = max_v - v
138 |     # print(f"final:{v}")
139 |     return ImageEnhance.Color(img).enhance(v)
140 | 
141 | 
142 | def img_aug_sharpness(img, scale=[0.05, 0.95]):
143 |     min_v, max_v = min(scale), max(scale)
144 |     v = float(max_v - min_v) * random.random()
145 |     v = max_v - v
146 |     # print(f"final:{v}")
147 |     return ImageEnhance.Sharpness(img).enhance(v)
148 | 
149 | 
150 | def img_aug_hue(img, scale=[0, 0.5]):
151 |     min_v, max_v = min(scale), max(scale)
152 |     v = float(max_v - min_v) * random.random()
153 |     v += min_v
154 |     if np.random.random() < 0.5:
155 |         hue_factor = -v
156 |     else:
157 |         hue_factor = v
158 |     # print(f"Final-V:{hue_factor}")
159 |     input_mode = img.mode
160 |     if input_mode in {"L", "1", "I", "F"}:
161 |         return img
162 |     h, s, v = img.convert("HSV").split()
163 |     np_h = np.array(h, dtype=np.uint8)
164 |     # uint8 addition take cares of rotation across boundaries
165 |     with np.errstate(over="ignore"):
166 |         np_h += np.uint8(hue_factor * 255)
167 |     h = Image.fromarray(np_h, "L")
168 |     img = Image.merge("HSV", (h, s, v)).convert(input_mode)
169 |     return img
170 | 
171 | 
172 | def img_aug_posterize(img, scale=[4, 8]):
173 |     min_v, max_v = min(scale), max(scale)
174 |     v = float(max_v - min_v) * random.random()
175 |     # print(min_v, max_v, v)
176 |     v = int(np.ceil(v))
177 |     v = max(1, v)
178 |     v = max_v - v
179 |     # print(f"final:{v}")
180 |     return ImageOps.posterize(img, v)
181 | 
182 | 
183 | def img_aug_solarize(img, scale=[1, 256]):
184 |     min_v, max_v = min(scale), max(scale)
185 |     v = float(max_v - min_v) * random.random()
186 |     # print(min_v, max_v, v)
187 |     v = int(np.ceil(v))
188 |     v = max(1, v)
189 |     v = max_v - v
190 |     # print(f"final:{v}")
191 |     return ImageOps.solarize(img, v)
192 | 
193 | 
194 | def get_augment_list():
195 |     l = [
196 |         (img_aug_identity, None),
197 |         (img_aug_autocontrast, None),
198 |         (img_aug_equalize, None),
199 |         (img_aug_blur, [0.1, 2.0]),
200 |         (img_aug_contrast, [0.05, 0.95]),
201 |         (img_aug_brightness, [0.05, 0.95]),
202 |         (img_aug_color, [0.05, 0.95]),
203 |         (img_aug_sharpness, [0.05, 0.95]),
204 |         (img_aug_posterize, [4, 8]),
205 |         (img_aug_solarize, [1, 256]),
206 |         (img_aug_hue, [0, 0.5])
207 |     ]
208 |     return l
209 | 
210 | 
211 | class strong_img_aug:
212 |     def __init__(self, num_augs=4, flag_using_random_num=True):
213 |         self.n = num_augs
214 |         self.augment_list = get_augment_list()
215 |         self.flag_using_random_num = flag_using_random_num
216 | 
217 |     def __call__(self, img):
218 |         if self.flag_using_random_num:
219 |             max_num = np.random.randint(1, high=self.n + 1)
220 |         else:
221 |             max_num = self.n
222 |         ops = random.choices(self.augment_list, k=max_num)
223 |         for op, scales in ops:
224 |             img = op(img, scales)
225 |         return img


--------------------------------------------------------------------------------
/evaluate.py:
--------------------------------------------------------------------------------
  1 | import argparse
  2 | import os
  3 | import numpy as np
  4 | import torch
  5 | import torch.distributed as dist
  6 | from util.dist_helper import setup_distributed
  7 | from model.semseg.deeplabv3plus import DeepLabV3Plus
  8 | 
  9 | from torch.utils.data import DataLoader
 10 | import yaml
 11 | from dataset.semi import SemiDataset
 12 | from util.utils import AverageMeter, intersectionAndUnion
 13 | 
 14 | 
 15 | def evaluate(model, loader, mode, cfg):
 16 |     return_dict = {}
 17 |     model.eval()
 18 |     assert mode in ['original', 'center_crop', 'sliding_window']
 19 |     intersection_meter = AverageMeter()
 20 |     union_meter = AverageMeter()
 21 | 
 22 |     with torch.no_grad():
 23 |         for img, mask, ids, img_ori in loader:
 24 |             img = img.cuda()
 25 |             b, _, h, w = img.shape
 26 |             if mode == 'sliding_window':
 27 |                 grid = cfg['crop_size']
 28 |                 final = torch.zeros(b, 19, h, w).cuda()
 29 |                 row = 0
 30 |                 while row < h:
 31 |                     col = 0
 32 |                     while col < w:
 33 |                         res = model(img[:, :, row: min(h, row + grid), col: min(w, col + grid)])
 34 |                         pred = res['out']
 35 |                         final[:, :, row: min(h, row + grid), col: min(w, col + grid)] += pred.softmax(dim=1)
 36 |                         col += int(grid * 2 / 3)
 37 |                     row += int(grid * 2 / 3)
 38 | 
 39 |                 pred = final.argmax(dim=1)
 40 | 
 41 |             else:
 42 |                 if mode == 'center_crop':
 43 |                     h, w = img.shape[-2:]
 44 |                     start_h, start_w = (h - cfg['crop_size']) // 2, (w - cfg['crop_size']) // 2
 45 |                     img = img[:, :, start_h:start_h + cfg['crop_size'], start_w:start_w + cfg['crop_size']]
 46 |                     mask = mask[:, start_h:start_h + cfg['crop_size'], start_w:start_w + cfg['crop_size']]
 47 | 
 48 |                 res = model(img)
 49 |                 pred = res['out'].argmax(dim=1)
 50 | 
 51 |             intersection, union, target = \
 52 |                 intersectionAndUnion(pred.cpu().numpy(), mask.numpy(), cfg['nclass'], 255)
 53 | 
 54 |             reduced_intersection = torch.from_numpy(intersection).cuda()
 55 |             reduced_union = torch.from_numpy(union).cuda()
 56 |             reduced_target = torch.from_numpy(target).cuda()
 57 | 
 58 |             dist.all_reduce(reduced_intersection)
 59 |             dist.all_reduce(reduced_union)
 60 |             dist.all_reduce(reduced_target)
 61 | 
 62 |             intersection_meter.update(reduced_intersection.cpu().numpy())
 63 |             union_meter.update(reduced_union.cpu().numpy())
 64 | 
 65 |     iou_class = intersection_meter.sum / (union_meter.sum + 1e-10)
 66 |     mIOU = np.mean(iou_class) * 100.0
 67 |     return_dict['iou_class'] = iou_class
 68 |     return_dict['mIOU'] = mIOU
 69 | 
 70 |     return return_dict
 71 | 
 72 | 
 73 | def main():
 74 |     parser = argparse.ArgumentParser(description='Semi-Supervised Semantic Segmentation')
 75 |     parser.add_argument('--config', type=str, required=True)
 76 |     parser.add_argument('--checkpoint_path', type=str, required=True)
 77 |     parser.add_argument('--local_rank', default=0, type=int)
 78 |     parser.add_argument('--port', default=None, type=int)
 79 |     args = parser.parse_args()
 80 |     setup_distributed(port=args.port)
 81 |     cfg = yaml.load(open(args.config, "r"), Loader=yaml.Loader)
 82 | 
 83 |     model = DeepLabV3Plus(cfg)
 84 |     model.load_state_dict(torch.load(args.checkpoint_path))
 85 |     model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model)
 86 |     model.cuda()
 87 | 
 88 |     local_rank = int(os.environ["LOCAL_RANK"])
 89 |     model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[local_rank],
 90 |                                                       output_device=local_rank, find_unused_parameters=False)
 91 | 
 92 |     valset = SemiDataset(cfg['dataset'], cfg['data_root'], 'val')
 93 |     valsampler = torch.utils.data.distributed.DistributedSampler(valset)
 94 |     valloader = DataLoader(valset, batch_size=1, pin_memory=True, num_workers=4,
 95 |                            drop_last=False, sampler=valsampler)
 96 | 
 97 |     model.eval()
 98 |     res_val = evaluate(model, valloader, 'original', cfg)
 99 |     mIOU = res_val['mIOU']
100 |     iou_class = res_val['iou_class']
101 |     print(mIOU)
102 |     print(iou_class)
103 | 
104 | 
105 | if __name__ == '__main__':
106 |     main()
107 | 


--------------------------------------------------------------------------------
/images/cvpr_pipeline.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/BBBBchan/CorrMatch/44f9f9e8648886621b1bdb4f95df9df773cf4f09/images/cvpr_pipeline.png


--------------------------------------------------------------------------------
/model/backbone/resnet.py:
--------------------------------------------------------------------------------
  1 | import torch
  2 | import torch.nn as nn
  3 | 
  4 | 
  5 | __all__ = ['ResNet', 'resnet50', 'resnet101']
  6 | 
  7 | 
  8 | def conv3x3(in_planes, out_planes, stride=1, groups=1, dilation=1):
  9 |     return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,
 10 |                      padding=dilation, groups=groups, bias=False, dilation=dilation)
 11 | 
 12 | 
 13 | def conv1x1(in_planes, out_planes, stride=1):
 14 |     return nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, bias=False)
 15 | 
 16 | 
 17 | class Bottleneck(nn.Module):
 18 |     expansion = 4
 19 | 
 20 |     def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1,
 21 |                  base_width=64, dilation=1, norm_layer=None):
 22 |         super(Bottleneck, self).__init__()
 23 |         if norm_layer is None:
 24 |             norm_layer = nn.BatchNorm2d
 25 |         width = int(planes * (base_width / 64.)) * groups
 26 | 
 27 |         self.conv1 = conv1x1(inplanes, width)
 28 |         self.bn1 = norm_layer(width)
 29 |         self.conv2 = conv3x3(width, width, stride, groups, dilation)
 30 |         self.bn2 = norm_layer(width)
 31 |         self.conv3 = conv1x1(width, planes * self.expansion)
 32 |         self.bn3 = norm_layer(planes * self.expansion)
 33 |         self.relu = nn.ReLU(inplace=True)
 34 |         self.downsample = downsample
 35 |         self.stride = stride
 36 | 
 37 |     def forward(self, x):
 38 |         identity = x
 39 | 
 40 |         out = self.conv1(x)
 41 |         out = self.bn1(out)
 42 |         out = self.relu(out)
 43 | 
 44 |         out = self.conv2(out)
 45 |         out = self.bn2(out)
 46 |         out = self.relu(out)
 47 | 
 48 |         out = self.conv3(out)
 49 |         out = self.bn3(out)
 50 | 
 51 |         if self.downsample is not None:
 52 |             identity = self.downsample(x)
 53 | 
 54 |         out += identity
 55 |         out = self.relu(out)
 56 | 
 57 |         return out
 58 | 
 59 | 
 60 | class ResNet(nn.Module):
 61 | 
 62 |     def __init__(self, block, layers, zero_init_residual=False, groups=1,
 63 |                  width_per_group=64, multi_grid=False, replace_stride_with_dilation=None, norm_layer=None):
 64 |         super(ResNet, self).__init__()
 65 | 
 66 |         if norm_layer is None:
 67 |             norm_layer = nn.BatchNorm2d
 68 |         self._norm_layer = norm_layer
 69 | 
 70 |         self.inplanes = 128
 71 |         self.dilation = 1
 72 |         if replace_stride_with_dilation is None:
 73 |             replace_stride_with_dilation = [False, False, False]
 74 |         if len(replace_stride_with_dilation) != 3:
 75 |             raise ValueError("replace_stride_with_dilation should be None "
 76 |                              "or a 3-element tuple, got {}".format(replace_stride_with_dilation))
 77 |         self.groups = groups
 78 |         self.base_width = width_per_group
 79 |         self.conv1 = nn.Sequential(
 80 |             nn.Conv2d(3, 64, kernel_size=3, stride=2, padding=1, bias=False),
 81 |             norm_layer(64),
 82 |             nn.ReLU(inplace=True),
 83 |             nn.Conv2d(64, 64, kernel_size=3, stride=1, padding=1, bias=False),
 84 |             norm_layer(64),
 85 |             nn.ReLU(inplace=True),
 86 |             nn.Conv2d(64, 128, kernel_size=3, stride=1, padding=1, bias=False),
 87 |         )
 88 |         self.bn1 = norm_layer(self.inplanes)
 89 |         self.relu = nn.ReLU(inplace=True)
 90 |         self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
 91 |         self.layer1 = self._make_layer(block, 64, layers[0])
 92 |         self.layer2 = self._make_layer(block, 128, layers[1], stride=2,
 93 |                                        dilate=replace_stride_with_dilation[0])
 94 |         self.layer3 = self._make_layer(block, 256, layers[2], stride=2,
 95 |                                        dilate=replace_stride_with_dilation[1])
 96 |         self.layer4 = self._make_layer(block, 512, layers[3], stride=2,
 97 |                                        dilate=replace_stride_with_dilation[2], multi_grid=multi_grid)
 98 | 
 99 |         for m in self.modules():
100 |             if isinstance(m, nn.Conv2d):
101 |                 nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
102 |             elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
103 |                 nn.init.constant_(m.weight, 1)
104 |                 nn.init.constant_(m.bias, 0)
105 | 
106 |         if zero_init_residual:
107 |             for m in self.modules():
108 |                 if isinstance(m, Bottleneck):
109 |                     nn.init.constant_(m.bn3.weight, 0)
110 | 
111 |     def _make_layer(self, block, planes, blocks, stride=1, dilate=False, multi_grid=False):
112 |         norm_layer = self._norm_layer
113 |         downsample = None
114 |         previous_dilation = self.dilation
115 |         if dilate:
116 |             self.dilation *= stride
117 |             stride = 1
118 |         if stride != 1 or self.inplanes != planes * block.expansion:
119 |             downsample = nn.Sequential(
120 |                 conv1x1(self.inplanes, planes * block.expansion, stride),
121 |                 norm_layer(planes * block.expansion),
122 |             )
123 | 
124 |         grids = [1] * blocks
125 |         if multi_grid:
126 |             grids = [2, 2, 4]
127 | 
128 |         layers = list()
129 |         layers.append(block(self.inplanes, planes, stride, downsample, self.groups,
130 |                             self.base_width, previous_dilation * grids[0], norm_layer))
131 |         self.inplanes = planes * block.expansion
132 |         for i in range(1, blocks):
133 |             layers.append(block(self.inplanes, planes, groups=self.groups,
134 |                                 base_width=self.base_width, dilation=self.dilation * grids[i],
135 |                                 norm_layer=norm_layer))
136 | 
137 |         return nn.Sequential(*layers)
138 | 
139 |     def base_forward(self, x):
140 |         x = self.conv1(x)
141 |         x = self.bn1(x)
142 |         x = self.relu(x)
143 |         x = self.maxpool(x)
144 | 
145 |         c1 = self.layer1(x)
146 |         c2 = self.layer2(c1)
147 |         c3 = self.layer3(c2)
148 |         c4 = self.layer4(c3)
149 | 
150 |         return c1, c2, c3, c4
151 | 
152 | 
153 | def _resnet(arch, block, layers, pretrained, **kwargs):
154 |     model = ResNet(block, layers, **kwargs)
155 |     if pretrained:
156 |         pretrained_path = "pretrained/%s.pth" % arch
157 |         state_dict = torch.load(pretrained_path)
158 |         model.load_state_dict(state_dict, strict=False)
159 |     return model
160 | 
161 | 
162 | def resnet50(pretrained=False, **kwargs):
163 |     return _resnet('resnet50', Bottleneck, [3, 4, 6, 3], pretrained, **kwargs)
164 | 
165 | 
166 | def resnet101(pretrained=False, **kwargs):
167 |     return _resnet('resnet101', Bottleneck, [3, 4, 23, 3], pretrained, **kwargs)
168 | 


--------------------------------------------------------------------------------
/model/backbone/xception.py:
--------------------------------------------------------------------------------
  1 | import math
  2 | import torch
  3 | import torch.nn as nn
  4 | import torch.nn.functional as F
  5 | import torch.utils.model_zoo as model_zoo
  6 | from torch.nn import init
  7 | 
  8 | bn_mom = 0.0003
  9 | __all__ = ['xception']
 10 | 
 11 | 
 12 | class SeparableConv2d(nn.Module):
 13 |     def __init__(self, in_channels, out_channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=False,
 14 |                  activate_first=True, inplace=True):
 15 |         super(SeparableConv2d, self).__init__()
 16 |         self.relu0 = nn.ReLU(inplace=inplace)
 17 |         self.depthwise = nn.Conv2d(in_channels, in_channels, kernel_size, stride, padding, dilation, groups=in_channels,
 18 |                                    bias=bias)
 19 |         self.bn1 = nn.BatchNorm2d(in_channels, momentum=bn_mom)
 20 |         self.relu1 = nn.ReLU(inplace=True)
 21 |         self.pointwise = nn.Conv2d(in_channels, out_channels, 1, 1, 0, 1, 1, bias=bias)
 22 |         self.bn2 = nn.BatchNorm2d(out_channels, momentum=bn_mom)
 23 |         self.relu2 = nn.ReLU(inplace=True)
 24 |         self.activate_first = activate_first
 25 | 
 26 |     def forward(self, x):
 27 |         if self.activate_first:
 28 |             x = self.relu0(x)
 29 |         x = self.depthwise(x)
 30 |         x = self.bn1(x)
 31 |         if not self.activate_first:
 32 |             x = self.relu1(x)
 33 |         x = self.pointwise(x)
 34 |         x = self.bn2(x)
 35 |         if not self.activate_first:
 36 |             x = self.relu2(x)
 37 |         return x
 38 | 
 39 | 
 40 | class Block(nn.Module):
 41 |     def __init__(self, in_filters, out_filters, strides=1, atrous=None, grow_first=True, activate_first=True,
 42 |                  inplace=True):
 43 |         super(Block, self).__init__()
 44 |         if atrous == None:
 45 |             atrous = [1] * 3
 46 |         elif isinstance(atrous, int):
 47 |             atrous_list = [atrous] * 3
 48 |             atrous = atrous_list
 49 |         idx = 0
 50 |         self.head_relu = True
 51 |         if out_filters != in_filters or strides != 1:
 52 |             self.skip = nn.Conv2d(in_filters, out_filters, 1, stride=strides, bias=False)
 53 |             self.skipbn = nn.BatchNorm2d(out_filters, momentum=bn_mom)
 54 |             self.head_relu = False
 55 |         else:
 56 |             self.skip = None
 57 | 
 58 |         self.hook_layer = None
 59 |         if grow_first:
 60 |             filters = out_filters
 61 |         else:
 62 |             filters = in_filters
 63 |         self.sepconv1 = SeparableConv2d(in_filters, filters, 3, stride=1, padding=1 * atrous[0], dilation=atrous[0],
 64 |                                         bias=False, activate_first=activate_first, inplace=self.head_relu)
 65 |         self.sepconv2 = SeparableConv2d(filters, out_filters, 3, stride=1, padding=1 * atrous[1], dilation=atrous[1],
 66 |                                         bias=False, activate_first=activate_first)
 67 |         self.sepconv3 = SeparableConv2d(out_filters, out_filters, 3, stride=strides, padding=1 * atrous[2],
 68 |                                         dilation=atrous[2], bias=False, activate_first=activate_first, inplace=inplace)
 69 | 
 70 |     def forward(self, inp):
 71 | 
 72 |         if self.skip is not None:
 73 |             skip = self.skip(inp)
 74 |             skip = self.skipbn(skip)
 75 |         else:
 76 |             skip = inp
 77 | 
 78 |         x = self.sepconv1(inp)
 79 |         x = self.sepconv2(x)
 80 |         self.hook_layer = x
 81 |         x = self.sepconv3(x)
 82 | 
 83 |         x += skip
 84 |         return x
 85 | 
 86 | 
 87 | class Xception(nn.Module):
 88 |     """
 89 |     Xception optimized for the ImageNet dataset, as specified in
 90 |     https://arxiv.org/pdf/1610.02357.pdf
 91 |     """
 92 | 
 93 |     def __init__(self, os):
 94 |         """ Constructor
 95 |         Args:
 96 |             num_classes: number of classes
 97 |         """
 98 |         super(Xception, self).__init__()
 99 | 
100 |         if os == 8:
101 |             stride_list = [2, 1, 1]
102 |         elif os == 16:
103 |             stride_list = [2, 2, 1]
104 |         else:
105 |             raise ValueError('xception.py: output stride=%d is not supported.' % os)
106 |         self.conv1 = nn.Conv2d(3, 32, 3, 2, 1, bias=False)
107 |         self.bn1 = nn.BatchNorm2d(32, momentum=bn_mom)
108 |         self.relu = nn.ReLU(inplace=True)
109 | 
110 |         self.conv2 = nn.Conv2d(32, 64, 3, 1, 1, bias=False)
111 |         self.bn2 = nn.BatchNorm2d(64, momentum=bn_mom)
112 | 
113 |         self.block1 = Block(64, 128, 2)
114 |         self.block2 = Block(128, 256, stride_list[0], inplace=False)
115 |         self.block3 = Block(256, 728, stride_list[1])
116 | 
117 |         rate = 16 // os
118 |         self.block4 = Block(728, 728, 1, atrous=rate)
119 |         self.block5 = Block(728, 728, 1, atrous=rate)
120 |         self.block6 = Block(728, 728, 1, atrous=rate)
121 |         self.block7 = Block(728, 728, 1, atrous=rate)
122 | 
123 |         self.block8 = Block(728, 728, 1, atrous=rate)
124 |         self.block9 = Block(728, 728, 1, atrous=rate)
125 |         self.block10 = Block(728, 728, 1, atrous=rate)
126 |         self.block11 = Block(728, 728, 1, atrous=rate)
127 | 
128 |         self.block12 = Block(728, 728, 1, atrous=rate)
129 |         self.block13 = Block(728, 728, 1, atrous=rate)
130 |         self.block14 = Block(728, 728, 1, atrous=rate)
131 |         self.block15 = Block(728, 728, 1, atrous=rate)
132 | 
133 |         self.block16 = Block(728, 728, 1, atrous=[1 * rate, 1 * rate, 1 * rate])
134 |         self.block17 = Block(728, 728, 1, atrous=[1 * rate, 1 * rate, 1 * rate])
135 |         self.block18 = Block(728, 728, 1, atrous=[1 * rate, 1 * rate, 1 * rate])
136 |         self.block19 = Block(728, 728, 1, atrous=[1 * rate, 1 * rate, 1 * rate])
137 | 
138 |         self.block20 = Block(728, 1024, stride_list[2], atrous=rate, grow_first=False)
139 | 
140 |         self.conv3 = SeparableConv2d(1024, 1536, 3, 1, 1 * rate, dilation=rate, activate_first=False)
141 | 
142 |         self.conv4 = SeparableConv2d(1536, 1536, 3, 1, 1 * rate, dilation=rate, activate_first=False)
143 | 
144 |         # do relu here
145 |         self.conv5 = SeparableConv2d(1536, 2048, 3, 1, 1 * rate, dilation=rate, activate_first=False)
146 |         self.layers = []
147 | 
148 |         # ------- init weights --------
149 |         for m in self.modules():
150 |             if isinstance(m, nn.Conv2d):
151 |                 n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
152 |                 m.weight.data.normal_(0, math.sqrt(2. / n))
153 |             elif isinstance(m, nn.BatchNorm2d):
154 |                 m.weight.data.fill_(1)
155 |                 m.bias.data.zero_()
156 |         # -----------------------------
157 | 
158 |     def base_forward(self, input):
159 |         self.layers = []
160 |         x = self.conv1(input)
161 |         x = self.bn1(x)
162 |         x = self.relu(x)
163 |         # self.layers.append(x)
164 |         x = self.conv2(x)
165 |         x = self.bn2(x)
166 |         x = self.relu(x)
167 | 
168 |         x = self.block1(x)
169 |         x = self.block2(x)
170 |         # self.layers.append(self.block2.hook_layer)
171 |         c1 = self.block2.hook_layer
172 |         x = self.block3(x)
173 |         # self.layers.append(self.block3.hook_layer)
174 |         x = self.block4(x)
175 |         x = self.block5(x)
176 |         x = self.block6(x)
177 |         x = self.block7(x)
178 |         x = self.block8(x)
179 |         x = self.block9(x)
180 |         x = self.block10(x)
181 |         x = self.block11(x)
182 |         x = self.block12(x)
183 |         x = self.block13(x)
184 |         x = self.block14(x)
185 |         x = self.block15(x)
186 |         x = self.block16(x)
187 |         x = self.block17(x)
188 |         x = self.block18(x)
189 |         x = self.block19(x)
190 |         x = self.block20(x)
191 | 
192 |         x = self.conv3(x)
193 | 
194 |         x = self.conv4(x)
195 | 
196 |         x = self.conv5(x)
197 | 
198 |         return c1, x
199 | 
200 | 
201 | def xception(pretrained=True, os=16):
202 |     model = Xception(os=os)
203 |     if pretrained:
204 |         old_dict = torch.load('pretrained/xception.pth')
205 |         model_dict = model.state_dict()
206 |         old_dict = {k: v for k, v in old_dict.items() if ('itr' not in k and 'tmp' not in k and 'track' not in k)}
207 |         model_dict.update(old_dict)
208 | 
209 |         model.load_state_dict(model_dict)
210 | 
211 |     return model
212 | 


--------------------------------------------------------------------------------
/model/semseg/deeplabv3plus.py:
--------------------------------------------------------------------------------
  1 | import model.backbone.resnet as resnet
  2 | from model.backbone.xception import xception
  3 | 
  4 | import torch
  5 | from torch import nn
  6 | import torch.nn.functional as F
  7 | import math
  8 | from einops import rearrange
  9 | 
 10 | 
 11 | class DeepLabV3Plus(nn.Module):
 12 |     def __init__(self, cfg):
 13 |         super(DeepLabV3Plus, self).__init__()
 14 |         self.is_corr = True
 15 | 
 16 |         if 'resnet' in cfg['backbone']:
 17 |             self.backbone = \
 18 |                 resnet.__dict__[cfg['backbone']](cfg['pretrain'], multi_grid=cfg['multi_grid'],
 19 |                                                  replace_stride_with_dilation=cfg['replace_stride_with_dilation'])
 20 |         else:
 21 |             assert cfg['backbone'] == 'xception'
 22 |             self.backbone = xception(True)
 23 | 
 24 |         low_channels = 256
 25 |         high_channels = 2048
 26 | 
 27 |         self.head = ASPPModule(high_channels, cfg['dilations'])
 28 | 
 29 |         self.reduce = nn.Sequential(nn.Conv2d(low_channels, 48, 1, bias=False),
 30 |                                     nn.BatchNorm2d(48),
 31 |                                     nn.ReLU(True))
 32 | 
 33 |         self.fuse = nn.Sequential(nn.Conv2d(high_channels // 8 + 48, 256, 3, padding=1, bias=False),
 34 |                                   nn.BatchNorm2d(256),
 35 |                                   nn.ReLU(True),
 36 |                                   nn.Conv2d(256, 256, 3, padding=1, bias=False),
 37 |                                   nn.BatchNorm2d(256),
 38 |                                   nn.ReLU(True))
 39 | 
 40 |         self.classifier = nn.Conv2d(256, cfg['nclass'], 1, bias=True)
 41 | 
 42 | 
 43 |         if self.is_corr:
 44 |             self.corr = Corr(nclass=cfg['nclass'])
 45 |             self.proj = nn.Sequential(
 46 |                 nn.Conv2d(2048, 256, kernel_size=3, stride=1, padding=1, bias=True),
 47 |                 nn.BatchNorm2d(256),
 48 |                 nn.ReLU(inplace=True),
 49 |                 nn.Dropout2d(0.1),
 50 |             )
 51 | 
 52 |     def forward(self, x, need_fp=False, use_corr=False):
 53 |         dict_return = {}
 54 |         h, w = x.shape[-2:]
 55 | 
 56 |         feats = self.backbone.base_forward(x)
 57 |         c1, c4 = feats[0], feats[-1]
 58 | 
 59 |         if need_fp:
 60 |             feats_decode = self._decode(torch.cat((c1, nn.Dropout2d(0.5)(c1))), torch.cat((c4, nn.Dropout2d(0.5)(c4))))
 61 |             outs = self.classifier(feats_decode)
 62 |             outs = F.interpolate(outs, size=(h, w), mode="bilinear", align_corners=True)
 63 |             out, out_fp = outs.chunk(2)
 64 |             if use_corr:
 65 |                 proj_feats = self.proj(c4)
 66 |                 corr_out_dict = self.corr(proj_feats, out)
 67 |                 dict_return['corr_map'] = corr_out_dict['corr_map']
 68 |                 corr_out = corr_out_dict['out']
 69 |                 corr_out = F.interpolate(corr_out, size=(h, w), mode="bilinear", align_corners=True)
 70 |                 dict_return['corr_out'] = corr_out
 71 |             dict_return['out'] = out
 72 |             dict_return['out_fp'] = out_fp
 73 | 
 74 |             return dict_return
 75 | 
 76 |         feats_decode = self._decode(c1, c4)
 77 |         out = self.classifier(feats_decode)
 78 |         out = F.interpolate(out, size=(h, w), mode="bilinear", align_corners=True)
 79 |         if use_corr:
 80 |             proj_feats = self.proj(c4)
 81 |             corr_out_dict = self.corr(proj_feats, out)
 82 |             dict_return['corr_map'] = corr_out_dict['corr_map']
 83 |             corr_out = corr_out_dict['out']
 84 |             corr_out = F.interpolate(corr_out, size=(h, w), mode="bilinear", align_corners=True)
 85 |             dict_return['corr_out'] = corr_out
 86 |         dict_return['out'] = out
 87 |         return dict_return
 88 | 
 89 |     def _decode(self, c1, c4):
 90 |         c4 = self.head(c4)
 91 |         c4 = F.interpolate(c4, size=c1.shape[-2:], mode="bilinear", align_corners=True)
 92 | 
 93 |         c1 = self.reduce(c1)
 94 | 
 95 |         feature = torch.cat([c1, c4], dim=1)
 96 |         feature = self.fuse(feature)
 97 | 
 98 |         return feature
 99 | 
100 | 
101 | def ASPPConv(in_channels, out_channels, atrous_rate):
102 |     block = nn.Sequential(nn.Conv2d(in_channels, out_channels, 3, padding=atrous_rate,
103 |                                     dilation=atrous_rate, bias=False),
104 |                           nn.BatchNorm2d(out_channels),
105 |                           nn.ReLU(True))
106 |     return block
107 | 
108 | 
109 | class ASPPPooling(nn.Module):
110 |     def __init__(self, in_channels, out_channels):
111 |         super(ASPPPooling, self).__init__()
112 |         self.gap = nn.Sequential(nn.AdaptiveAvgPool2d(1),
113 |                                  nn.Conv2d(in_channels, out_channels, 1, bias=False),
114 |                                  nn.BatchNorm2d(out_channels),
115 |                                  nn.ReLU(True))
116 | 
117 |     def forward(self, x):
118 |         h, w = x.shape[-2:]
119 |         pool = self.gap(x)
120 |         return F.interpolate(pool, (h, w), mode="bilinear", align_corners=True)
121 | 
122 | 
123 | class ASPPModule(nn.Module):
124 |     def __init__(self, in_channels, atrous_rates):
125 |         super(ASPPModule, self).__init__()
126 |         out_channels = in_channels // 8
127 |         rate1, rate2, rate3 = atrous_rates
128 | 
129 |         self.b0 = nn.Sequential(nn.Conv2d(in_channels, out_channels, 1, bias=False),
130 |                                 nn.BatchNorm2d(out_channels),
131 |                                 nn.ReLU(True))
132 |         self.b1 = ASPPConv(in_channels, out_channels, rate1)
133 |         self.b2 = ASPPConv(in_channels, out_channels, rate2)
134 |         self.b3 = ASPPConv(in_channels, out_channels, rate3)
135 |         self.b4 = ASPPPooling(in_channels, out_channels)
136 | 
137 |         self.project = nn.Sequential(nn.Conv2d(5 * out_channels, out_channels, 1, bias=False),
138 |                                      nn.BatchNorm2d(out_channels),
139 |                                      nn.ReLU(True))
140 | 
141 |     def forward(self, x):
142 |         feat0 = self.b0(x)
143 |         feat1 = self.b1(x)
144 |         feat2 = self.b2(x)
145 |         feat3 = self.b3(x)
146 |         feat4 = self.b4(x)
147 |         y = torch.cat((feat0, feat1, feat2, feat3, feat4), 1)
148 |         return self.project(y)
149 | 
150 | 
151 | class Corr(nn.Module):
152 |     def __init__(self, nclass=21):
153 |         super(Corr, self).__init__()
154 |         self.nclass = nclass
155 |         self.conv1 = nn.Conv2d(256, self.nclass, kernel_size=1, stride=1, padding=0, bias=True)
156 |         self.conv2 = nn.Conv2d(256, self.nclass, kernel_size=1, stride=1, padding=0, bias=True)
157 | 
158 |     def forward(self, feature_in, out):
159 |         dict_return = {}
160 |         h_in, w_in = math.ceil(feature_in.shape[2] / (1)), math.ceil(feature_in.shape[3] / (1))
161 |         h_out, w_out = out.shape[2], out.shape[3]
162 |         out = F.interpolate(out.detach(), (h_in, w_in), mode='bilinear', align_corners=True)
163 |         feature = F.interpolate(feature_in, (h_in, w_in), mode='bilinear', align_corners=True)
164 |         f1 = rearrange(self.conv1(feature), 'n c h w -> n c (h w)')
165 |         f2 = rearrange(self.conv2(feature), 'n c h w -> n c (h w)')
166 |         out_temp = rearrange(out, 'n c h w -> n c (h w)')
167 |         corr_map = torch.matmul(f1.transpose(1, 2), f2) / torch.sqrt(torch.tensor(f1.shape[1]).float())
168 |         corr_map = F.softmax(corr_map, dim=-1)
169 |         corr_map_sample = self.sample(corr_map.detach(), h_in, w_in)
170 |         dict_return['corr_map'] = self.normalize_corr_map(corr_map_sample, h_in, w_in, h_out, w_out)
171 |         dict_return['out'] = rearrange(torch.matmul(out_temp, corr_map), 'n c (h w) -> n c h w', h=h_in, w=w_in)
172 |         return dict_return
173 | 
174 |     def sample(self, corr_map, h_in, w_in):
175 |         index = torch.randint(0, h_in * w_in - 1, [128])
176 |         corr_map_sample = corr_map[:, index.long(), :]
177 |         return corr_map_sample
178 | 
179 |     def normalize_corr_map(self, corr_map, h_in, w_in, h_out, w_out):
180 |         n, m, hw = corr_map.shape
181 |         corr_map = rearrange(corr_map, 'n m (h w) -> (n m) 1 h w', h=h_in, w=w_in)
182 |         corr_map = F.interpolate(corr_map, (h_out, w_out), mode='bilinear', align_corners=True)
183 | 
184 |         corr_map = rearrange(corr_map, '(n m) 1 h w -> (n m) (h w)', n=n, m=m)
185 |         range_ = torch.max(corr_map, dim=1, keepdim=True)[0] - torch.min(corr_map, dim=1, keepdim=True)[0]
186 |         temp_map = ((- torch.min(corr_map, dim=1, keepdim=True)[0]) + corr_map) / range_
187 |         corr_map = (temp_map > 0.5)
188 |         norm_corr_map = rearrange(corr_map, '(n m) (h w) -> n m h w', n=n, m=m, h=h_out, w=w_out)
189 |         return norm_corr_map
190 | 
191 | 
192 | 


--------------------------------------------------------------------------------
/model/semseg/deeplabv3plus_vis.py:
--------------------------------------------------------------------------------
  1 | import model.backbone.resnet as resnet
  2 | from model.backbone.xception import xception
  3 | 
  4 | import torch
  5 | from torch import nn
  6 | import torch.nn.functional as F
  7 | import math
  8 | from einops import rearrange
  9 | 
 10 | 
 11 | class DeepLabV3Plus(nn.Module):
 12 |     def __init__(self, cfg):
 13 |         super(DeepLabV3Plus, self).__init__()
 14 |         self.is_corr = True
 15 | 
 16 |         if 'resnet' in cfg['backbone']:
 17 |             self.backbone = \
 18 |                 resnet.__dict__[cfg['backbone']](cfg['pretrain'], multi_grid=cfg['multi_grid'],
 19 |                                                  replace_stride_with_dilation=cfg['replace_stride_with_dilation'])
 20 |         else:
 21 |             assert cfg['backbone'] == 'xception'
 22 |             self.backbone = xception(True)
 23 | 
 24 |         low_channels = 256
 25 |         high_channels = 2048
 26 | 
 27 |         self.head = ASPPModule(high_channels, cfg['dilations'])
 28 | 
 29 |         self.reduce = nn.Sequential(nn.Conv2d(low_channels, 48, 1, bias=False),
 30 |                                     nn.BatchNorm2d(48),
 31 |                                     nn.ReLU(True))
 32 | 
 33 |         self.fuse = nn.Sequential(nn.Conv2d(high_channels // 8 + 48, 256, 3, padding=1, bias=False),
 34 |                                   nn.BatchNorm2d(256),
 35 |                                   nn.ReLU(True),
 36 |                                   nn.Conv2d(256, 256, 3, padding=1, bias=False),
 37 |                                   nn.BatchNorm2d(256),
 38 |                                   nn.ReLU(True))
 39 | 
 40 |         self.classifier = nn.Conv2d(256, cfg['nclass'], 1, bias=True)
 41 | 
 42 |         if self.is_corr:
 43 |             self.corr = Corr(nclass=cfg['nclass'])
 44 |             self.proj = nn.Sequential(
 45 |                 nn.Conv2d(2048, 256, kernel_size=3, stride=1, padding=1, bias=True),
 46 |                 nn.BatchNorm2d(256),
 47 |                 nn.ReLU(inplace=True),
 48 |                 nn.Dropout2d(0.1),
 49 |             )
 50 | 
 51 |     def forward(self, x, need_fp=False, use_corr=False):
 52 |         dict_return = {}
 53 |         h, w = x.shape[-2:]
 54 | 
 55 |         feats = self.backbone.base_forward(x)
 56 |         c1, c4 = feats[0], feats[-1]
 57 | 
 58 |         if need_fp:
 59 |             feats_decode = self._decode(torch.cat((c1, nn.Dropout2d(0.5)(c1))), torch.cat((c4, nn.Dropout2d(0.5)(c4))))
 60 |             outs = self.classifier(feats_decode)
 61 |             outs = F.interpolate(outs, size=(h, w), mode="bilinear", align_corners=True)
 62 |             out, out_fp = outs.chunk(2)
 63 |             if use_corr:
 64 |                 proj_feats = self.proj(c4)
 65 |                 corr_out, corr_map = self.corr(proj_feats, out)
 66 |                 corr_out = F.interpolate(corr_out, size=(h, w), mode="bilinear", align_corners=True)
 67 |                 dict_return['corr_out'] = corr_out
 68 |                 dict_return['corr_map'] = corr_map
 69 |                 dict_return['c4'] = c4
 70 |             dict_return['out'] = out
 71 |             dict_return['out_fp'] = out_fp
 72 | 
 73 |             return dict_return
 74 | 
 75 |         feats_decode = self._decode(c1, c4)
 76 |         out = self.classifier(feats_decode)
 77 |         out = F.interpolate(out, size=(h, w), mode="bilinear", align_corners=True)
 78 |         if use_corr:
 79 |             proj_feats = self.proj(c4)
 80 |             corr_out, corr_map = self.corr(proj_feats, out)
 81 |             corr_out = F.interpolate(corr_out, size=(h, w), mode="bilinear", align_corners=True)
 82 |             dict_return['corr_out'] = corr_out
 83 |             dict_return['corr_map'] = corr_map
 84 |             dict_return['c4'] = c4
 85 |         dict_return['out'] = out
 86 |         return dict_return
 87 | 
 88 |     def _decode(self, c1, c4):
 89 |         c4 = self.head(c4)
 90 |         c4 = F.interpolate(c4, size=c1.shape[-2:], mode="bilinear", align_corners=True)
 91 | 
 92 |         c1 = self.reduce(c1)
 93 | 
 94 |         feature = torch.cat([c1, c4], dim=1)
 95 |         feature = self.fuse(feature)
 96 | 
 97 |         return feature
 98 | 
 99 | 
100 | def ASPPConv(in_channels, out_channels, atrous_rate):
101 |     block = nn.Sequential(nn.Conv2d(in_channels, out_channels, 3, padding=atrous_rate,
102 |                                     dilation=atrous_rate, bias=False),
103 |                           nn.BatchNorm2d(out_channels),
104 |                           nn.ReLU(True))
105 |     return block
106 | 
107 | 
108 | class ASPPPooling(nn.Module):
109 |     def __init__(self, in_channels, out_channels):
110 |         super(ASPPPooling, self).__init__()
111 |         self.gap = nn.Sequential(nn.AdaptiveAvgPool2d(1),
112 |                                  nn.Conv2d(in_channels, out_channels, 1, bias=False),
113 |                                  nn.BatchNorm2d(out_channels),
114 |                                  nn.ReLU(True))
115 | 
116 |     def forward(self, x):
117 |         h, w = x.shape[-2:]
118 |         pool = self.gap(x)
119 |         return F.interpolate(pool, (h, w), mode="bilinear", align_corners=True)
120 | 
121 | 
122 | class ASPPModule(nn.Module):
123 |     def __init__(self, in_channels, atrous_rates):
124 |         super(ASPPModule, self).__init__()
125 |         out_channels = in_channels // 8
126 |         rate1, rate2, rate3 = atrous_rates
127 | 
128 |         self.b0 = nn.Sequential(nn.Conv2d(in_channels, out_channels, 1, bias=False),
129 |                                 nn.BatchNorm2d(out_channels),
130 |                                 nn.ReLU(True))
131 |         self.b1 = ASPPConv(in_channels, out_channels, rate1)
132 |         self.b2 = ASPPConv(in_channels, out_channels, rate2)
133 |         self.b3 = ASPPConv(in_channels, out_channels, rate3)
134 |         self.b4 = ASPPPooling(in_channels, out_channels)
135 | 
136 |         self.project = nn.Sequential(nn.Conv2d(5 * out_channels, out_channels, 1, bias=False),
137 |                                      nn.BatchNorm2d(out_channels),
138 |                                      nn.ReLU(True))
139 | 
140 |     def forward(self, x):
141 |         feat0 = self.b0(x)
142 |         feat1 = self.b1(x)
143 |         feat2 = self.b2(x)
144 |         feat3 = self.b3(x)
145 |         feat4 = self.b4(x)
146 |         y = torch.cat((feat0, feat1, feat2, feat3, feat4), 1)
147 |         return self.project(y)
148 | 
149 | 
150 | class Corr(nn.Module):
151 |     def __init__(self, nclass=21):
152 |         super(Corr, self).__init__()
153 |         self.nclass = nclass
154 |         self.conv1 = nn.Conv2d(256, self.nclass, kernel_size=1, stride=1, padding=0, bias=True)
155 |         self.conv2 = nn.Conv2d(256, self.nclass, kernel_size=1, stride=1, padding=0, bias=True)
156 | 
157 |     def forward(self, feature_in, out):
158 |         h_in, w_in = math.ceil(feature_in.shape[2] / (1)), math.ceil(feature_in.shape[3] / (1))
159 |         out = F.interpolate(out.detach(), (h_in, w_in), mode='bilinear', align_corners=True)
160 |         feature = F.interpolate(feature_in, (h_in, w_in), mode='bilinear', align_corners=True)
161 |         f1 = rearrange(self.conv1(feature), 'n c h w -> n c (h w)')
162 |         f2 = rearrange(self.conv2(feature), 'n c h w -> n c (h w)')
163 |         out_temp = rearrange(out, 'n c h w -> n c (h w)')
164 |         corr_map = torch.matmul(f1.transpose(1, 2), f2) / torch.sqrt(torch.tensor(f1.shape[1]).float())
165 |         corr_map = F.softmax(corr_map, dim=-1)
166 |         out = rearrange(torch.matmul(out_temp, corr_map), 'n c (h w) -> n c h w', h=h_in, w=w_in)
167 |         return out, corr_map.detach()
168 | 
169 | 


--------------------------------------------------------------------------------
/partitions/cityscapes/1_16/labeled.txt:
--------------------------------------------------------------------------------
  1 | leftImg8bit/train/jena/jena_000078_000019_leftImg8bit.png gtFine/train/jena/jena_000078_000019_gtFine_labelTrainIds.png
  2 | leftImg8bit/train/jena/jena_000065_000019_leftImg8bit.png gtFine/train/jena/jena_000065_000019_gtFine_labelTrainIds.png
  3 | leftImg8bit/train/jena/jena_000005_000019_leftImg8bit.png gtFine/train/jena/jena_000005_000019_gtFine_labelTrainIds.png
  4 | leftImg8bit/train/jena/jena_000102_000019_leftImg8bit.png gtFine/train/jena/jena_000102_000019_gtFine_labelTrainIds.png
  5 | leftImg8bit/train/jena/jena_000022_000019_leftImg8bit.png gtFine/train/jena/jena_000022_000019_gtFine_labelTrainIds.png
  6 | leftImg8bit/train/jena/jena_000011_000019_leftImg8bit.png gtFine/train/jena/jena_000011_000019_gtFine_labelTrainIds.png
  7 | leftImg8bit/train/jena/jena_000108_000019_leftImg8bit.png gtFine/train/jena/jena_000108_000019_gtFine_labelTrainIds.png
  8 | leftImg8bit/train/jena/jena_000025_000019_leftImg8bit.png gtFine/train/jena/jena_000025_000019_gtFine_labelTrainIds.png
  9 | leftImg8bit/train/stuttgart/stuttgart_000191_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000191_000019_gtFine_labelTrainIds.png
 10 | leftImg8bit/train/stuttgart/stuttgart_000067_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000067_000019_gtFine_labelTrainIds.png
 11 | leftImg8bit/train/stuttgart/stuttgart_000037_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000037_000019_gtFine_labelTrainIds.png
 12 | leftImg8bit/train/stuttgart/stuttgart_000046_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000046_000019_gtFine_labelTrainIds.png
 13 | leftImg8bit/train/stuttgart/stuttgart_000028_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000028_000019_gtFine_labelTrainIds.png
 14 | leftImg8bit/train/stuttgart/stuttgart_000016_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000016_000019_gtFine_labelTrainIds.png
 15 | leftImg8bit/train/stuttgart/stuttgart_000004_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000004_000019_gtFine_labelTrainIds.png
 16 | leftImg8bit/train/stuttgart/stuttgart_000005_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000005_000019_gtFine_labelTrainIds.png
 17 | leftImg8bit/train/stuttgart/stuttgart_000159_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000159_000019_gtFine_labelTrainIds.png
 18 | leftImg8bit/train/stuttgart/stuttgart_000068_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000068_000019_gtFine_labelTrainIds.png
 19 | leftImg8bit/train/stuttgart/stuttgart_000180_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000180_000019_gtFine_labelTrainIds.png
 20 | leftImg8bit/train/stuttgart/stuttgart_000109_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000109_000019_gtFine_labelTrainIds.png
 21 | leftImg8bit/train/ulm/ulm_000001_000019_leftImg8bit.png gtFine/train/ulm/ulm_000001_000019_gtFine_labelTrainIds.png
 22 | leftImg8bit/train/ulm/ulm_000027_000019_leftImg8bit.png gtFine/train/ulm/ulm_000027_000019_gtFine_labelTrainIds.png
 23 | leftImg8bit/train/ulm/ulm_000076_000019_leftImg8bit.png gtFine/train/ulm/ulm_000076_000019_gtFine_labelTrainIds.png
 24 | leftImg8bit/train/ulm/ulm_000053_000019_leftImg8bit.png gtFine/train/ulm/ulm_000053_000019_gtFine_labelTrainIds.png
 25 | leftImg8bit/train/ulm/ulm_000060_000019_leftImg8bit.png gtFine/train/ulm/ulm_000060_000019_gtFine_labelTrainIds.png
 26 | leftImg8bit/train/ulm/ulm_000051_000019_leftImg8bit.png gtFine/train/ulm/ulm_000051_000019_gtFine_labelTrainIds.png
 27 | leftImg8bit/train/dusseldorf/dusseldorf_000194_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000194_000019_gtFine_labelTrainIds.png
 28 | leftImg8bit/train/dusseldorf/dusseldorf_000077_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000077_000019_gtFine_labelTrainIds.png
 29 | leftImg8bit/train/dusseldorf/dusseldorf_000103_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000103_000019_gtFine_labelTrainIds.png
 30 | leftImg8bit/train/dusseldorf/dusseldorf_000163_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000163_000019_gtFine_labelTrainIds.png
 31 | leftImg8bit/train/dusseldorf/dusseldorf_000124_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000124_000019_gtFine_labelTrainIds.png
 32 | leftImg8bit/train/dusseldorf/dusseldorf_000079_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000079_000019_gtFine_labelTrainIds.png
 33 | leftImg8bit/train/dusseldorf/dusseldorf_000091_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000091_000019_gtFine_labelTrainIds.png
 34 | leftImg8bit/train/dusseldorf/dusseldorf_000026_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000026_000019_gtFine_labelTrainIds.png
 35 | leftImg8bit/train/dusseldorf/dusseldorf_000203_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000203_000019_gtFine_labelTrainIds.png
 36 | leftImg8bit/train/dusseldorf/dusseldorf_000129_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000129_000019_gtFine_labelTrainIds.png
 37 | leftImg8bit/train/dusseldorf/dusseldorf_000123_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000123_000019_gtFine_labelTrainIds.png
 38 | leftImg8bit/train/dusseldorf/dusseldorf_000184_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000184_000019_gtFine_labelTrainIds.png
 39 | leftImg8bit/train/dusseldorf/dusseldorf_000096_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000096_000019_gtFine_labelTrainIds.png
 40 | leftImg8bit/train/dusseldorf/dusseldorf_000084_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000084_000019_gtFine_labelTrainIds.png
 41 | leftImg8bit/train/darmstadt/darmstadt_000039_000019_leftImg8bit.png gtFine/train/darmstadt/darmstadt_000039_000019_gtFine_labelTrainIds.png
 42 | leftImg8bit/train/darmstadt/darmstadt_000014_000019_leftImg8bit.png gtFine/train/darmstadt/darmstadt_000014_000019_gtFine_labelTrainIds.png
 43 | leftImg8bit/train/darmstadt/darmstadt_000070_000019_leftImg8bit.png gtFine/train/darmstadt/darmstadt_000070_000019_gtFine_labelTrainIds.png
 44 | leftImg8bit/train/darmstadt/darmstadt_000049_000019_leftImg8bit.png gtFine/train/darmstadt/darmstadt_000049_000019_gtFine_labelTrainIds.png
 45 | leftImg8bit/train/darmstadt/darmstadt_000004_000019_leftImg8bit.png gtFine/train/darmstadt/darmstadt_000004_000019_gtFine_labelTrainIds.png
 46 | leftImg8bit/train/zurich/zurich_000037_000019_leftImg8bit.png gtFine/train/zurich/zurich_000037_000019_gtFine_labelTrainIds.png
 47 | leftImg8bit/train/zurich/zurich_000091_000019_leftImg8bit.png gtFine/train/zurich/zurich_000091_000019_gtFine_labelTrainIds.png
 48 | leftImg8bit/train/zurich/zurich_000084_000019_leftImg8bit.png gtFine/train/zurich/zurich_000084_000019_gtFine_labelTrainIds.png
 49 | leftImg8bit/train/zurich/zurich_000010_000019_leftImg8bit.png gtFine/train/zurich/zurich_000010_000019_gtFine_labelTrainIds.png
 50 | leftImg8bit/train/zurich/zurich_000067_000019_leftImg8bit.png gtFine/train/zurich/zurich_000067_000019_gtFine_labelTrainIds.png
 51 | leftImg8bit/train/zurich/zurich_000107_000019_leftImg8bit.png gtFine/train/zurich/zurich_000107_000019_gtFine_labelTrainIds.png
 52 | leftImg8bit/train/zurich/zurich_000057_000019_leftImg8bit.png gtFine/train/zurich/zurich_000057_000019_gtFine_labelTrainIds.png
 53 | leftImg8bit/train/zurich/zurich_000041_000019_leftImg8bit.png gtFine/train/zurich/zurich_000041_000019_gtFine_labelTrainIds.png
 54 | leftImg8bit/train/bremen/bremen_000164_000019_leftImg8bit.png gtFine/train/bremen/bremen_000164_000019_gtFine_labelTrainIds.png
 55 | leftImg8bit/train/bremen/bremen_000265_000019_leftImg8bit.png gtFine/train/bremen/bremen_000265_000019_gtFine_labelTrainIds.png
 56 | leftImg8bit/train/bremen/bremen_000260_000019_leftImg8bit.png gtFine/train/bremen/bremen_000260_000019_gtFine_labelTrainIds.png
 57 | leftImg8bit/train/bremen/bremen_000285_000019_leftImg8bit.png gtFine/train/bremen/bremen_000285_000019_gtFine_labelTrainIds.png
 58 | leftImg8bit/train/bremen/bremen_000255_000019_leftImg8bit.png gtFine/train/bremen/bremen_000255_000019_gtFine_labelTrainIds.png
 59 | leftImg8bit/train/bremen/bremen_000056_000019_leftImg8bit.png gtFine/train/bremen/bremen_000056_000019_gtFine_labelTrainIds.png
 60 | leftImg8bit/train/bremen/bremen_000303_000019_leftImg8bit.png gtFine/train/bremen/bremen_000303_000019_gtFine_labelTrainIds.png
 61 | leftImg8bit/train/bremen/bremen_000185_000019_leftImg8bit.png gtFine/train/bremen/bremen_000185_000019_gtFine_labelTrainIds.png
 62 | leftImg8bit/train/bremen/bremen_000114_000019_leftImg8bit.png gtFine/train/bremen/bremen_000114_000019_gtFine_labelTrainIds.png
 63 | leftImg8bit/train/bremen/bremen_000233_000019_leftImg8bit.png gtFine/train/bremen/bremen_000233_000019_gtFine_labelTrainIds.png
 64 | leftImg8bit/train/bremen/bremen_000273_000019_leftImg8bit.png gtFine/train/bremen/bremen_000273_000019_gtFine_labelTrainIds.png
 65 | leftImg8bit/train/bremen/bremen_000306_000019_leftImg8bit.png gtFine/train/bremen/bremen_000306_000019_gtFine_labelTrainIds.png
 66 | leftImg8bit/train/bremen/bremen_000187_000019_leftImg8bit.png gtFine/train/bremen/bremen_000187_000019_gtFine_labelTrainIds.png
 67 | leftImg8bit/train/bremen/bremen_000281_000019_leftImg8bit.png gtFine/train/bremen/bremen_000281_000019_gtFine_labelTrainIds.png
 68 | leftImg8bit/train/bremen/bremen_000080_000019_leftImg8bit.png gtFine/train/bremen/bremen_000080_000019_gtFine_labelTrainIds.png
 69 | leftImg8bit/train/bremen/bremen_000231_000019_leftImg8bit.png gtFine/train/bremen/bremen_000231_000019_gtFine_labelTrainIds.png
 70 | leftImg8bit/train/bremen/bremen_000052_000019_leftImg8bit.png gtFine/train/bremen/bremen_000052_000019_gtFine_labelTrainIds.png
 71 | leftImg8bit/train/bremen/bremen_000180_000019_leftImg8bit.png gtFine/train/bremen/bremen_000180_000019_gtFine_labelTrainIds.png
 72 | leftImg8bit/train/bremen/bremen_000121_000019_leftImg8bit.png gtFine/train/bremen/bremen_000121_000019_gtFine_labelTrainIds.png
 73 | leftImg8bit/train/bremen/bremen_000153_000019_leftImg8bit.png gtFine/train/bremen/bremen_000153_000019_gtFine_labelTrainIds.png
 74 | leftImg8bit/train/bochum/bochum_000000_022210_leftImg8bit.png gtFine/train/bochum/bochum_000000_022210_gtFine_labelTrainIds.png
 75 | leftImg8bit/train/bochum/bochum_000000_000313_leftImg8bit.png gtFine/train/bochum/bochum_000000_000313_gtFine_labelTrainIds.png
 76 | leftImg8bit/train/bochum/bochum_000000_027057_leftImg8bit.png gtFine/train/bochum/bochum_000000_027057_gtFine_labelTrainIds.png
 77 | leftImg8bit/train/bochum/bochum_000000_021606_leftImg8bit.png gtFine/train/bochum/bochum_000000_021606_gtFine_labelTrainIds.png
 78 | leftImg8bit/train/bochum/bochum_000000_028764_leftImg8bit.png gtFine/train/bochum/bochum_000000_028764_gtFine_labelTrainIds.png
 79 | leftImg8bit/train/bochum/bochum_000000_002562_leftImg8bit.png gtFine/train/bochum/bochum_000000_002562_gtFine_labelTrainIds.png
 80 | leftImg8bit/train/hamburg/hamburg_000000_093572_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_093572_gtFine_labelTrainIds.png
 81 | leftImg8bit/train/hamburg/hamburg_000000_000629_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_000629_gtFine_labelTrainIds.png
 82 | leftImg8bit/train/hamburg/hamburg_000000_067799_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_067799_gtFine_labelTrainIds.png
 83 | leftImg8bit/train/hamburg/hamburg_000000_044747_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_044747_gtFine_labelTrainIds.png
 84 | leftImg8bit/train/hamburg/hamburg_000000_088054_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_088054_gtFine_labelTrainIds.png
 85 | leftImg8bit/train/hamburg/hamburg_000000_079657_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_079657_gtFine_labelTrainIds.png
 86 | leftImg8bit/train/hamburg/hamburg_000000_046619_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_046619_gtFine_labelTrainIds.png
 87 | leftImg8bit/train/hamburg/hamburg_000000_006192_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_006192_gtFine_labelTrainIds.png
 88 | leftImg8bit/train/hamburg/hamburg_000000_074545_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_074545_gtFine_labelTrainIds.png
 89 | leftImg8bit/train/hamburg/hamburg_000000_046078_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_046078_gtFine_labelTrainIds.png
 90 | leftImg8bit/train/hamburg/hamburg_000000_037741_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_037741_gtFine_labelTrainIds.png
 91 | leftImg8bit/train/hamburg/hamburg_000000_069289_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_069289_gtFine_labelTrainIds.png
 92 | leftImg8bit/train/hamburg/hamburg_000000_088939_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_088939_gtFine_labelTrainIds.png
 93 | leftImg8bit/train/hamburg/hamburg_000000_014940_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_014940_gtFine_labelTrainIds.png
 94 | leftImg8bit/train/hamburg/hamburg_000000_016447_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_016447_gtFine_labelTrainIds.png
 95 | leftImg8bit/train/tubingen/tubingen_000104_000019_leftImg8bit.png gtFine/train/tubingen/tubingen_000104_000019_gtFine_labelTrainIds.png
 96 | leftImg8bit/train/tubingen/tubingen_000046_000019_leftImg8bit.png gtFine/train/tubingen/tubingen_000046_000019_gtFine_labelTrainIds.png
 97 | leftImg8bit/train/tubingen/tubingen_000130_000019_leftImg8bit.png gtFine/train/tubingen/tubingen_000130_000019_gtFine_labelTrainIds.png
 98 | leftImg8bit/train/tubingen/tubingen_000079_000019_leftImg8bit.png gtFine/train/tubingen/tubingen_000079_000019_gtFine_labelTrainIds.png
 99 | leftImg8bit/train/tubingen/tubingen_000063_000019_leftImg8bit.png gtFine/train/tubingen/tubingen_000063_000019_gtFine_labelTrainIds.png
100 | leftImg8bit/train/tubingen/tubingen_000087_000019_leftImg8bit.png gtFine/train/tubingen/tubingen_000087_000019_gtFine_labelTrainIds.png
101 | leftImg8bit/train/tubingen/tubingen_000061_000019_leftImg8bit.png gtFine/train/tubingen/tubingen_000061_000019_gtFine_labelTrainIds.png
102 | leftImg8bit/train/tubingen/tubingen_000074_000019_leftImg8bit.png gtFine/train/tubingen/tubingen_000074_000019_gtFine_labelTrainIds.png
103 | leftImg8bit/train/tubingen/tubingen_000017_000019_leftImg8bit.png gtFine/train/tubingen/tubingen_000017_000019_gtFine_labelTrainIds.png
104 | leftImg8bit/train/aachen/aachen_000142_000019_leftImg8bit.png gtFine/train/aachen/aachen_000142_000019_gtFine_labelTrainIds.png
105 | leftImg8bit/train/aachen/aachen_000029_000019_leftImg8bit.png gtFine/train/aachen/aachen_000029_000019_gtFine_labelTrainIds.png
106 | leftImg8bit/train/aachen/aachen_000141_000019_leftImg8bit.png gtFine/train/aachen/aachen_000141_000019_gtFine_labelTrainIds.png
107 | leftImg8bit/train/aachen/aachen_000136_000019_leftImg8bit.png gtFine/train/aachen/aachen_000136_000019_gtFine_labelTrainIds.png
108 | leftImg8bit/train/aachen/aachen_000098_000019_leftImg8bit.png gtFine/train/aachen/aachen_000098_000019_gtFine_labelTrainIds.png
109 | leftImg8bit/train/aachen/aachen_000080_000019_leftImg8bit.png gtFine/train/aachen/aachen_000080_000019_gtFine_labelTrainIds.png
110 | leftImg8bit/train/aachen/aachen_000157_000019_leftImg8bit.png gtFine/train/aachen/aachen_000157_000019_gtFine_labelTrainIds.png
111 | leftImg8bit/train/aachen/aachen_000169_000019_leftImg8bit.png gtFine/train/aachen/aachen_000169_000019_gtFine_labelTrainIds.png
112 | leftImg8bit/train/aachen/aachen_000147_000019_leftImg8bit.png gtFine/train/aachen/aachen_000147_000019_gtFine_labelTrainIds.png
113 | leftImg8bit/train/aachen/aachen_000002_000019_leftImg8bit.png gtFine/train/aachen/aachen_000002_000019_gtFine_labelTrainIds.png
114 | leftImg8bit/train/aachen/aachen_000007_000019_leftImg8bit.png gtFine/train/aachen/aachen_000007_000019_gtFine_labelTrainIds.png
115 | leftImg8bit/train/krefeld/krefeld_000000_023338_leftImg8bit.png gtFine/train/krefeld/krefeld_000000_023338_gtFine_labelTrainIds.png
116 | leftImg8bit/train/krefeld/krefeld_000000_030560_leftImg8bit.png gtFine/train/krefeld/krefeld_000000_030560_gtFine_labelTrainIds.png
117 | leftImg8bit/train/krefeld/krefeld_000000_035124_leftImg8bit.png gtFine/train/krefeld/krefeld_000000_035124_gtFine_labelTrainIds.png
118 | leftImg8bit/train/krefeld/krefeld_000000_000316_leftImg8bit.png gtFine/train/krefeld/krefeld_000000_000316_gtFine_labelTrainIds.png
119 | leftImg8bit/train/krefeld/krefeld_000000_007325_leftImg8bit.png gtFine/train/krefeld/krefeld_000000_007325_gtFine_labelTrainIds.png
120 | leftImg8bit/train/krefeld/krefeld_000000_023698_leftImg8bit.png gtFine/train/krefeld/krefeld_000000_023698_gtFine_labelTrainIds.png
121 | leftImg8bit/train/hanover/hanover_000000_044085_leftImg8bit.png gtFine/train/hanover/hanover_000000_044085_gtFine_labelTrainIds.png
122 | leftImg8bit/train/hanover/hanover_000000_046200_leftImg8bit.png gtFine/train/hanover/hanover_000000_046200_gtFine_labelTrainIds.png
123 | leftImg8bit/train/hanover/hanover_000000_000712_leftImg8bit.png gtFine/train/hanover/hanover_000000_000712_gtFine_labelTrainIds.png
124 | leftImg8bit/train/hanover/hanover_000000_052649_leftImg8bit.png gtFine/train/hanover/hanover_000000_052649_gtFine_labelTrainIds.png
125 | leftImg8bit/train/hanover/hanover_000000_035768_leftImg8bit.png gtFine/train/hanover/hanover_000000_035768_gtFine_labelTrainIds.png
126 | leftImg8bit/train/hanover/hanover_000000_043822_leftImg8bit.png gtFine/train/hanover/hanover_000000_043822_gtFine_labelTrainIds.png
127 | leftImg8bit/train/hanover/hanover_000000_049005_leftImg8bit.png gtFine/train/hanover/hanover_000000_049005_gtFine_labelTrainIds.png
128 | leftImg8bit/train/hanover/hanover_000000_014919_leftImg8bit.png gtFine/train/hanover/hanover_000000_014919_gtFine_labelTrainIds.png
129 | leftImg8bit/train/hanover/hanover_000000_023239_leftImg8bit.png gtFine/train/hanover/hanover_000000_023239_gtFine_labelTrainIds.png
130 | leftImg8bit/train/hanover/hanover_000000_007780_leftImg8bit.png gtFine/train/hanover/hanover_000000_007780_gtFine_labelTrainIds.png
131 | leftImg8bit/train/hanover/hanover_000000_055937_leftImg8bit.png gtFine/train/hanover/hanover_000000_055937_gtFine_labelTrainIds.png
132 | leftImg8bit/train/hanover/hanover_000000_002357_leftImg8bit.png gtFine/train/hanover/hanover_000000_002357_gtFine_labelTrainIds.png
133 | leftImg8bit/train/weimar/weimar_000077_000019_leftImg8bit.png gtFine/train/weimar/weimar_000077_000019_gtFine_labelTrainIds.png
134 | leftImg8bit/train/weimar/weimar_000091_000019_leftImg8bit.png gtFine/train/weimar/weimar_000091_000019_gtFine_labelTrainIds.png
135 | leftImg8bit/train/weimar/weimar_000034_000019_leftImg8bit.png gtFine/train/weimar/weimar_000034_000019_gtFine_labelTrainIds.png
136 | leftImg8bit/train/weimar/weimar_000117_000019_leftImg8bit.png gtFine/train/weimar/weimar_000117_000019_gtFine_labelTrainIds.png
137 | leftImg8bit/train/weimar/weimar_000019_000019_leftImg8bit.png gtFine/train/weimar/weimar_000019_000019_gtFine_labelTrainIds.png
138 | leftImg8bit/train/weimar/weimar_000125_000019_leftImg8bit.png gtFine/train/weimar/weimar_000125_000019_gtFine_labelTrainIds.png
139 | leftImg8bit/train/weimar/weimar_000067_000019_leftImg8bit.png gtFine/train/weimar/weimar_000067_000019_gtFine_labelTrainIds.png
140 | leftImg8bit/train/weimar/weimar_000012_000019_leftImg8bit.png gtFine/train/weimar/weimar_000012_000019_gtFine_labelTrainIds.png
141 | leftImg8bit/train/weimar/weimar_000103_000019_leftImg8bit.png gtFine/train/weimar/weimar_000103_000019_gtFine_labelTrainIds.png
142 | leftImg8bit/train/strasbourg/strasbourg_000001_054275_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_054275_gtFine_labelTrainIds.png
143 | leftImg8bit/train/strasbourg/strasbourg_000001_049399_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_049399_gtFine_labelTrainIds.png
144 | leftImg8bit/train/strasbourg/strasbourg_000001_002949_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_002949_gtFine_labelTrainIds.png
145 | leftImg8bit/train/strasbourg/strasbourg_000001_017540_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_017540_gtFine_labelTrainIds.png
146 | leftImg8bit/train/strasbourg/strasbourg_000000_021651_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_021651_gtFine_labelTrainIds.png
147 | leftImg8bit/train/strasbourg/strasbourg_000000_017159_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_017159_gtFine_labelTrainIds.png
148 | leftImg8bit/train/strasbourg/strasbourg_000001_031427_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_031427_gtFine_labelTrainIds.png
149 | leftImg8bit/train/strasbourg/strasbourg_000001_045481_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_045481_gtFine_labelTrainIds.png
150 | leftImg8bit/train/strasbourg/strasbourg_000001_002354_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_002354_gtFine_labelTrainIds.png
151 | leftImg8bit/train/strasbourg/strasbourg_000000_015131_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_015131_gtFine_labelTrainIds.png
152 | leftImg8bit/train/strasbourg/strasbourg_000001_036697_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_036697_gtFine_labelTrainIds.png
153 | leftImg8bit/train/strasbourg/strasbourg_000001_024379_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_024379_gtFine_labelTrainIds.png
154 | leftImg8bit/train/strasbourg/strasbourg_000000_019050_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_019050_gtFine_labelTrainIds.png
155 | leftImg8bit/train/strasbourg/strasbourg_000001_002519_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_002519_gtFine_labelTrainIds.png
156 | leftImg8bit/train/strasbourg/strasbourg_000000_011225_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_011225_gtFine_labelTrainIds.png
157 | leftImg8bit/train/strasbourg/strasbourg_000000_017593_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_017593_gtFine_labelTrainIds.png
158 | leftImg8bit/train/strasbourg/strasbourg_000001_016253_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_016253_gtFine_labelTrainIds.png
159 | leftImg8bit/train/strasbourg/strasbourg_000000_018153_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_018153_gtFine_labelTrainIds.png
160 | leftImg8bit/train/strasbourg/strasbourg_000000_008784_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_008784_gtFine_labelTrainIds.png
161 | leftImg8bit/train/strasbourg/strasbourg_000001_000710_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_000710_gtFine_labelTrainIds.png
162 | leftImg8bit/train/strasbourg/strasbourg_000000_025351_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_025351_gtFine_labelTrainIds.png
163 | leftImg8bit/train/strasbourg/strasbourg_000000_023694_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_023694_gtFine_labelTrainIds.png
164 | leftImg8bit/train/strasbourg/strasbourg_000001_000778_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_000778_gtFine_labelTrainIds.png
165 | leftImg8bit/train/cologne/cologne_000062_000019_leftImg8bit.png gtFine/train/cologne/cologne_000062_000019_gtFine_labelTrainIds.png
166 | leftImg8bit/train/cologne/cologne_000109_000019_leftImg8bit.png gtFine/train/cologne/cologne_000109_000019_gtFine_labelTrainIds.png
167 | leftImg8bit/train/cologne/cologne_000071_000019_leftImg8bit.png gtFine/train/cologne/cologne_000071_000019_gtFine_labelTrainIds.png
168 | leftImg8bit/train/cologne/cologne_000019_000019_leftImg8bit.png gtFine/train/cologne/cologne_000019_000019_gtFine_labelTrainIds.png
169 | leftImg8bit/train/cologne/cologne_000050_000019_leftImg8bit.png gtFine/train/cologne/cologne_000050_000019_gtFine_labelTrainIds.png
170 | leftImg8bit/train/cologne/cologne_000057_000019_leftImg8bit.png gtFine/train/cologne/cologne_000057_000019_gtFine_labelTrainIds.png
171 | leftImg8bit/train/cologne/cologne_000117_000019_leftImg8bit.png gtFine/train/cologne/cologne_000117_000019_gtFine_labelTrainIds.png
172 | leftImg8bit/train/cologne/cologne_000108_000019_leftImg8bit.png gtFine/train/cologne/cologne_000108_000019_gtFine_labelTrainIds.png
173 | leftImg8bit/train/cologne/cologne_000070_000019_leftImg8bit.png gtFine/train/cologne/cologne_000070_000019_gtFine_labelTrainIds.png
174 | leftImg8bit/train/cologne/cologne_000086_000019_leftImg8bit.png gtFine/train/cologne/cologne_000086_000019_gtFine_labelTrainIds.png
175 | leftImg8bit/train/erfurt/erfurt_000017_000019_leftImg8bit.png gtFine/train/erfurt/erfurt_000017_000019_gtFine_labelTrainIds.png
176 | leftImg8bit/train/erfurt/erfurt_000102_000019_leftImg8bit.png gtFine/train/erfurt/erfurt_000102_000019_gtFine_labelTrainIds.png
177 | leftImg8bit/train/erfurt/erfurt_000097_000019_leftImg8bit.png gtFine/train/erfurt/erfurt_000097_000019_gtFine_labelTrainIds.png
178 | leftImg8bit/train/erfurt/erfurt_000005_000019_leftImg8bit.png gtFine/train/erfurt/erfurt_000005_000019_gtFine_labelTrainIds.png
179 | leftImg8bit/train/erfurt/erfurt_000049_000019_leftImg8bit.png gtFine/train/erfurt/erfurt_000049_000019_gtFine_labelTrainIds.png
180 | leftImg8bit/train/erfurt/erfurt_000096_000019_leftImg8bit.png gtFine/train/erfurt/erfurt_000096_000019_gtFine_labelTrainIds.png
181 | leftImg8bit/train/erfurt/erfurt_000093_000019_leftImg8bit.png gtFine/train/erfurt/erfurt_000093_000019_gtFine_labelTrainIds.png
182 | leftImg8bit/train/monchengladbach/monchengladbach_000000_002478_leftImg8bit.png gtFine/train/monchengladbach/monchengladbach_000000_002478_gtFine_labelTrainIds.png
183 | leftImg8bit/train/monchengladbach/monchengladbach_000000_007098_leftImg8bit.png gtFine/train/monchengladbach/monchengladbach_000000_007098_gtFine_labelTrainIds.png
184 | leftImg8bit/train/monchengladbach/monchengladbach_000001_000054_leftImg8bit.png gtFine/train/monchengladbach/monchengladbach_000001_000054_gtFine_labelTrainIds.png
185 | leftImg8bit/train/monchengladbach/monchengladbach_000000_013352_leftImg8bit.png gtFine/train/monchengladbach/monchengladbach_000000_013352_gtFine_labelTrainIds.png
186 | leftImg8bit/train/monchengladbach/monchengladbach_000000_018720_leftImg8bit.png gtFine/train/monchengladbach/monchengladbach_000000_018720_gtFine_labelTrainIds.png
187 | 


--------------------------------------------------------------------------------
/partitions/cityscapes/1_30/labeled.txt:
--------------------------------------------------------------------------------
  1 | leftImg8bit/train/weimar/weimar_000085_000019_leftImg8bit.png gtFine/train/weimar/weimar_000085_000019_gtFine_labelTrainIds.png
  2 | leftImg8bit/train/bochum/bochum_000000_021070_leftImg8bit.png gtFine/train/bochum/bochum_000000_021070_gtFine_labelTrainIds.png
  3 | leftImg8bit/train/bremen/bremen_000128_000019_leftImg8bit.png gtFine/train/bremen/bremen_000128_000019_gtFine_labelTrainIds.png
  4 | leftImg8bit/train/stuttgart/stuttgart_000119_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000119_000019_gtFine_labelTrainIds.png
  5 | leftImg8bit/train/hamburg/hamburg_000000_088783_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_088783_gtFine_labelTrainIds.png
  6 | leftImg8bit/train/bremen/bremen_000307_000019_leftImg8bit.png gtFine/train/bremen/bremen_000307_000019_gtFine_labelTrainIds.png
  7 | leftImg8bit/train/strasbourg/strasbourg_000001_014258_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_014258_gtFine_labelTrainIds.png
  8 | leftImg8bit/train/aachen/aachen_000093_000019_leftImg8bit.png gtFine/train/aachen/aachen_000093_000019_gtFine_labelTrainIds.png
  9 | leftImg8bit/train/zurich/zurich_000078_000019_leftImg8bit.png gtFine/train/zurich/zurich_000078_000019_gtFine_labelTrainIds.png
 10 | leftImg8bit/train/ulm/ulm_000034_000019_leftImg8bit.png gtFine/train/ulm/ulm_000034_000019_gtFine_labelTrainIds.png
 11 | leftImg8bit/train/monchengladbach/monchengladbach_000000_017950_leftImg8bit.png gtFine/train/monchengladbach/monchengladbach_000000_017950_gtFine_labelTrainIds.png
 12 | leftImg8bit/train/tubingen/tubingen_000111_000019_leftImg8bit.png gtFine/train/tubingen/tubingen_000111_000019_gtFine_labelTrainIds.png
 13 | leftImg8bit/train/weimar/weimar_000091_000019_leftImg8bit.png gtFine/train/weimar/weimar_000091_000019_gtFine_labelTrainIds.png
 14 | leftImg8bit/train/strasbourg/strasbourg_000000_034652_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_034652_gtFine_labelTrainIds.png
 15 | leftImg8bit/train/weimar/weimar_000071_000019_leftImg8bit.png gtFine/train/weimar/weimar_000071_000019_gtFine_labelTrainIds.png
 16 | leftImg8bit/train/bremen/bremen_000181_000019_leftImg8bit.png gtFine/train/bremen/bremen_000181_000019_gtFine_labelTrainIds.png
 17 | leftImg8bit/train/aachen/aachen_000010_000019_leftImg8bit.png gtFine/train/aachen/aachen_000010_000019_gtFine_labelTrainIds.png
 18 | leftImg8bit/train/krefeld/krefeld_000000_003096_leftImg8bit.png gtFine/train/krefeld/krefeld_000000_003096_gtFine_labelTrainIds.png
 19 | leftImg8bit/train/monchengladbach/monchengladbach_000000_020303_leftImg8bit.png gtFine/train/monchengladbach/monchengladbach_000000_020303_gtFine_labelTrainIds.png
 20 | leftImg8bit/train/strasbourg/strasbourg_000001_026355_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_026355_gtFine_labelTrainIds.png
 21 | leftImg8bit/train/aachen/aachen_000005_000019_leftImg8bit.png gtFine/train/aachen/aachen_000005_000019_gtFine_labelTrainIds.png
 22 | leftImg8bit/train/hamburg/hamburg_000000_090742_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_090742_gtFine_labelTrainIds.png
 23 | leftImg8bit/train/stuttgart/stuttgart_000028_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000028_000019_gtFine_labelTrainIds.png
 24 | leftImg8bit/train/strasbourg/strasbourg_000001_003676_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_003676_gtFine_labelTrainIds.png
 25 | leftImg8bit/train/dusseldorf/dusseldorf_000214_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000214_000019_gtFine_labelTrainIds.png
 26 | leftImg8bit/train/dusseldorf/dusseldorf_000212_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000212_000019_gtFine_labelTrainIds.png
 27 | leftImg8bit/train/ulm/ulm_000035_000019_leftImg8bit.png gtFine/train/ulm/ulm_000035_000019_gtFine_labelTrainIds.png
 28 | leftImg8bit/train/tubingen/tubingen_000036_000019_leftImg8bit.png gtFine/train/tubingen/tubingen_000036_000019_gtFine_labelTrainIds.png
 29 | leftImg8bit/train/strasbourg/strasbourg_000000_014101_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_014101_gtFine_labelTrainIds.png
 30 | leftImg8bit/train/erfurt/erfurt_000108_000019_leftImg8bit.png gtFine/train/erfurt/erfurt_000108_000019_gtFine_labelTrainIds.png
 31 | leftImg8bit/train/cologne/cologne_000029_000019_leftImg8bit.png gtFine/train/cologne/cologne_000029_000019_gtFine_labelTrainIds.png
 32 | leftImg8bit/train/ulm/ulm_000007_000019_leftImg8bit.png gtFine/train/ulm/ulm_000007_000019_gtFine_labelTrainIds.png
 33 | leftImg8bit/train/erfurt/erfurt_000095_000019_leftImg8bit.png gtFine/train/erfurt/erfurt_000095_000019_gtFine_labelTrainIds.png
 34 | leftImg8bit/train/tubingen/tubingen_000038_000019_leftImg8bit.png gtFine/train/tubingen/tubingen_000038_000019_gtFine_labelTrainIds.png
 35 | leftImg8bit/train/strasbourg/strasbourg_000000_021651_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_021651_gtFine_labelTrainIds.png
 36 | leftImg8bit/train/aachen/aachen_000042_000019_leftImg8bit.png gtFine/train/aachen/aachen_000042_000019_gtFine_labelTrainIds.png
 37 | leftImg8bit/train/stuttgart/stuttgart_000125_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000125_000019_gtFine_labelTrainIds.png
 38 | leftImg8bit/train/hamburg/hamburg_000000_098862_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_098862_gtFine_labelTrainIds.png
 39 | leftImg8bit/train/dusseldorf/dusseldorf_000002_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000002_000019_gtFine_labelTrainIds.png
 40 | leftImg8bit/train/krefeld/krefeld_000000_019697_leftImg8bit.png gtFine/train/krefeld/krefeld_000000_019697_gtFine_labelTrainIds.png
 41 | leftImg8bit/train/dusseldorf/dusseldorf_000081_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000081_000019_gtFine_labelTrainIds.png
 42 | leftImg8bit/train/erfurt/erfurt_000085_000019_leftImg8bit.png gtFine/train/erfurt/erfurt_000085_000019_gtFine_labelTrainIds.png
 43 | leftImg8bit/train/jena/jena_000085_000019_leftImg8bit.png gtFine/train/jena/jena_000085_000019_gtFine_labelTrainIds.png
 44 | leftImg8bit/train/hamburg/hamburg_000000_042885_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_042885_gtFine_labelTrainIds.png
 45 | leftImg8bit/train/bremen/bremen_000080_000019_leftImg8bit.png gtFine/train/bremen/bremen_000080_000019_gtFine_labelTrainIds.png
 46 | leftImg8bit/train/dusseldorf/dusseldorf_000105_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000105_000019_gtFine_labelTrainIds.png
 47 | leftImg8bit/train/strasbourg/strasbourg_000001_003991_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_003991_gtFine_labelTrainIds.png
 48 | leftImg8bit/train/weimar/weimar_000025_000019_leftImg8bit.png gtFine/train/weimar/weimar_000025_000019_gtFine_labelTrainIds.png
 49 | leftImg8bit/train/strasbourg/strasbourg_000000_034040_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_034040_gtFine_labelTrainIds.png
 50 | leftImg8bit/train/hanover/hanover_000000_034141_leftImg8bit.png gtFine/train/hanover/hanover_000000_034141_gtFine_labelTrainIds.png
 51 | leftImg8bit/train/strasbourg/strasbourg_000001_022836_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_022836_gtFine_labelTrainIds.png
 52 | leftImg8bit/train/krefeld/krefeld_000000_024604_leftImg8bit.png gtFine/train/krefeld/krefeld_000000_024604_gtFine_labelTrainIds.png
 53 | leftImg8bit/train/cologne/cologne_000055_000019_leftImg8bit.png gtFine/train/cologne/cologne_000055_000019_gtFine_labelTrainIds.png
 54 | leftImg8bit/train/stuttgart/stuttgart_000118_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000118_000019_gtFine_labelTrainIds.png
 55 | leftImg8bit/train/dusseldorf/dusseldorf_000101_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000101_000019_gtFine_labelTrainIds.png
 56 | leftImg8bit/train/ulm/ulm_000074_000019_leftImg8bit.png gtFine/train/ulm/ulm_000074_000019_gtFine_labelTrainIds.png
 57 | leftImg8bit/train/jena/jena_000000_000019_leftImg8bit.png gtFine/train/jena/jena_000000_000019_gtFine_labelTrainIds.png
 58 | leftImg8bit/train/bremen/bremen_000048_000019_leftImg8bit.png gtFine/train/bremen/bremen_000048_000019_gtFine_labelTrainIds.png
 59 | leftImg8bit/train/cologne/cologne_000117_000019_leftImg8bit.png gtFine/train/cologne/cologne_000117_000019_gtFine_labelTrainIds.png
 60 | leftImg8bit/train/dusseldorf/dusseldorf_000171_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000171_000019_gtFine_labelTrainIds.png
 61 | leftImg8bit/train/dusseldorf/dusseldorf_000197_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000197_000019_gtFine_labelTrainIds.png
 62 | leftImg8bit/train/tubingen/tubingen_000142_000019_leftImg8bit.png gtFine/train/tubingen/tubingen_000142_000019_gtFine_labelTrainIds.png
 63 | leftImg8bit/train/monchengladbach/monchengladbach_000000_009690_leftImg8bit.png gtFine/train/monchengladbach/monchengladbach_000000_009690_gtFine_labelTrainIds.png
 64 | leftImg8bit/train/cologne/cologne_000014_000019_leftImg8bit.png gtFine/train/cologne/cologne_000014_000019_gtFine_labelTrainIds.png
 65 | leftImg8bit/train/aachen/aachen_000036_000019_leftImg8bit.png gtFine/train/aachen/aachen_000036_000019_gtFine_labelTrainIds.png
 66 | leftImg8bit/train/strasbourg/strasbourg_000000_015602_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_015602_gtFine_labelTrainIds.png
 67 | leftImg8bit/train/aachen/aachen_000143_000019_leftImg8bit.png gtFine/train/aachen/aachen_000143_000019_gtFine_labelTrainIds.png
 68 | leftImg8bit/train/hamburg/hamburg_000000_062039_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_062039_gtFine_labelTrainIds.png
 69 | leftImg8bit/train/stuttgart/stuttgart_000145_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000145_000019_gtFine_labelTrainIds.png
 70 | leftImg8bit/train/bremen/bremen_000297_000019_leftImg8bit.png gtFine/train/bremen/bremen_000297_000019_gtFine_labelTrainIds.png
 71 | leftImg8bit/train/zurich/zurich_000027_000019_leftImg8bit.png gtFine/train/zurich/zurich_000027_000019_gtFine_labelTrainIds.png
 72 | leftImg8bit/train/aachen/aachen_000024_000019_leftImg8bit.png gtFine/train/aachen/aachen_000024_000019_gtFine_labelTrainIds.png
 73 | leftImg8bit/train/cologne/cologne_000011_000019_leftImg8bit.png gtFine/train/cologne/cologne_000011_000019_gtFine_labelTrainIds.png
 74 | leftImg8bit/train/hanover/hanover_000000_029769_leftImg8bit.png gtFine/train/hanover/hanover_000000_029769_gtFine_labelTrainIds.png
 75 | leftImg8bit/train/ulm/ulm_000021_000019_leftImg8bit.png gtFine/train/ulm/ulm_000021_000019_gtFine_labelTrainIds.png
 76 | leftImg8bit/train/monchengladbach/monchengladbach_000000_006169_leftImg8bit.png gtFine/train/monchengladbach/monchengladbach_000000_006169_gtFine_labelTrainIds.png
 77 | leftImg8bit/train/darmstadt/darmstadt_000035_000019_leftImg8bit.png gtFine/train/darmstadt/darmstadt_000035_000019_gtFine_labelTrainIds.png
 78 | leftImg8bit/train/krefeld/krefeld_000000_022162_leftImg8bit.png gtFine/train/krefeld/krefeld_000000_022162_gtFine_labelTrainIds.png
 79 | leftImg8bit/train/bremen/bremen_000263_000019_leftImg8bit.png gtFine/train/bremen/bremen_000263_000019_gtFine_labelTrainIds.png
 80 | leftImg8bit/train/krefeld/krefeld_000000_015868_leftImg8bit.png gtFine/train/krefeld/krefeld_000000_015868_gtFine_labelTrainIds.png
 81 | leftImg8bit/train/darmstadt/darmstadt_000023_000019_leftImg8bit.png gtFine/train/darmstadt/darmstadt_000023_000019_gtFine_labelTrainIds.png
 82 | leftImg8bit/train/aachen/aachen_000049_000019_leftImg8bit.png gtFine/train/aachen/aachen_000049_000019_gtFine_labelTrainIds.png
 83 | leftImg8bit/train/hamburg/hamburg_000000_105464_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_105464_gtFine_labelTrainIds.png
 84 | leftImg8bit/train/strasbourg/strasbourg_000000_035571_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000000_035571_gtFine_labelTrainIds.png
 85 | leftImg8bit/train/bochum/bochum_000000_027951_leftImg8bit.png gtFine/train/bochum/bochum_000000_027951_gtFine_labelTrainIds.png
 86 | leftImg8bit/train/bremen/bremen_000231_000019_leftImg8bit.png gtFine/train/bremen/bremen_000231_000019_gtFine_labelTrainIds.png
 87 | leftImg8bit/train/strasbourg/strasbourg_000001_042309_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_042309_gtFine_labelTrainIds.png
 88 | leftImg8bit/train/dusseldorf/dusseldorf_000155_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000155_000019_gtFine_labelTrainIds.png
 89 | leftImg8bit/train/dusseldorf/dusseldorf_000185_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000185_000019_gtFine_labelTrainIds.png
 90 | leftImg8bit/train/aachen/aachen_000164_000019_leftImg8bit.png gtFine/train/aachen/aachen_000164_000019_gtFine_labelTrainIds.png
 91 | leftImg8bit/train/hanover/hanover_000000_043822_leftImg8bit.png gtFine/train/hanover/hanover_000000_043822_gtFine_labelTrainIds.png
 92 | leftImg8bit/train/hanover/hanover_000000_014713_leftImg8bit.png gtFine/train/hanover/hanover_000000_014713_gtFine_labelTrainIds.png
 93 | leftImg8bit/train/dusseldorf/dusseldorf_000110_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000110_000019_gtFine_labelTrainIds.png
 94 | leftImg8bit/train/stuttgart/stuttgart_000034_000019_leftImg8bit.png gtFine/train/stuttgart/stuttgart_000034_000019_gtFine_labelTrainIds.png
 95 | leftImg8bit/train/strasbourg/strasbourg_000001_052297_leftImg8bit.png gtFine/train/strasbourg/strasbourg_000001_052297_gtFine_labelTrainIds.png
 96 | leftImg8bit/train/hamburg/hamburg_000000_065983_leftImg8bit.png gtFine/train/hamburg/hamburg_000000_065983_gtFine_labelTrainIds.png
 97 | leftImg8bit/train/dusseldorf/dusseldorf_000133_000019_leftImg8bit.png gtFine/train/dusseldorf/dusseldorf_000133_000019_gtFine_labelTrainIds.png
 98 | leftImg8bit/train/ulm/ulm_000044_000019_leftImg8bit.png gtFine/train/ulm/ulm_000044_000019_gtFine_labelTrainIds.png
 99 | leftImg8bit/train/jena/jena_000090_000019_leftImg8bit.png gtFine/train/jena/jena_000090_000019_gtFine_labelTrainIds.png
100 | leftImg8bit/train/bremen/bremen_000177_000019_leftImg8bit.png gtFine/train/bremen/bremen_000177_000019_gtFine_labelTrainIds.png


--------------------------------------------------------------------------------
/partitions/coco/1_256/labeled.txt:
--------------------------------------------------------------------------------
  1 | train2017/000000539326.jpg masks/000000539326.png
  2 | train2017/000000436349.jpg masks/000000436349.png
  3 | train2017/000000153931.jpg masks/000000153931.png
  4 | train2017/000000244816.jpg masks/000000244816.png
  5 | train2017/000000382263.jpg masks/000000382263.png
  6 | train2017/000000326531.jpg masks/000000326531.png
  7 | train2017/000000546702.jpg masks/000000546702.png
  8 | train2017/000000023501.jpg masks/000000023501.png
  9 | train2017/000000058731.jpg masks/000000058731.png
 10 | train2017/000000071215.jpg masks/000000071215.png
 11 | train2017/000000432762.jpg masks/000000432762.png
 12 | train2017/000000475711.jpg masks/000000475711.png
 13 | train2017/000000462401.jpg masks/000000462401.png
 14 | train2017/000000257632.jpg masks/000000257632.png
 15 | train2017/000000445433.jpg masks/000000445433.png
 16 | train2017/000000218459.jpg masks/000000218459.png
 17 | train2017/000000383893.jpg masks/000000383893.png
 18 | train2017/000000151733.jpg masks/000000151733.png
 19 | train2017/000000535766.jpg masks/000000535766.png
 20 | train2017/000000487018.jpg masks/000000487018.png
 21 | train2017/000000044432.jpg masks/000000044432.png
 22 | train2017/000000428686.jpg masks/000000428686.png
 23 | train2017/000000157036.jpg masks/000000157036.png
 24 | train2017/000000352980.jpg masks/000000352980.png
 25 | train2017/000000105936.jpg masks/000000105936.png
 26 | train2017/000000491453.jpg masks/000000491453.png
 27 | train2017/000000302467.jpg masks/000000302467.png
 28 | train2017/000000133532.jpg masks/000000133532.png
 29 | train2017/000000262508.jpg masks/000000262508.png
 30 | train2017/000000125160.jpg masks/000000125160.png
 31 | train2017/000000461340.jpg masks/000000461340.png
 32 | train2017/000000472472.jpg masks/000000472472.png
 33 | train2017/000000223751.jpg masks/000000223751.png
 34 | train2017/000000431241.jpg masks/000000431241.png
 35 | train2017/000000268400.jpg masks/000000268400.png
 36 | train2017/000000376843.jpg masks/000000376843.png
 37 | train2017/000000338427.jpg masks/000000338427.png
 38 | train2017/000000538596.jpg masks/000000538596.png
 39 | train2017/000000132993.jpg masks/000000132993.png
 40 | train2017/000000037401.jpg masks/000000037401.png
 41 | train2017/000000310808.jpg masks/000000310808.png
 42 | train2017/000000298799.jpg masks/000000298799.png
 43 | train2017/000000465625.jpg masks/000000465625.png
 44 | train2017/000000178637.jpg masks/000000178637.png
 45 | train2017/000000298008.jpg masks/000000298008.png
 46 | train2017/000000295480.jpg masks/000000295480.png
 47 | train2017/000000002402.jpg masks/000000002402.png
 48 | train2017/000000103642.jpg masks/000000103642.png
 49 | train2017/000000049445.jpg masks/000000049445.png
 50 | train2017/000000379211.jpg masks/000000379211.png
 51 | train2017/000000210580.jpg masks/000000210580.png
 52 | train2017/000000177011.jpg masks/000000177011.png
 53 | train2017/000000156503.jpg masks/000000156503.png
 54 | train2017/000000149250.jpg masks/000000149250.png
 55 | train2017/000000130538.jpg masks/000000130538.png
 56 | train2017/000000182403.jpg masks/000000182403.png
 57 | train2017/000000310416.jpg masks/000000310416.png
 58 | train2017/000000292009.jpg masks/000000292009.png
 59 | train2017/000000093793.jpg masks/000000093793.png
 60 | train2017/000000265646.jpg masks/000000265646.png
 61 | train2017/000000385194.jpg masks/000000385194.png
 62 | train2017/000000575355.jpg masks/000000575355.png
 63 | train2017/000000322478.jpg masks/000000322478.png
 64 | train2017/000000451219.jpg masks/000000451219.png
 65 | train2017/000000114414.jpg masks/000000114414.png
 66 | train2017/000000068381.jpg masks/000000068381.png
 67 | train2017/000000570115.jpg masks/000000570115.png
 68 | train2017/000000287578.jpg masks/000000287578.png
 69 | train2017/000000366795.jpg masks/000000366795.png
 70 | train2017/000000575391.jpg masks/000000575391.png
 71 | train2017/000000292688.jpg masks/000000292688.png
 72 | train2017/000000265051.jpg masks/000000265051.png
 73 | train2017/000000029085.jpg masks/000000029085.png
 74 | train2017/000000541478.jpg masks/000000541478.png
 75 | train2017/000000056509.jpg masks/000000056509.png
 76 | train2017/000000472749.jpg masks/000000472749.png
 77 | train2017/000000281564.jpg masks/000000281564.png
 78 | train2017/000000052250.jpg masks/000000052250.png
 79 | train2017/000000265315.jpg masks/000000265315.png
 80 | train2017/000000055567.jpg masks/000000055567.png
 81 | train2017/000000433667.jpg masks/000000433667.png
 82 | train2017/000000506311.jpg masks/000000506311.png
 83 | train2017/000000041840.jpg masks/000000041840.png
 84 | train2017/000000360540.jpg masks/000000360540.png
 85 | train2017/000000219904.jpg masks/000000219904.png
 86 | train2017/000000447087.jpg masks/000000447087.png
 87 | train2017/000000263779.jpg masks/000000263779.png
 88 | train2017/000000218405.jpg masks/000000218405.png
 89 | train2017/000000267055.jpg masks/000000267055.png
 90 | train2017/000000080406.jpg masks/000000080406.png
 91 | train2017/000000323374.jpg masks/000000323374.png
 92 | train2017/000000108589.jpg masks/000000108589.png
 93 | train2017/000000297204.jpg masks/000000297204.png
 94 | train2017/000000217323.jpg masks/000000217323.png
 95 | train2017/000000509588.jpg masks/000000509588.png
 96 | train2017/000000477491.jpg masks/000000477491.png
 97 | train2017/000000156963.jpg masks/000000156963.png
 98 | train2017/000000138789.jpg masks/000000138789.png
 99 | train2017/000000532625.jpg masks/000000532625.png
100 | train2017/000000365164.jpg masks/000000365164.png
101 | train2017/000000139758.jpg masks/000000139758.png
102 | train2017/000000089027.jpg masks/000000089027.png
103 | train2017/000000081966.jpg masks/000000081966.png
104 | train2017/000000482472.jpg masks/000000482472.png
105 | train2017/000000291477.jpg masks/000000291477.png
106 | train2017/000000222305.jpg masks/000000222305.png
107 | train2017/000000188606.jpg masks/000000188606.png
108 | train2017/000000278260.jpg masks/000000278260.png
109 | train2017/000000133990.jpg masks/000000133990.png
110 | train2017/000000500646.jpg masks/000000500646.png
111 | train2017/000000391170.jpg masks/000000391170.png
112 | train2017/000000229388.jpg masks/000000229388.png
113 | train2017/000000008066.jpg masks/000000008066.png
114 | train2017/000000232723.jpg masks/000000232723.png
115 | train2017/000000038858.jpg masks/000000038858.png
116 | train2017/000000431597.jpg masks/000000431597.png
117 | train2017/000000528764.jpg masks/000000528764.png
118 | train2017/000000308905.jpg masks/000000308905.png
119 | train2017/000000008055.jpg masks/000000008055.png
120 | train2017/000000152436.jpg masks/000000152436.png
121 | train2017/000000102862.jpg masks/000000102862.png
122 | train2017/000000184211.jpg masks/000000184211.png
123 | train2017/000000392683.jpg masks/000000392683.png
124 | train2017/000000362079.jpg masks/000000362079.png
125 | train2017/000000495756.jpg masks/000000495756.png
126 | train2017/000000435109.jpg masks/000000435109.png
127 | train2017/000000376393.jpg masks/000000376393.png
128 | train2017/000000378055.jpg masks/000000378055.png
129 | train2017/000000398289.jpg masks/000000398289.png
130 | train2017/000000490437.jpg masks/000000490437.png
131 | train2017/000000338224.jpg masks/000000338224.png
132 | train2017/000000413044.jpg masks/000000413044.png
133 | train2017/000000537349.jpg masks/000000537349.png
134 | train2017/000000501059.jpg masks/000000501059.png
135 | train2017/000000469467.jpg masks/000000469467.png
136 | train2017/000000187156.jpg masks/000000187156.png
137 | train2017/000000188787.jpg masks/000000188787.png
138 | train2017/000000336129.jpg masks/000000336129.png
139 | train2017/000000020334.jpg masks/000000020334.png
140 | train2017/000000349934.jpg masks/000000349934.png
141 | train2017/000000416625.jpg masks/000000416625.png
142 | train2017/000000189391.jpg masks/000000189391.png
143 | train2017/000000001622.jpg masks/000000001622.png
144 | train2017/000000367804.jpg masks/000000367804.png
145 | train2017/000000116503.jpg masks/000000116503.png
146 | train2017/000000090516.jpg masks/000000090516.png
147 | train2017/000000426849.jpg masks/000000426849.png
148 | train2017/000000528516.jpg masks/000000528516.png
149 | train2017/000000385339.jpg masks/000000385339.png
150 | train2017/000000052543.jpg masks/000000052543.png
151 | train2017/000000411695.jpg masks/000000411695.png
152 | train2017/000000547742.jpg masks/000000547742.png
153 | train2017/000000407509.jpg masks/000000407509.png
154 | train2017/000000313733.jpg masks/000000313733.png
155 | train2017/000000298372.jpg masks/000000298372.png
156 | train2017/000000239196.jpg masks/000000239196.png
157 | train2017/000000371950.jpg masks/000000371950.png
158 | train2017/000000349559.jpg masks/000000349559.png
159 | train2017/000000564682.jpg masks/000000564682.png
160 | train2017/000000112536.jpg masks/000000112536.png
161 | train2017/000000113977.jpg masks/000000113977.png
162 | train2017/000000256903.jpg masks/000000256903.png
163 | train2017/000000208955.jpg masks/000000208955.png
164 | train2017/000000497259.jpg masks/000000497259.png
165 | train2017/000000171179.jpg masks/000000171179.png
166 | train2017/000000097734.jpg masks/000000097734.png
167 | train2017/000000253838.jpg masks/000000253838.png
168 | train2017/000000501995.jpg masks/000000501995.png
169 | train2017/000000032655.jpg masks/000000032655.png
170 | train2017/000000483794.jpg masks/000000483794.png
171 | train2017/000000409862.jpg masks/000000409862.png
172 | train2017/000000243090.jpg masks/000000243090.png
173 | train2017/000000034855.jpg masks/000000034855.png
174 | train2017/000000081605.jpg masks/000000081605.png
175 | train2017/000000032458.jpg masks/000000032458.png
176 | train2017/000000565951.jpg masks/000000565951.png
177 | train2017/000000114424.jpg masks/000000114424.png
178 | train2017/000000353148.jpg masks/000000353148.png
179 | train2017/000000310843.jpg masks/000000310843.png
180 | train2017/000000349154.jpg masks/000000349154.png
181 | train2017/000000449238.jpg masks/000000449238.png
182 | train2017/000000399269.jpg masks/000000399269.png
183 | train2017/000000464430.jpg masks/000000464430.png
184 | train2017/000000161823.jpg masks/000000161823.png
185 | train2017/000000508750.jpg masks/000000508750.png
186 | train2017/000000195180.jpg masks/000000195180.png
187 | train2017/000000064474.jpg masks/000000064474.png
188 | train2017/000000462134.jpg masks/000000462134.png
189 | train2017/000000281678.jpg masks/000000281678.png
190 | train2017/000000237064.jpg masks/000000237064.png
191 | train2017/000000545329.jpg masks/000000545329.png
192 | train2017/000000335085.jpg masks/000000335085.png
193 | train2017/000000449879.jpg masks/000000449879.png
194 | train2017/000000126563.jpg masks/000000126563.png
195 | train2017/000000313553.jpg masks/000000313553.png
196 | train2017/000000439465.jpg masks/000000439465.png
197 | train2017/000000157184.jpg masks/000000157184.png
198 | train2017/000000117131.jpg masks/000000117131.png
199 | train2017/000000205460.jpg masks/000000205460.png
200 | train2017/000000112996.jpg masks/000000112996.png
201 | train2017/000000228664.jpg masks/000000228664.png
202 | train2017/000000053275.jpg masks/000000053275.png
203 | train2017/000000375321.jpg masks/000000375321.png
204 | train2017/000000318723.jpg masks/000000318723.png
205 | train2017/000000197584.jpg masks/000000197584.png
206 | train2017/000000247925.jpg masks/000000247925.png
207 | train2017/000000391651.jpg masks/000000391651.png
208 | train2017/000000286018.jpg masks/000000286018.png
209 | train2017/000000432174.jpg masks/000000432174.png
210 | train2017/000000156809.jpg masks/000000156809.png
211 | train2017/000000386847.jpg masks/000000386847.png
212 | train2017/000000353124.jpg masks/000000353124.png
213 | train2017/000000015107.jpg masks/000000015107.png
214 | train2017/000000512458.jpg masks/000000512458.png
215 | train2017/000000518287.jpg masks/000000518287.png
216 | train2017/000000244671.jpg masks/000000244671.png
217 | train2017/000000105689.jpg masks/000000105689.png
218 | train2017/000000259755.jpg masks/000000259755.png
219 | train2017/000000196715.jpg masks/000000196715.png
220 | train2017/000000273378.jpg masks/000000273378.png
221 | train2017/000000559470.jpg masks/000000559470.png
222 | train2017/000000132306.jpg masks/000000132306.png
223 | train2017/000000287574.jpg masks/000000287574.png
224 | train2017/000000139355.jpg masks/000000139355.png
225 | train2017/000000007167.jpg masks/000000007167.png
226 | train2017/000000448064.jpg masks/000000448064.png
227 | train2017/000000008190.jpg masks/000000008190.png
228 | train2017/000000204775.jpg masks/000000204775.png
229 | train2017/000000014337.jpg masks/000000014337.png
230 | train2017/000000446075.jpg masks/000000446075.png
231 | train2017/000000085338.jpg masks/000000085338.png
232 | train2017/000000527338.jpg masks/000000527338.png
233 | train2017/000000569591.jpg masks/000000569591.png
234 | train2017/000000253773.jpg masks/000000253773.png
235 | train2017/000000557556.jpg masks/000000557556.png
236 | train2017/000000002470.jpg masks/000000002470.png
237 | train2017/000000055559.jpg masks/000000055559.png
238 | train2017/000000516823.jpg masks/000000516823.png
239 | train2017/000000507516.jpg masks/000000507516.png
240 | train2017/000000251474.jpg masks/000000251474.png
241 | train2017/000000377570.jpg masks/000000377570.png
242 | train2017/000000003751.jpg masks/000000003751.png
243 | train2017/000000327623.jpg masks/000000327623.png
244 | train2017/000000282293.jpg masks/000000282293.png
245 | train2017/000000577520.jpg masks/000000577520.png
246 | train2017/000000132094.jpg masks/000000132094.png
247 | train2017/000000402172.jpg masks/000000402172.png
248 | train2017/000000486580.jpg masks/000000486580.png
249 | train2017/000000478287.jpg masks/000000478287.png
250 | train2017/000000272673.jpg masks/000000272673.png
251 | train2017/000000410023.jpg masks/000000410023.png
252 | train2017/000000281625.jpg masks/000000281625.png
253 | train2017/000000356456.jpg masks/000000356456.png
254 | train2017/000000368564.jpg masks/000000368564.png
255 | train2017/000000407387.jpg masks/000000407387.png
256 | train2017/000000064264.jpg masks/000000064264.png
257 | train2017/000000438346.jpg masks/000000438346.png
258 | train2017/000000171729.jpg masks/000000171729.png
259 | train2017/000000460413.jpg masks/000000460413.png
260 | train2017/000000005210.jpg masks/000000005210.png
261 | train2017/000000394245.jpg masks/000000394245.png
262 | train2017/000000218302.jpg masks/000000218302.png
263 | train2017/000000154847.jpg masks/000000154847.png
264 | train2017/000000022816.jpg masks/000000022816.png
265 | train2017/000000469870.jpg masks/000000469870.png
266 | train2017/000000023121.jpg masks/000000023121.png
267 | train2017/000000309454.jpg masks/000000309454.png
268 | train2017/000000252604.jpg masks/000000252604.png
269 | train2017/000000504733.jpg masks/000000504733.png
270 | train2017/000000490860.jpg masks/000000490860.png
271 | train2017/000000053150.jpg masks/000000053150.png
272 | train2017/000000422469.jpg masks/000000422469.png
273 | train2017/000000330932.jpg masks/000000330932.png
274 | train2017/000000552142.jpg masks/000000552142.png
275 | train2017/000000500890.jpg masks/000000500890.png
276 | train2017/000000462984.jpg masks/000000462984.png
277 | train2017/000000470770.jpg masks/000000470770.png
278 | train2017/000000133874.jpg masks/000000133874.png
279 | train2017/000000042944.jpg masks/000000042944.png
280 | train2017/000000542338.jpg masks/000000542338.png
281 | train2017/000000492325.jpg masks/000000492325.png
282 | train2017/000000275316.jpg masks/000000275316.png
283 | train2017/000000058524.jpg masks/000000058524.png
284 | train2017/000000051808.jpg masks/000000051808.png
285 | train2017/000000275102.jpg masks/000000275102.png
286 | train2017/000000268192.jpg masks/000000268192.png
287 | train2017/000000072229.jpg masks/000000072229.png
288 | train2017/000000075110.jpg masks/000000075110.png
289 | train2017/000000355697.jpg masks/000000355697.png
290 | train2017/000000171467.jpg masks/000000171467.png
291 | train2017/000000474439.jpg masks/000000474439.png
292 | train2017/000000220156.jpg masks/000000220156.png
293 | train2017/000000320533.jpg masks/000000320533.png
294 | train2017/000000503906.jpg masks/000000503906.png
295 | train2017/000000569722.jpg masks/000000569722.png
296 | train2017/000000415269.jpg masks/000000415269.png
297 | train2017/000000441480.jpg masks/000000441480.png
298 | train2017/000000580681.jpg masks/000000580681.png
299 | train2017/000000288099.jpg masks/000000288099.png
300 | train2017/000000528080.jpg masks/000000528080.png
301 | train2017/000000428973.jpg masks/000000428973.png
302 | train2017/000000413757.jpg masks/000000413757.png
303 | train2017/000000468405.jpg masks/000000468405.png
304 | train2017/000000513237.jpg masks/000000513237.png
305 | train2017/000000104046.jpg masks/000000104046.png
306 | train2017/000000539975.jpg masks/000000539975.png
307 | train2017/000000068183.jpg masks/000000068183.png
308 | train2017/000000116588.jpg masks/000000116588.png
309 | train2017/000000138644.jpg masks/000000138644.png
310 | train2017/000000223636.jpg masks/000000223636.png
311 | train2017/000000191803.jpg masks/000000191803.png
312 | train2017/000000293514.jpg masks/000000293514.png
313 | train2017/000000164348.jpg masks/000000164348.png
314 | train2017/000000450340.jpg masks/000000450340.png
315 | train2017/000000411187.jpg masks/000000411187.png
316 | train2017/000000471913.jpg masks/000000471913.png
317 | train2017/000000499423.jpg masks/000000499423.png
318 | train2017/000000300981.jpg masks/000000300981.png
319 | train2017/000000096111.jpg masks/000000096111.png
320 | train2017/000000260883.jpg masks/000000260883.png
321 | train2017/000000262175.jpg masks/000000262175.png
322 | train2017/000000207513.jpg masks/000000207513.png
323 | train2017/000000264733.jpg masks/000000264733.png
324 | train2017/000000430345.jpg masks/000000430345.png
325 | train2017/000000296030.jpg masks/000000296030.png
326 | train2017/000000551227.jpg masks/000000551227.png
327 | train2017/000000097777.jpg masks/000000097777.png
328 | train2017/000000252342.jpg masks/000000252342.png
329 | train2017/000000085372.jpg masks/000000085372.png
330 | train2017/000000473196.jpg masks/000000473196.png
331 | train2017/000000220224.jpg masks/000000220224.png
332 | train2017/000000064681.jpg masks/000000064681.png
333 | train2017/000000027902.jpg masks/000000027902.png
334 | train2017/000000117182.jpg masks/000000117182.png
335 | train2017/000000199363.jpg masks/000000199363.png
336 | train2017/000000459659.jpg masks/000000459659.png
337 | train2017/000000083174.jpg masks/000000083174.png
338 | train2017/000000203653.jpg masks/000000203653.png
339 | train2017/000000096030.jpg masks/000000096030.png
340 | train2017/000000370868.jpg masks/000000370868.png
341 | train2017/000000239611.jpg masks/000000239611.png
342 | train2017/000000063446.jpg masks/000000063446.png
343 | train2017/000000200231.jpg masks/000000200231.png
344 | train2017/000000530630.jpg masks/000000530630.png
345 | train2017/000000540265.jpg masks/000000540265.png
346 | train2017/000000321508.jpg masks/000000321508.png
347 | train2017/000000057481.jpg masks/000000057481.png
348 | train2017/000000269946.jpg masks/000000269946.png
349 | train2017/000000003637.jpg masks/000000003637.png
350 | train2017/000000498915.jpg masks/000000498915.png
351 | train2017/000000571215.jpg masks/000000571215.png
352 | train2017/000000358706.jpg masks/000000358706.png
353 | train2017/000000476417.jpg masks/000000476417.png
354 | train2017/000000047905.jpg masks/000000047905.png
355 | train2017/000000481981.jpg masks/000000481981.png
356 | train2017/000000245425.jpg masks/000000245425.png
357 | train2017/000000325693.jpg masks/000000325693.png
358 | train2017/000000472796.jpg masks/000000472796.png
359 | train2017/000000196571.jpg masks/000000196571.png
360 | train2017/000000507536.jpg masks/000000507536.png
361 | train2017/000000175908.jpg masks/000000175908.png
362 | train2017/000000535457.jpg masks/000000535457.png
363 | train2017/000000283960.jpg masks/000000283960.png
364 | train2017/000000365206.jpg masks/000000365206.png
365 | train2017/000000469242.jpg masks/000000469242.png
366 | train2017/000000083670.jpg masks/000000083670.png
367 | train2017/000000570483.jpg masks/000000570483.png
368 | train2017/000000478009.jpg masks/000000478009.png
369 | train2017/000000543618.jpg masks/000000543618.png
370 | train2017/000000089949.jpg masks/000000089949.png
371 | train2017/000000314126.jpg masks/000000314126.png
372 | train2017/000000486471.jpg masks/000000486471.png
373 | train2017/000000401515.jpg masks/000000401515.png
374 | train2017/000000374340.jpg masks/000000374340.png
375 | train2017/000000327819.jpg masks/000000327819.png
376 | train2017/000000447089.jpg masks/000000447089.png
377 | train2017/000000537918.jpg masks/000000537918.png
378 | train2017/000000370426.jpg masks/000000370426.png
379 | train2017/000000576322.jpg masks/000000576322.png
380 | train2017/000000560784.jpg masks/000000560784.png
381 | train2017/000000378029.jpg masks/000000378029.png
382 | train2017/000000092335.jpg masks/000000092335.png
383 | train2017/000000343606.jpg masks/000000343606.png
384 | train2017/000000393086.jpg masks/000000393086.png
385 | train2017/000000339512.jpg masks/000000339512.png
386 | train2017/000000031893.jpg masks/000000031893.png
387 | train2017/000000478537.jpg masks/000000478537.png
388 | train2017/000000127304.jpg masks/000000127304.png
389 | train2017/000000158038.jpg masks/000000158038.png
390 | train2017/000000415770.jpg masks/000000415770.png
391 | train2017/000000181815.jpg masks/000000181815.png
392 | train2017/000000492974.jpg masks/000000492974.png
393 | train2017/000000407522.jpg masks/000000407522.png
394 | train2017/000000332832.jpg masks/000000332832.png
395 | train2017/000000121361.jpg masks/000000121361.png
396 | train2017/000000003514.jpg masks/000000003514.png
397 | train2017/000000191310.jpg masks/000000191310.png
398 | train2017/000000420573.jpg masks/000000420573.png
399 | train2017/000000146646.jpg masks/000000146646.png
400 | train2017/000000474963.jpg masks/000000474963.png
401 | train2017/000000359085.jpg masks/000000359085.png
402 | train2017/000000273369.jpg masks/000000273369.png
403 | train2017/000000090888.jpg masks/000000090888.png
404 | train2017/000000235038.jpg masks/000000235038.png
405 | train2017/000000227928.jpg masks/000000227928.png
406 | train2017/000000563257.jpg masks/000000563257.png
407 | train2017/000000254304.jpg masks/000000254304.png
408 | train2017/000000383866.jpg masks/000000383866.png
409 | train2017/000000302680.jpg masks/000000302680.png
410 | train2017/000000422657.jpg masks/000000422657.png
411 | train2017/000000542573.jpg masks/000000542573.png
412 | train2017/000000495055.jpg masks/000000495055.png
413 | train2017/000000378657.jpg masks/000000378657.png
414 | train2017/000000268406.jpg masks/000000268406.png
415 | train2017/000000383733.jpg masks/000000383733.png
416 | train2017/000000334416.jpg masks/000000334416.png
417 | train2017/000000505994.jpg masks/000000505994.png
418 | train2017/000000001205.jpg masks/000000001205.png
419 | train2017/000000234424.jpg masks/000000234424.png
420 | train2017/000000535902.jpg masks/000000535902.png
421 | train2017/000000003464.jpg masks/000000003464.png
422 | train2017/000000282444.jpg masks/000000282444.png
423 | train2017/000000302924.jpg masks/000000302924.png
424 | train2017/000000545364.jpg masks/000000545364.png
425 | train2017/000000063121.jpg masks/000000063121.png
426 | train2017/000000062324.jpg masks/000000062324.png
427 | train2017/000000043227.jpg masks/000000043227.png
428 | train2017/000000062581.jpg masks/000000062581.png
429 | train2017/000000234501.jpg masks/000000234501.png
430 | train2017/000000011360.jpg masks/000000011360.png
431 | train2017/000000491850.jpg masks/000000491850.png
432 | train2017/000000194419.jpg masks/000000194419.png
433 | train2017/000000312269.jpg masks/000000312269.png
434 | train2017/000000519827.jpg masks/000000519827.png
435 | train2017/000000527961.jpg masks/000000527961.png
436 | train2017/000000560532.jpg masks/000000560532.png
437 | train2017/000000462536.jpg masks/000000462536.png
438 | train2017/000000117407.jpg masks/000000117407.png
439 | train2017/000000104885.jpg masks/000000104885.png
440 | train2017/000000161885.jpg masks/000000161885.png
441 | train2017/000000029579.jpg masks/000000029579.png
442 | train2017/000000424002.jpg masks/000000424002.png
443 | train2017/000000515040.jpg masks/000000515040.png
444 | train2017/000000173948.jpg masks/000000173948.png
445 | train2017/000000203081.jpg masks/000000203081.png
446 | train2017/000000434423.jpg masks/000000434423.png
447 | train2017/000000022979.jpg masks/000000022979.png
448 | train2017/000000008458.jpg masks/000000008458.png
449 | train2017/000000309791.jpg masks/000000309791.png
450 | train2017/000000559586.jpg masks/000000559586.png
451 | train2017/000000118181.jpg masks/000000118181.png
452 | train2017/000000348076.jpg masks/000000348076.png
453 | train2017/000000519429.jpg masks/000000519429.png
454 | train2017/000000296584.jpg masks/000000296584.png
455 | train2017/000000118724.jpg masks/000000118724.png
456 | train2017/000000126931.jpg masks/000000126931.png
457 | train2017/000000383660.jpg masks/000000383660.png
458 | train2017/000000265449.jpg masks/000000265449.png
459 | train2017/000000083878.jpg masks/000000083878.png
460 | train2017/000000283152.jpg masks/000000283152.png
461 | train2017/000000211725.jpg masks/000000211725.png
462 | train2017/000000567342.jpg masks/000000567342.png
463 | train2017/000000191280.jpg masks/000000191280.png


--------------------------------------------------------------------------------
/partitions/coco/1_512/labeled.txt:
--------------------------------------------------------------------------------
  1 | train2017/000000177254.jpg masks/000000177254.png
  2 | train2017/000000210467.jpg masks/000000210467.png
  3 | train2017/000000468760.jpg masks/000000468760.png
  4 | train2017/000000482382.jpg masks/000000482382.png
  5 | train2017/000000575018.jpg masks/000000575018.png
  6 | train2017/000000568440.jpg masks/000000568440.png
  7 | train2017/000000206705.jpg masks/000000206705.png
  8 | train2017/000000316826.jpg masks/000000316826.png
  9 | train2017/000000562930.jpg masks/000000562930.png
 10 | train2017/000000318594.jpg masks/000000318594.png
 11 | train2017/000000128405.jpg masks/000000128405.png
 12 | train2017/000000079247.jpg masks/000000079247.png
 13 | train2017/000000066866.jpg masks/000000066866.png
 14 | train2017/000000530433.jpg masks/000000530433.png
 15 | train2017/000000236010.jpg masks/000000236010.png
 16 | train2017/000000306926.jpg masks/000000306926.png
 17 | train2017/000000455040.jpg masks/000000455040.png
 18 | train2017/000000485752.jpg masks/000000485752.png
 19 | train2017/000000491453.jpg masks/000000491453.png
 20 | train2017/000000302767.jpg masks/000000302767.png
 21 | train2017/000000435328.jpg masks/000000435328.png
 22 | train2017/000000390671.jpg masks/000000390671.png
 23 | train2017/000000174304.jpg masks/000000174304.png
 24 | train2017/000000310753.jpg masks/000000310753.png
 25 | train2017/000000094145.jpg masks/000000094145.png
 26 | train2017/000000450355.jpg masks/000000450355.png
 27 | train2017/000000032948.jpg masks/000000032948.png
 28 | train2017/000000456658.jpg masks/000000456658.png
 29 | train2017/000000363191.jpg masks/000000363191.png
 30 | train2017/000000202774.jpg masks/000000202774.png
 31 | train2017/000000234839.jpg masks/000000234839.png
 32 | train2017/000000485139.jpg masks/000000485139.png
 33 | train2017/000000435681.jpg masks/000000435681.png
 34 | train2017/000000329245.jpg masks/000000329245.png
 35 | train2017/000000240147.jpg masks/000000240147.png
 36 | train2017/000000077137.jpg masks/000000077137.png
 37 | train2017/000000383185.jpg masks/000000383185.png
 38 | train2017/000000183114.jpg masks/000000183114.png
 39 | train2017/000000578521.jpg masks/000000578521.png
 40 | train2017/000000239208.jpg masks/000000239208.png
 41 | train2017/000000037748.jpg masks/000000037748.png
 42 | train2017/000000511324.jpg masks/000000511324.png
 43 | train2017/000000493552.jpg masks/000000493552.png
 44 | train2017/000000290875.jpg masks/000000290875.png
 45 | train2017/000000296706.jpg masks/000000296706.png
 46 | train2017/000000032724.jpg masks/000000032724.png
 47 | train2017/000000440694.jpg masks/000000440694.png
 48 | train2017/000000238212.jpg masks/000000238212.png
 49 | train2017/000000283748.jpg masks/000000283748.png
 50 | train2017/000000308350.jpg masks/000000308350.png
 51 | train2017/000000088768.jpg masks/000000088768.png
 52 | train2017/000000473103.jpg masks/000000473103.png
 53 | train2017/000000278171.jpg masks/000000278171.png
 54 | train2017/000000142454.jpg masks/000000142454.png
 55 | train2017/000000267825.jpg masks/000000267825.png
 56 | train2017/000000450774.jpg masks/000000450774.png
 57 | train2017/000000046753.jpg masks/000000046753.png
 58 | train2017/000000041366.jpg masks/000000041366.png
 59 | train2017/000000474356.jpg masks/000000474356.png
 60 | train2017/000000381106.jpg masks/000000381106.png
 61 | train2017/000000477179.jpg masks/000000477179.png
 62 | train2017/000000313727.jpg masks/000000313727.png
 63 | train2017/000000103730.jpg masks/000000103730.png
 64 | train2017/000000554743.jpg masks/000000554743.png
 65 | train2017/000000035672.jpg masks/000000035672.png
 66 | train2017/000000411291.jpg masks/000000411291.png
 67 | train2017/000000555920.jpg masks/000000555920.png
 68 | train2017/000000256662.jpg masks/000000256662.png
 69 | train2017/000000438301.jpg masks/000000438301.png
 70 | train2017/000000480344.jpg masks/000000480344.png
 71 | train2017/000000436695.jpg masks/000000436695.png
 72 | train2017/000000003103.jpg masks/000000003103.png
 73 | train2017/000000131696.jpg masks/000000131696.png
 74 | train2017/000000401826.jpg masks/000000401826.png
 75 | train2017/000000379630.jpg masks/000000379630.png
 76 | train2017/000000212429.jpg masks/000000212429.png
 77 | train2017/000000222309.jpg masks/000000222309.png
 78 | train2017/000000181359.jpg masks/000000181359.png
 79 | train2017/000000015274.jpg masks/000000015274.png
 80 | train2017/000000410229.jpg masks/000000410229.png
 81 | train2017/000000423718.jpg masks/000000423718.png
 82 | train2017/000000378911.jpg masks/000000378911.png
 83 | train2017/000000459487.jpg masks/000000459487.png
 84 | train2017/000000298008.jpg masks/000000298008.png
 85 | train2017/000000542814.jpg masks/000000542814.png
 86 | train2017/000000520208.jpg masks/000000520208.png
 87 | train2017/000000507249.jpg masks/000000507249.png
 88 | train2017/000000542662.jpg masks/000000542662.png
 89 | train2017/000000541113.jpg masks/000000541113.png
 90 | train2017/000000547039.jpg masks/000000547039.png
 91 | train2017/000000423620.jpg masks/000000423620.png
 92 | train2017/000000129314.jpg masks/000000129314.png
 93 | train2017/000000051310.jpg masks/000000051310.png
 94 | train2017/000000300170.jpg masks/000000300170.png
 95 | train2017/000000093391.jpg masks/000000093391.png
 96 | train2017/000000156282.jpg masks/000000156282.png
 97 | train2017/000000044151.jpg masks/000000044151.png
 98 | train2017/000000260883.jpg masks/000000260883.png
 99 | train2017/000000297180.jpg masks/000000297180.png
100 | train2017/000000521982.jpg masks/000000521982.png
101 | train2017/000000051800.jpg masks/000000051800.png
102 | train2017/000000525899.jpg masks/000000525899.png
103 | train2017/000000348648.jpg masks/000000348648.png
104 | train2017/000000567858.jpg masks/000000567858.png
105 | train2017/000000151566.jpg masks/000000151566.png
106 | train2017/000000211609.jpg masks/000000211609.png
107 | train2017/000000091796.jpg masks/000000091796.png
108 | train2017/000000393108.jpg masks/000000393108.png
109 | train2017/000000255550.jpg masks/000000255550.png
110 | train2017/000000000009.jpg masks/000000000009.png
111 | train2017/000000354424.jpg masks/000000354424.png
112 | train2017/000000518907.jpg masks/000000518907.png
113 | train2017/000000245409.jpg masks/000000245409.png
114 | train2017/000000518365.jpg masks/000000518365.png
115 | train2017/000000150769.jpg masks/000000150769.png
116 | train2017/000000362301.jpg masks/000000362301.png
117 | train2017/000000126833.jpg masks/000000126833.png
118 | train2017/000000202128.jpg masks/000000202128.png
119 | train2017/000000344415.jpg masks/000000344415.png
120 | train2017/000000505010.jpg masks/000000505010.png
121 | train2017/000000194885.jpg masks/000000194885.png
122 | train2017/000000459347.jpg masks/000000459347.png
123 | train2017/000000238029.jpg masks/000000238029.png
124 | train2017/000000267457.jpg masks/000000267457.png
125 | train2017/000000358642.jpg masks/000000358642.png
126 | train2017/000000173644.jpg masks/000000173644.png
127 | train2017/000000496064.jpg masks/000000496064.png
128 | train2017/000000145502.jpg masks/000000145502.png
129 | train2017/000000117026.jpg masks/000000117026.png
130 | train2017/000000408705.jpg masks/000000408705.png
131 | train2017/000000426414.jpg masks/000000426414.png
132 | train2017/000000036207.jpg masks/000000036207.png
133 | train2017/000000314907.jpg masks/000000314907.png
134 | train2017/000000024815.jpg masks/000000024815.png
135 | train2017/000000420827.jpg masks/000000420827.png
136 | train2017/000000312216.jpg masks/000000312216.png
137 | train2017/000000097659.jpg masks/000000097659.png
138 | train2017/000000132297.jpg masks/000000132297.png
139 | train2017/000000445723.jpg masks/000000445723.png
140 | train2017/000000341011.jpg masks/000000341011.png
141 | train2017/000000447911.jpg masks/000000447911.png
142 | train2017/000000294545.jpg masks/000000294545.png
143 | train2017/000000228580.jpg masks/000000228580.png
144 | train2017/000000205911.jpg masks/000000205911.png
145 | train2017/000000289883.jpg masks/000000289883.png
146 | train2017/000000125291.jpg masks/000000125291.png
147 | train2017/000000092377.jpg masks/000000092377.png
148 | train2017/000000227003.jpg masks/000000227003.png
149 | train2017/000000488965.jpg masks/000000488965.png
150 | train2017/000000141538.jpg masks/000000141538.png
151 | train2017/000000185328.jpg masks/000000185328.png
152 | train2017/000000477016.jpg masks/000000477016.png
153 | train2017/000000070527.jpg masks/000000070527.png
154 | train2017/000000271235.jpg masks/000000271235.png
155 | train2017/000000265303.jpg masks/000000265303.png
156 | train2017/000000405582.jpg masks/000000405582.png
157 | train2017/000000522489.jpg masks/000000522489.png
158 | train2017/000000543233.jpg masks/000000543233.png
159 | train2017/000000382670.jpg masks/000000382670.png
160 | train2017/000000255827.jpg masks/000000255827.png
161 | train2017/000000436759.jpg masks/000000436759.png
162 | train2017/000000260882.jpg masks/000000260882.png
163 | train2017/000000329762.jpg masks/000000329762.png
164 | train2017/000000226211.jpg masks/000000226211.png
165 | train2017/000000094610.jpg masks/000000094610.png
166 | train2017/000000017311.jpg masks/000000017311.png
167 | train2017/000000389480.jpg masks/000000389480.png
168 | train2017/000000211002.jpg masks/000000211002.png
169 | train2017/000000542694.jpg masks/000000542694.png
170 | train2017/000000125244.jpg masks/000000125244.png
171 | train2017/000000477946.jpg masks/000000477946.png
172 | train2017/000000072708.jpg masks/000000072708.png
173 | train2017/000000142246.jpg masks/000000142246.png
174 | train2017/000000328427.jpg masks/000000328427.png
175 | train2017/000000274474.jpg masks/000000274474.png
176 | train2017/000000578023.jpg masks/000000578023.png
177 | train2017/000000272481.jpg masks/000000272481.png
178 | train2017/000000081001.jpg masks/000000081001.png
179 | train2017/000000231991.jpg masks/000000231991.png
180 | train2017/000000379158.jpg masks/000000379158.png
181 | train2017/000000333181.jpg masks/000000333181.png
182 | train2017/000000173776.jpg masks/000000173776.png
183 | train2017/000000143401.jpg masks/000000143401.png
184 | train2017/000000183617.jpg masks/000000183617.png
185 | train2017/000000161013.jpg masks/000000161013.png
186 | train2017/000000438492.jpg masks/000000438492.png
187 | train2017/000000260321.jpg masks/000000260321.png
188 | train2017/000000564865.jpg masks/000000564865.png
189 | train2017/000000008676.jpg masks/000000008676.png
190 | train2017/000000113598.jpg masks/000000113598.png
191 | train2017/000000290285.jpg masks/000000290285.png
192 | train2017/000000480317.jpg masks/000000480317.png
193 | train2017/000000371514.jpg masks/000000371514.png
194 | train2017/000000320872.jpg masks/000000320872.png
195 | train2017/000000391857.jpg masks/000000391857.png
196 | train2017/000000182620.jpg masks/000000182620.png
197 | train2017/000000038850.jpg masks/000000038850.png
198 | train2017/000000241867.jpg masks/000000241867.png
199 | train2017/000000545370.jpg masks/000000545370.png
200 | train2017/000000008027.jpg masks/000000008027.png
201 | train2017/000000436392.jpg masks/000000436392.png
202 | train2017/000000134144.jpg masks/000000134144.png
203 | train2017/000000406268.jpg masks/000000406268.png
204 | train2017/000000324616.jpg masks/000000324616.png
205 | train2017/000000325768.jpg masks/000000325768.png
206 | train2017/000000410739.jpg masks/000000410739.png
207 | train2017/000000255950.jpg masks/000000255950.png
208 | train2017/000000219407.jpg masks/000000219407.png
209 | train2017/000000537543.jpg masks/000000537543.png
210 | train2017/000000043719.jpg masks/000000043719.png
211 | train2017/000000027626.jpg masks/000000027626.png
212 | train2017/000000286671.jpg masks/000000286671.png
213 | train2017/000000569234.jpg masks/000000569234.png
214 | train2017/000000347407.jpg masks/000000347407.png
215 | train2017/000000261702.jpg masks/000000261702.png
216 | train2017/000000092761.jpg masks/000000092761.png
217 | train2017/000000477434.jpg masks/000000477434.png
218 | train2017/000000556886.jpg masks/000000556886.png
219 | train2017/000000478055.jpg masks/000000478055.png
220 | train2017/000000007023.jpg masks/000000007023.png
221 | train2017/000000448529.jpg masks/000000448529.png
222 | train2017/000000225505.jpg masks/000000225505.png
223 | train2017/000000397375.jpg masks/000000397375.png
224 | train2017/000000028317.jpg masks/000000028317.png
225 | train2017/000000555358.jpg masks/000000555358.png
226 | train2017/000000562632.jpg masks/000000562632.png
227 | train2017/000000405278.jpg masks/000000405278.png
228 | train2017/000000565203.jpg masks/000000565203.png
229 | train2017/000000132527.jpg masks/000000132527.png
230 | train2017/000000364525.jpg masks/000000364525.png
231 | train2017/000000525636.jpg masks/000000525636.png
232 | train2017/000000143516.jpg masks/000000143516.png


--------------------------------------------------------------------------------
/partitions/pascal/183/labeled.txt:
--------------------------------------------------------------------------------
  1 | JPEGImages/2008_006748.jpg SegmentationClass/2008_006748.png
  2 | JPEGImages/2011_000453.jpg SegmentationClass/2011_000453.png
  3 | JPEGImages/2010_002338.jpg SegmentationClass/2010_002338.png
  4 | JPEGImages/2009_000161.jpg SegmentationClass/2009_000161.png
  5 | JPEGImages/2009_003860.jpg SegmentationClass/2009_003860.png
  6 | JPEGImages/2007_008945.jpg SegmentationClass/2007_008945.png
  7 | JPEGImages/2007_002488.jpg SegmentationClass/2007_002488.png
  8 | JPEGImages/2008_001056.jpg SegmentationClass/2008_001056.png
  9 | JPEGImages/2008_000515.jpg SegmentationClass/2008_000515.png
 10 | JPEGImages/2010_002286.jpg SegmentationClass/2010_002286.png
 11 | JPEGImages/2009_000603.jpg SegmentationClass/2009_000603.png
 12 | JPEGImages/2010_005500.jpg SegmentationClass/2010_005500.png
 13 | JPEGImages/2009_000015.jpg SegmentationClass/2009_000015.png
 14 | JPEGImages/2008_001358.jpg SegmentationClass/2008_001358.png
 15 | JPEGImages/2011_001400.jpg SegmentationClass/2011_001400.png
 16 | JPEGImages/2010_001386.jpg SegmentationClass/2010_001386.png
 17 | JPEGImages/2007_009580.jpg SegmentationClass/2007_009580.png
 18 | JPEGImages/2007_006899.jpg SegmentationClass/2007_006899.png
 19 | JPEGImages/2007_006317.jpg SegmentationClass/2007_006317.png
 20 | JPEGImages/2008_003252.jpg SegmentationClass/2008_003252.png
 21 | JPEGImages/2010_005232.jpg SegmentationClass/2010_005232.png
 22 | JPEGImages/2011_002350.jpg SegmentationClass/2011_002350.png
 23 | JPEGImages/2011_002920.jpg SegmentationClass/2011_002920.png
 24 | JPEGImages/2009_004117.jpg SegmentationClass/2009_004117.png
 25 | JPEGImages/2011_002246.jpg SegmentationClass/2011_002246.png
 26 | JPEGImages/2008_003200.jpg SegmentationClass/2008_003200.png
 27 | JPEGImages/2007_005988.jpg SegmentationClass/2007_005988.png
 28 | JPEGImages/2009_001339.jpg SegmentationClass/2009_001339.png
 29 | JPEGImages/2010_005429.jpg SegmentationClass/2010_005429.png
 30 | JPEGImages/2009_001117.jpg SegmentationClass/2009_001117.png
 31 | JPEGImages/2008_005668.jpg SegmentationClass/2008_005668.png
 32 | JPEGImages/2008_002338.jpg SegmentationClass/2008_002338.png
 33 | JPEGImages/2007_005273.jpg SegmentationClass/2007_005273.png
 34 | JPEGImages/2011_000382.jpg SegmentationClass/2011_000382.png
 35 | JPEGImages/2009_000408.jpg SegmentationClass/2009_000408.png
 36 | JPEGImages/2009_000420.jpg SegmentationClass/2009_000420.png
 37 | JPEGImages/2010_003380.jpg SegmentationClass/2010_003380.png
 38 | JPEGImages/2007_005262.jpg SegmentationClass/2007_005262.png
 39 | JPEGImages/2010_003799.jpg SegmentationClass/2010_003799.png
 40 | JPEGImages/2008_005843.jpg SegmentationClass/2008_005843.png
 41 | JPEGImages/2007_003715.jpg SegmentationClass/2007_003715.png
 42 | JPEGImages/2009_004425.jpg SegmentationClass/2009_004425.png
 43 | JPEGImages/2009_003172.jpg SegmentationClass/2009_003172.png
 44 | JPEGImages/2011_000152.jpg SegmentationClass/2011_000152.png
 45 | JPEGImages/2007_000528.jpg SegmentationClass/2007_000528.png
 46 | JPEGImages/2011_003246.jpg SegmentationClass/2011_003246.png
 47 | JPEGImages/2007_008307.jpg SegmentationClass/2007_008307.png
 48 | JPEGImages/2007_007585.jpg SegmentationClass/2007_007585.png
 49 | JPEGImages/2011_001810.jpg SegmentationClass/2011_001810.png
 50 | JPEGImages/2008_006490.jpg SegmentationClass/2008_006490.png
 51 | JPEGImages/2007_009052.jpg SegmentationClass/2007_009052.png
 52 | JPEGImages/2009_003975.jpg SegmentationClass/2009_003975.png
 53 | JPEGImages/2010_000075.jpg SegmentationClass/2010_000075.png
 54 | JPEGImages/2007_002967.jpg SegmentationClass/2007_002967.png
 55 | JPEGImages/2007_003118.jpg SegmentationClass/2007_003118.png
 56 | JPEGImages/2008_005214.jpg SegmentationClass/2008_005214.png
 57 | JPEGImages/2007_000738.jpg SegmentationClass/2007_000738.png
 58 | JPEGImages/2009_000285.jpg SegmentationClass/2009_000285.png
 59 | JPEGImages/2007_007481.jpg SegmentationClass/2007_007481.png
 60 | JPEGImages/2010_001418.jpg SegmentationClass/2010_001418.png
 61 | JPEGImages/2011_002335.jpg SegmentationClass/2011_002335.png
 62 | JPEGImages/2007_002273.jpg SegmentationClass/2007_002273.png
 63 | JPEGImages/2011_001166.jpg SegmentationClass/2011_001166.png
 64 | JPEGImages/2011_003038.jpg SegmentationClass/2011_003038.png
 65 | JPEGImages/2008_005006.jpg SegmentationClass/2008_005006.png
 66 | JPEGImages/2010_000117.jpg SegmentationClass/2010_000117.png
 67 | JPEGImages/2010_002625.jpg SegmentationClass/2010_002625.png
 68 | JPEGImages/2009_002993.jpg SegmentationClass/2009_002993.png
 69 | JPEGImages/2007_005790.jpg SegmentationClass/2007_005790.png
 70 | JPEGImages/2008_003180.jpg SegmentationClass/2008_003180.png
 71 | JPEGImages/2009_001095.jpg SegmentationClass/2009_001095.png
 72 | JPEGImages/2010_001329.jpg SegmentationClass/2010_001329.png
 73 | JPEGImages/2011_002559.jpg SegmentationClass/2011_002559.png
 74 | JPEGImages/2009_001070.jpg SegmentationClass/2009_001070.png
 75 | JPEGImages/2010_005318.jpg SegmentationClass/2010_005318.png
 76 | JPEGImages/2011_000122.jpg SegmentationClass/2011_000122.png
 77 | JPEGImages/2009_003697.jpg SegmentationClass/2009_003697.png
 78 | JPEGImages/2010_002892.jpg SegmentationClass/2010_002892.png
 79 | JPEGImages/2009_002019.jpg SegmentationClass/2009_002019.png
 80 | JPEGImages/2008_002471.jpg SegmentationClass/2008_002471.png
 81 | JPEGImages/2007_005314.jpg SegmentationClass/2007_005314.png
 82 | JPEGImages/2007_007398.jpg SegmentationClass/2007_007398.png
 83 | JPEGImages/2008_008343.jpg SegmentationClass/2008_008343.png
 84 | JPEGImages/2007_003207.jpg SegmentationClass/2007_003207.png
 85 | JPEGImages/2009_002588.jpg SegmentationClass/2009_002588.png
 86 | JPEGImages/2009_002422.jpg SegmentationClass/2009_002422.png
 87 | JPEGImages/2008_002697.jpg SegmentationClass/2008_002697.png
 88 | JPEGImages/2011_001754.jpg SegmentationClass/2011_001754.png
 89 | JPEGImages/2008_007375.jpg SegmentationClass/2008_007375.png
 90 | JPEGImages/2008_003060.jpg SegmentationClass/2008_003060.png
 91 | JPEGImages/2011_001259.jpg SegmentationClass/2011_001259.png
 92 | JPEGImages/2010_004171.jpg SegmentationClass/2010_004171.png
 93 | JPEGImages/2009_000655.jpg SegmentationClass/2009_000655.png
 94 | JPEGImages/2010_004478.jpg SegmentationClass/2010_004478.png
 95 | JPEGImages/2011_001855.jpg SegmentationClass/2011_001855.png
 96 | JPEGImages/2007_009649.jpg SegmentationClass/2007_009649.png
 97 | JPEGImages/2007_003267.jpg SegmentationClass/2007_003267.png
 98 | JPEGImages/2008_001188.jpg SegmentationClass/2008_001188.png
 99 | JPEGImages/2010_004900.jpg SegmentationClass/2010_004900.png
100 | JPEGImages/2011_000449.jpg SegmentationClass/2011_000449.png
101 | JPEGImages/2008_004869.jpg SegmentationClass/2008_004869.png
102 | JPEGImages/2008_004838.jpg SegmentationClass/2008_004838.png
103 | JPEGImages/2010_002907.jpg SegmentationClass/2010_002907.png
104 | JPEGImages/2008_004259.jpg SegmentationClass/2008_004259.png
105 | JPEGImages/2008_003429.jpg SegmentationClass/2008_003429.png
106 | JPEGImages/2011_000999.jpg SegmentationClass/2011_000999.png
107 | JPEGImages/2011_001632.jpg SegmentationClass/2011_001632.png
108 | JPEGImages/2011_000577.jpg SegmentationClass/2011_000577.png
109 | JPEGImages/2010_005505.jpg SegmentationClass/2010_005505.png
110 | JPEGImages/2010_004365.jpg SegmentationClass/2010_004365.png
111 | JPEGImages/2007_007948.jpg SegmentationClass/2007_007948.png
112 | JPEGImages/2008_005713.jpg SegmentationClass/2008_005713.png
113 | JPEGImages/2008_006920.jpg SegmentationClass/2008_006920.png
114 | JPEGImages/2010_000132.jpg SegmentationClass/2010_000132.png
115 | JPEGImages/2008_006182.jpg SegmentationClass/2008_006182.png
116 | JPEGImages/2007_005212.jpg SegmentationClass/2007_005212.png
117 | JPEGImages/2009_000746.jpg SegmentationClass/2009_000746.png
118 | JPEGImages/2009_005236.jpg SegmentationClass/2009_005236.png
119 | JPEGImages/2010_005876.jpg SegmentationClass/2010_005876.png
120 | JPEGImages/2007_002462.jpg SegmentationClass/2007_002462.png
121 | JPEGImages/2009_000409.jpg SegmentationClass/2009_000409.png
122 | JPEGImages/2008_008525.jpg SegmentationClass/2008_008525.png
123 | JPEGImages/2011_002114.jpg SegmentationClass/2011_002114.png
124 | JPEGImages/2010_003887.jpg SegmentationClass/2010_003887.png
125 | JPEGImages/2009_005287.jpg SegmentationClass/2009_005287.png
126 | JPEGImages/2010_002815.jpg SegmentationClass/2010_002815.png
127 | JPEGImages/2010_002203.jpg SegmentationClass/2010_002203.png
128 | JPEGImages/2010_003717.jpg SegmentationClass/2010_003717.png
129 | JPEGImages/2010_005277.jpg SegmentationClass/2010_005277.png
130 | JPEGImages/2008_000832.jpg SegmentationClass/2008_000832.png
131 | JPEGImages/2009_003757.jpg SegmentationClass/2009_003757.png
132 | JPEGImages/2007_006477.jpg SegmentationClass/2007_006477.png
133 | JPEGImages/2008_003480.jpg SegmentationClass/2008_003480.png
134 | JPEGImages/2009_003783.jpg SegmentationClass/2009_003783.png
135 | JPEGImages/2009_003646.jpg SegmentationClass/2009_003646.png
136 | JPEGImages/2011_000068.jpg SegmentationClass/2011_000068.png
137 | JPEGImages/2010_002573.jpg SegmentationClass/2010_002573.png
138 | JPEGImages/2010_001160.jpg SegmentationClass/2010_001160.png
139 | JPEGImages/2011_002447.jpg SegmentationClass/2011_002447.png
140 | JPEGImages/2008_002215.jpg SegmentationClass/2008_002215.png
141 | JPEGImages/2011_002709.jpg SegmentationClass/2011_002709.png
142 | JPEGImages/2010_004721.jpg SegmentationClass/2010_004721.png
143 | JPEGImages/2008_003065.jpg SegmentationClass/2008_003065.png
144 | JPEGImages/2009_002314.jpg SegmentationClass/2009_002314.png
145 | JPEGImages/2007_009605.jpg SegmentationClass/2007_009605.png
146 | JPEGImages/2007_007230.jpg SegmentationClass/2007_007230.png
147 | JPEGImages/2008_003373.jpg SegmentationClass/2008_003373.png
148 | JPEGImages/2011_002135.jpg SegmentationClass/2011_002135.png
149 | JPEGImages/2008_003939.jpg SegmentationClass/2008_003939.png
150 | JPEGImages/2010_001706.jpg SegmentationClass/2010_001706.png
151 | JPEGImages/2008_002218.jpg SegmentationClass/2008_002218.png
152 | JPEGImages/2009_003034.jpg SegmentationClass/2009_003034.png
153 | JPEGImages/2010_001247.jpg SegmentationClass/2010_001247.png
154 | JPEGImages/2010_001922.jpg SegmentationClass/2010_001922.png
155 | JPEGImages/2007_009832.jpg SegmentationClass/2007_009832.png
156 | JPEGImages/2011_000208.jpg SegmentationClass/2011_000208.png
157 | JPEGImages/2010_002527.jpg SegmentationClass/2010_002527.png
158 | JPEGImages/2010_002154.jpg SegmentationClass/2010_002154.png
159 | JPEGImages/2009_001203.jpg SegmentationClass/2009_001203.png
160 | JPEGImages/2007_003251.jpg SegmentationClass/2007_003251.png
161 | JPEGImages/2010_002962.jpg SegmentationClass/2010_002962.png
162 | JPEGImages/2007_000515.jpg SegmentationClass/2007_000515.png
163 | JPEGImages/2009_002387.jpg SegmentationClass/2009_002387.png
164 | JPEGImages/2007_006400.jpg SegmentationClass/2007_006400.png
165 | JPEGImages/2009_003249.jpg SegmentationClass/2009_003249.png
166 | JPEGImages/2009_000454.jpg SegmentationClass/2009_000454.png
167 | JPEGImages/2009_004539.jpg SegmentationClass/2009_004539.png
168 | JPEGImages/2009_001544.jpg SegmentationClass/2009_001544.png
169 | JPEGImages/2009_004446.jpg SegmentationClass/2009_004446.png
170 | JPEGImages/2010_004683.jpg SegmentationClass/2010_004683.png
171 | JPEGImages/2007_007878.jpg SegmentationClass/2007_007878.png
172 | JPEGImages/2011_002291.jpg SegmentationClass/2011_002291.png
173 | JPEGImages/2011_000108.jpg SegmentationClass/2011_000108.png
174 | JPEGImages/2010_004154.jpg SegmentationClass/2010_004154.png
175 | JPEGImages/2010_001347.jpg SegmentationClass/2010_001347.png
176 | JPEGImages/2008_004892.jpg SegmentationClass/2008_004892.png
177 | JPEGImages/2007_005264.jpg SegmentationClass/2007_005264.png
178 | JPEGImages/2009_004561.jpg SegmentationClass/2009_004561.png
179 | JPEGImages/2011_000468.jpg SegmentationClass/2011_000468.png
180 | JPEGImages/2011_000920.jpg SegmentationClass/2011_000920.png
181 | JPEGImages/2011_001928.jpg SegmentationClass/2011_001928.png
182 | JPEGImages/2008_003562.jpg SegmentationClass/2008_003562.png
183 | JPEGImages/2009_001036.jpg SegmentationClass/2009_001036.png


--------------------------------------------------------------------------------
/partitions/pascal/366/labeled.txt:
--------------------------------------------------------------------------------
  1 | JPEGImages/2008_004112.jpg SegmentationClass/2008_004112.png
  2 | JPEGImages/2010_005028.jpg SegmentationClass/2010_005028.png
  3 | JPEGImages/2010_004933.jpg SegmentationClass/2010_004933.png
  4 | JPEGImages/2009_003345.jpg SegmentationClass/2009_003345.png
  5 | JPEGImages/2008_002641.jpg SegmentationClass/2008_002641.png
  6 | JPEGImages/2011_002410.jpg SegmentationClass/2011_002410.png
  7 | JPEGImages/2010_002382.jpg SegmentationClass/2010_002382.png
  8 | JPEGImages/2007_002024.jpg SegmentationClass/2007_002024.png
  9 | JPEGImages/2009_004095.jpg SegmentationClass/2009_004095.png
 10 | JPEGImages/2009_001961.jpg SegmentationClass/2009_001961.png
 11 | JPEGImages/2007_007783.jpg SegmentationClass/2007_007783.png
 12 | JPEGImages/2011_001765.jpg SegmentationClass/2011_001765.png
 13 | JPEGImages/2011_000216.jpg SegmentationClass/2011_000216.png
 14 | JPEGImages/2011_000453.jpg SegmentationClass/2011_000453.png
 15 | JPEGImages/2011_000252.jpg SegmentationClass/2011_000252.png
 16 | JPEGImages/2009_003369.jpg SegmentationClass/2009_003369.png
 17 | JPEGImages/2007_002361.jpg SegmentationClass/2007_002361.png
 18 | JPEGImages/2011_003025.jpg SegmentationClass/2011_003025.png
 19 | JPEGImages/2007_003541.jpg SegmentationClass/2007_003541.png
 20 | JPEGImages/2009_001251.jpg SegmentationClass/2009_001251.png
 21 | JPEGImages/2007_009759.jpg SegmentationClass/2007_009759.png
 22 | JPEGImages/2007_003529.jpg SegmentationClass/2007_003529.png
 23 | JPEGImages/2010_004805.jpg SegmentationClass/2010_004805.png
 24 | JPEGImages/2008_001119.jpg SegmentationClass/2008_001119.png
 25 | JPEGImages/2008_000259.jpg SegmentationClass/2008_000259.png
 26 | JPEGImages/2008_002650.jpg SegmentationClass/2008_002650.png
 27 | JPEGImages/2009_003012.jpg SegmentationClass/2009_003012.png
 28 | JPEGImages/2010_004361.jpg SegmentationClass/2010_004361.png
 29 | JPEGImages/2008_004026.jpg SegmentationClass/2008_004026.png
 30 | JPEGImages/2008_003083.jpg SegmentationClass/2008_003083.png
 31 | JPEGImages/2011_000550.jpg SegmentationClass/2011_000550.png
 32 | JPEGImages/2010_004808.jpg SegmentationClass/2010_004808.png
 33 | JPEGImages/2011_002503.jpg SegmentationClass/2011_002503.png
 34 | JPEGImages/2011_002770.jpg SegmentationClass/2011_002770.png
 35 | JPEGImages/2010_001183.jpg SegmentationClass/2010_001183.png
 36 | JPEGImages/2011_000790.jpg SegmentationClass/2011_000790.png
 37 | JPEGImages/2009_004328.jpg SegmentationClass/2009_004328.png
 38 | JPEGImages/2008_000764.jpg SegmentationClass/2008_000764.png
 39 | JPEGImages/2010_002055.jpg SegmentationClass/2010_002055.png
 40 | JPEGImages/2009_003090.jpg SegmentationClass/2009_003090.png
 41 | JPEGImages/2008_000540.jpg SegmentationClass/2008_000540.png
 42 | JPEGImages/2009_001885.jpg SegmentationClass/2009_001885.png
 43 | JPEGImages/2010_005223.jpg SegmentationClass/2010_005223.png
 44 | JPEGImages/2009_001002.jpg SegmentationClass/2009_001002.png
 45 | JPEGImages/2010_000466.jpg SegmentationClass/2010_000466.png
 46 | JPEGImages/2008_003068.jpg SegmentationClass/2008_003068.png
 47 | JPEGImages/2008_007357.jpg SegmentationClass/2008_007357.png
 48 | JPEGImages/2008_007165.jpg SegmentationClass/2008_007165.png
 49 | JPEGImages/2009_005037.jpg SegmentationClass/2009_005037.png
 50 | JPEGImages/2007_004003.jpg SegmentationClass/2007_004003.png
 51 | JPEGImages/2011_000468.jpg SegmentationClass/2011_000468.png
 52 | JPEGImages/2008_001106.jpg SegmentationClass/2008_001106.png
 53 | JPEGImages/2010_002139.jpg SegmentationClass/2010_002139.png
 54 | JPEGImages/2011_002149.jpg SegmentationClass/2011_002149.png
 55 | JPEGImages/2010_005744.jpg SegmentationClass/2010_005744.png
 56 | JPEGImages/2008_000711.jpg SegmentationClass/2008_000711.png
 57 | JPEGImages/2008_004892.jpg SegmentationClass/2008_004892.png
 58 | JPEGImages/2010_002107.jpg SegmentationClass/2010_002107.png
 59 | JPEGImages/2007_004476.jpg SegmentationClass/2007_004476.png
 60 | JPEGImages/2009_004887.jpg SegmentationClass/2009_004887.png
 61 | JPEGImages/2008_003769.jpg SegmentationClass/2008_003769.png
 62 | JPEGImages/2010_000986.jpg SegmentationClass/2010_000986.png
 63 | JPEGImages/2011_002119.jpg SegmentationClass/2011_002119.png
 64 | JPEGImages/2008_002182.jpg SegmentationClass/2008_002182.png
 65 | JPEGImages/2008_000436.jpg SegmentationClass/2008_000436.png
 66 | JPEGImages/2009_000887.jpg SegmentationClass/2009_000887.png
 67 | JPEGImages/2008_002868.jpg SegmentationClass/2008_002868.png
 68 | JPEGImages/2008_008462.jpg SegmentationClass/2008_008462.png
 69 | JPEGImages/2007_009030.jpg SegmentationClass/2007_009030.png
 70 | JPEGImages/2008_001413.jpg SegmentationClass/2008_001413.png
 71 | JPEGImages/2010_005927.jpg SegmentationClass/2010_005927.png
 72 | JPEGImages/2009_000774.jpg SegmentationClass/2009_000774.png
 73 | JPEGImages/2011_000197.jpg SegmentationClass/2011_000197.png
 74 | JPEGImages/2008_002834.jpg SegmentationClass/2008_002834.png
 75 | JPEGImages/2011_000551.jpg SegmentationClass/2011_000551.png
 76 | JPEGImages/2009_001636.jpg SegmentationClass/2009_001636.png
 77 | JPEGImages/2009_002072.jpg SegmentationClass/2009_002072.png
 78 | JPEGImages/2008_002749.jpg SegmentationClass/2008_002749.png
 79 | JPEGImages/2007_002142.jpg SegmentationClass/2007_002142.png
 80 | JPEGImages/2007_006483.jpg SegmentationClass/2007_006483.png
 81 | JPEGImages/2008_001523.jpg SegmentationClass/2008_001523.png
 82 | JPEGImages/2007_003205.jpg SegmentationClass/2007_003205.png
 83 | JPEGImages/2009_001270.jpg SegmentationClass/2009_001270.png
 84 | JPEGImages/2007_009649.jpg SegmentationClass/2007_009649.png
 85 | JPEGImages/2009_002849.jpg SegmentationClass/2009_002849.png
 86 | JPEGImages/2010_005450.jpg SegmentationClass/2010_005450.png
 87 | JPEGImages/2009_005269.jpg SegmentationClass/2009_005269.png
 88 | JPEGImages/2010_003799.jpg SegmentationClass/2010_003799.png
 89 | JPEGImages/2010_003230.jpg SegmentationClass/2010_003230.png
 90 | JPEGImages/2010_005796.jpg SegmentationClass/2010_005796.png
 91 | JPEGImages/2009_001735.jpg SegmentationClass/2009_001735.png
 92 | JPEGImages/2008_005770.jpg SegmentationClass/2008_005770.png
 93 | JPEGImages/2008_005367.jpg SegmentationClass/2008_005367.png
 94 | JPEGImages/2011_000243.jpg SegmentationClass/2011_000243.png
 95 | JPEGImages/2010_004694.jpg SegmentationClass/2010_004694.png
 96 | JPEGImages/2007_009630.jpg SegmentationClass/2007_009630.png
 97 | JPEGImages/2008_001188.jpg SegmentationClass/2008_001188.png
 98 | JPEGImages/2011_000646.jpg SegmentationClass/2011_000646.png
 99 | JPEGImages/2008_004663.jpg SegmentationClass/2008_004663.png
100 | JPEGImages/2007_006615.jpg SegmentationClass/2007_006615.png
101 | JPEGImages/2007_007021.jpg SegmentationClass/2007_007021.png
102 | JPEGImages/2008_002338.jpg SegmentationClass/2008_002338.png
103 | JPEGImages/2010_001347.jpg SegmentationClass/2010_001347.png
104 | JPEGImages/2009_002613.jpg SegmentationClass/2009_002613.png
105 | JPEGImages/2009_001782.jpg SegmentationClass/2009_001782.png
106 | JPEGImages/2010_004210.jpg SegmentationClass/2010_004210.png
107 | JPEGImages/2008_000309.jpg SegmentationClass/2008_000309.png
108 | JPEGImages/2007_007250.jpg SegmentationClass/2007_007250.png
109 | JPEGImages/2009_001444.jpg SegmentationClass/2009_001444.png
110 | JPEGImages/2009_001036.jpg SegmentationClass/2009_001036.png
111 | JPEGImages/2007_006212.jpg SegmentationClass/2007_006212.png
112 | JPEGImages/2011_000920.jpg SegmentationClass/2011_000920.png
113 | JPEGImages/2010_001273.jpg SegmentationClass/2010_001273.png
114 | JPEGImages/2010_001619.jpg SegmentationClass/2010_001619.png
115 | JPEGImages/2007_009597.jpg SegmentationClass/2007_009597.png
116 | JPEGImages/2008_005541.jpg SegmentationClass/2008_005541.png
117 | JPEGImages/2009_000684.jpg SegmentationClass/2009_000684.png
118 | JPEGImages/2008_006655.jpg SegmentationClass/2008_006655.png
119 | JPEGImages/2008_000832.jpg SegmentationClass/2008_000832.png
120 | JPEGImages/2010_001944.jpg SegmentationClass/2010_001944.png
121 | JPEGImages/2007_009322.jpg SegmentationClass/2007_009322.png
122 | JPEGImages/2010_001660.jpg SegmentationClass/2010_001660.png
123 | JPEGImages/2008_004430.jpg SegmentationClass/2008_004430.png
124 | JPEGImages/2009_005234.jpg SegmentationClass/2009_005234.png
125 | JPEGImages/2007_000648.jpg SegmentationClass/2007_000648.png
126 | JPEGImages/2007_009435.jpg SegmentationClass/2007_009435.png
127 | JPEGImages/2010_002018.jpg SegmentationClass/2010_002018.png
128 | JPEGImages/2008_002218.jpg SegmentationClass/2008_002218.png
129 | JPEGImages/2009_001937.jpg SegmentationClass/2009_001937.png
130 | JPEGImages/2007_000645.jpg SegmentationClass/2007_000645.png
131 | JPEGImages/2010_004365.jpg SegmentationClass/2010_004365.png
132 | JPEGImages/2010_002457.jpg SegmentationClass/2010_002457.png
133 | JPEGImages/2007_000170.jpg SegmentationClass/2007_000170.png
134 | JPEGImages/2007_002273.jpg SegmentationClass/2007_002273.png
135 | JPEGImages/2007_007902.jpg SegmentationClass/2007_007902.png
136 | JPEGImages/2010_002418.jpg SegmentationClass/2010_002418.png
137 | JPEGImages/2009_003340.jpg SegmentationClass/2009_003340.png
138 | JPEGImages/2011_000258.jpg SegmentationClass/2011_000258.png
139 | JPEGImages/2007_001416.jpg SegmentationClass/2007_001416.png
140 | JPEGImages/2010_000675.jpg SegmentationClass/2010_000675.png
141 | JPEGImages/2009_000720.jpg SegmentationClass/2009_000720.png
142 | JPEGImages/2010_004900.jpg SegmentationClass/2010_004900.png
143 | JPEGImages/2010_005898.jpg SegmentationClass/2010_005898.png
144 | JPEGImages/2007_003330.jpg SegmentationClass/2007_003330.png
145 | JPEGImages/2009_002588.jpg SegmentationClass/2009_002588.png
146 | JPEGImages/2010_004306.jpg SegmentationClass/2010_004306.png
147 | JPEGImages/2009_000553.jpg SegmentationClass/2009_000553.png
148 | JPEGImages/2007_001917.jpg SegmentationClass/2007_001917.png
149 | JPEGImages/2009_003519.jpg SegmentationClass/2009_003519.png
150 | JPEGImages/2009_004171.jpg SegmentationClass/2009_004171.png
151 | JPEGImages/2010_005952.jpg SegmentationClass/2010_005952.png
152 | JPEGImages/2007_009779.jpg SegmentationClass/2007_009779.png
153 | JPEGImages/2009_000444.jpg SegmentationClass/2009_000444.png
154 | JPEGImages/2009_003539.jpg SegmentationClass/2009_003539.png
155 | JPEGImages/2011_002222.jpg SegmentationClass/2011_002222.png
156 | JPEGImages/2010_000269.jpg SegmentationClass/2010_000269.png
157 | JPEGImages/2007_007585.jpg SegmentationClass/2007_007585.png
158 | JPEGImages/2007_009527.jpg SegmentationClass/2007_009527.png
159 | JPEGImages/2007_008927.jpg SegmentationClass/2007_008927.png
160 | JPEGImages/2011_000400.jpg SegmentationClass/2011_000400.png
161 | JPEGImages/2007_003580.jpg SegmentationClass/2007_003580.png
162 | JPEGImages/2009_003142.jpg SegmentationClass/2009_003142.png
163 | JPEGImages/2010_000978.jpg SegmentationClass/2010_000978.png
164 | JPEGImages/2007_001825.jpg SegmentationClass/2007_001825.png
165 | JPEGImages/2009_005287.jpg SegmentationClass/2009_005287.png
166 | JPEGImages/2009_000409.jpg SegmentationClass/2009_000409.png
167 | JPEGImages/2008_002064.jpg SegmentationClass/2008_002064.png
168 | JPEGImages/2008_003939.jpg SegmentationClass/2008_003939.png
169 | JPEGImages/2008_000923.jpg SegmentationClass/2008_000923.png
170 | JPEGImages/2010_002551.jpg SegmentationClass/2010_002551.png
171 | JPEGImages/2007_007698.jpg SegmentationClass/2007_007698.png
172 | JPEGImages/2009_003361.jpg SegmentationClass/2009_003361.png
173 | JPEGImages/2011_003184.jpg SegmentationClass/2011_003184.png
174 | JPEGImages/2007_004998.jpg SegmentationClass/2007_004998.png
175 | JPEGImages/2008_000290.jpg SegmentationClass/2008_000290.png
176 | JPEGImages/2011_001432.jpg SegmentationClass/2011_001432.png
177 | JPEGImages/2008_005006.jpg SegmentationClass/2008_005006.png
178 | JPEGImages/2008_001716.jpg SegmentationClass/2008_001716.png
179 | JPEGImages/2007_006899.jpg SegmentationClass/2007_006899.png
180 | JPEGImages/2010_003384.jpg SegmentationClass/2010_003384.png
181 | JPEGImages/2010_004625.jpg SegmentationClass/2010_004625.png
182 | JPEGImages/2009_001070.jpg SegmentationClass/2009_001070.png
183 | JPEGImages/2007_007890.jpg SegmentationClass/2007_007890.png
184 | JPEGImages/2008_003208.jpg SegmentationClass/2008_003208.png
185 | JPEGImages/2008_005953.jpg SegmentationClass/2008_005953.png
186 | JPEGImages/2010_001457.jpg SegmentationClass/2010_001457.png
187 | JPEGImages/2008_004259.jpg SegmentationClass/2008_004259.png
188 | JPEGImages/2011_000006.jpg SegmentationClass/2011_000006.png
189 | JPEGImages/2011_000651.jpg SegmentationClass/2011_000651.png
190 | JPEGImages/2010_001732.jpg SegmentationClass/2010_001732.png
191 | JPEGImages/2009_001615.jpg SegmentationClass/2009_001615.png
192 | JPEGImages/2011_001904.jpg SegmentationClass/2011_001904.png
193 | JPEGImages/2008_000470.jpg SegmentationClass/2008_000470.png
194 | JPEGImages/2011_000379.jpg SegmentationClass/2011_000379.png
195 | JPEGImages/2010_000810.jpg SegmentationClass/2010_000810.png
196 | JPEGImages/2009_004105.jpg SegmentationClass/2009_004105.png
197 | JPEGImages/2008_008263.jpg SegmentationClass/2008_008263.png
198 | JPEGImages/2007_006865.jpg SegmentationClass/2007_006865.png
199 | JPEGImages/2010_002166.jpg SegmentationClass/2010_002166.png
200 | JPEGImages/2008_005300.jpg SegmentationClass/2008_005300.png
201 | JPEGImages/2008_000238.jpg SegmentationClass/2008_000238.png
202 | JPEGImages/2007_000515.jpg SegmentationClass/2007_000515.png
203 | JPEGImages/2010_000195.jpg SegmentationClass/2010_000195.png
204 | JPEGImages/2010_004938.jpg SegmentationClass/2010_004938.png
205 | JPEGImages/2007_006605.jpg SegmentationClass/2007_006605.png
206 | JPEGImages/2007_009436.jpg SegmentationClass/2007_009436.png
207 | JPEGImages/2009_002567.jpg SegmentationClass/2009_002567.png
208 | JPEGImages/2009_002519.jpg SegmentationClass/2009_002519.png
209 | JPEGImages/2009_000635.jpg SegmentationClass/2009_000635.png
210 | JPEGImages/2011_000999.jpg SegmentationClass/2011_000999.png
211 | JPEGImages/2007_009594.jpg SegmentationClass/2007_009594.png
212 | JPEGImages/2009_001339.jpg SegmentationClass/2009_001339.png
213 | JPEGImages/2010_002527.jpg SegmentationClass/2010_002527.png
214 | JPEGImages/2007_002914.jpg SegmentationClass/2007_002914.png
215 | JPEGImages/2007_000733.jpg SegmentationClass/2007_000733.png
216 | JPEGImages/2007_000549.jpg SegmentationClass/2007_000549.png
217 | JPEGImages/2008_005196.jpg SegmentationClass/2008_005196.png
218 | JPEGImages/2011_001336.jpg SegmentationClass/2011_001336.png
219 | JPEGImages/2009_001502.jpg SegmentationClass/2009_001502.png
220 | JPEGImages/2009_005000.jpg SegmentationClass/2009_005000.png
221 | JPEGImages/2007_008948.jpg SegmentationClass/2007_008948.png
222 | JPEGImages/2007_004707.jpg SegmentationClass/2007_004707.png
223 | JPEGImages/2011_000637.jpg SegmentationClass/2011_000637.png
224 | JPEGImages/2011_000221.jpg SegmentationClass/2011_000221.png
225 | JPEGImages/2007_008932.jpg SegmentationClass/2007_008932.png
226 | JPEGImages/2007_001149.jpg SegmentationClass/2007_001149.png
227 | JPEGImages/2010_001261.jpg SegmentationClass/2010_001261.png
228 | JPEGImages/2011_000208.jpg SegmentationClass/2011_000208.png
229 | JPEGImages/2011_001412.jpg SegmentationClass/2011_001412.png
230 | JPEGImages/2008_001169.jpg SegmentationClass/2008_001169.png
231 | JPEGImages/2007_002967.jpg SegmentationClass/2007_002967.png
232 | JPEGImages/2008_005345.jpg SegmentationClass/2008_005345.png
233 | JPEGImages/2008_000074.jpg SegmentationClass/2008_000074.png
234 | JPEGImages/2008_004080.jpg SegmentationClass/2008_004080.png
235 | JPEGImages/2010_000772.jpg SegmentationClass/2010_000772.png
236 | JPEGImages/2010_002054.jpg SegmentationClass/2010_002054.png
237 | JPEGImages/2010_004704.jpg SegmentationClass/2010_004704.png
238 | JPEGImages/2007_007930.jpg SegmentationClass/2007_007930.png
239 | JPEGImages/2007_009709.jpg SegmentationClass/2007_009709.png
240 | JPEGImages/2011_001621.jpg SegmentationClass/2011_001621.png
241 | JPEGImages/2009_001104.jpg SegmentationClass/2009_001104.png
242 | JPEGImages/2007_007947.jpg SegmentationClass/2007_007947.png
243 | JPEGImages/2010_001618.jpg SegmentationClass/2010_001618.png
244 | JPEGImages/2007_006490.jpg SegmentationClass/2007_006490.png
245 | JPEGImages/2007_005086.jpg SegmentationClass/2007_005086.png
246 | JPEGImages/2008_003998.jpg SegmentationClass/2008_003998.png
247 | JPEGImages/2009_004919.jpg SegmentationClass/2009_004919.png
248 | JPEGImages/2007_007908.jpg SegmentationClass/2007_007908.png
249 | JPEGImages/2011_002956.jpg SegmentationClass/2011_002956.png
250 | JPEGImages/2010_004560.jpg SegmentationClass/2010_004560.png
251 | JPEGImages/2008_001498.jpg SegmentationClass/2008_001498.png
252 | JPEGImages/2009_001868.jpg SegmentationClass/2009_001868.png
253 | JPEGImages/2009_000400.jpg SegmentationClass/2009_000400.png
254 | JPEGImages/2009_002472.jpg SegmentationClass/2009_002472.png
255 | JPEGImages/2009_003860.jpg SegmentationClass/2009_003860.png
256 | JPEGImages/2008_002079.jpg SegmentationClass/2008_002079.png
257 | JPEGImages/2009_003757.jpg SegmentationClass/2009_003757.png
258 | JPEGImages/2009_000690.jpg SegmentationClass/2009_000690.png
259 | JPEGImages/2008_003373.jpg SegmentationClass/2008_003373.png
260 | JPEGImages/2009_002763.jpg SegmentationClass/2009_002763.png
261 | JPEGImages/2010_005055.jpg SegmentationClass/2010_005055.png
262 | JPEGImages/2010_003670.jpg SegmentationClass/2010_003670.png
263 | JPEGImages/2009_002885.jpg SegmentationClass/2009_002885.png
264 | JPEGImages/2008_000131.jpg SegmentationClass/2008_000131.png
265 | JPEGImages/2009_000987.jpg SegmentationClass/2009_000987.png
266 | JPEGImages/2008_003691.jpg SegmentationClass/2008_003691.png
267 | JPEGImages/2008_000103.jpg SegmentationClass/2008_000103.png
268 | JPEGImages/2008_002032.jpg SegmentationClass/2008_002032.png
269 | JPEGImages/2009_001283.jpg SegmentationClass/2009_001283.png
270 | JPEGImages/2008_003814.jpg SegmentationClass/2008_003814.png
271 | JPEGImages/2008_007201.jpg SegmentationClass/2008_007201.png
272 | JPEGImages/2011_001959.jpg SegmentationClass/2011_001959.png
273 | JPEGImages/2011_001542.jpg SegmentationClass/2011_001542.png
274 | JPEGImages/2009_001888.jpg SegmentationClass/2009_001888.png
275 | JPEGImages/2008_001926.jpg SegmentationClass/2008_001926.png
276 | JPEGImages/2011_002389.jpg SegmentationClass/2011_002389.png
277 | JPEGImages/2010_001177.jpg SegmentationClass/2010_001177.png
278 | JPEGImages/2010_004598.jpg SegmentationClass/2010_004598.png
279 | JPEGImages/2011_001536.jpg SegmentationClass/2011_001536.png
280 | JPEGImages/2011_003246.jpg SegmentationClass/2011_003246.png
281 | JPEGImages/2010_001515.jpg SegmentationClass/2010_001515.png
282 | JPEGImages/2007_007891.jpg SegmentationClass/2007_007891.png
283 | JPEGImages/2007_003788.jpg SegmentationClass/2007_003788.png
284 | JPEGImages/2008_004776.jpg SegmentationClass/2008_004776.png
285 | JPEGImages/2009_003088.jpg SegmentationClass/2009_003088.png
286 | JPEGImages/2009_000655.jpg SegmentationClass/2009_000655.png
287 | JPEGImages/2011_003121.jpg SegmentationClass/2011_003121.png
288 | JPEGImages/2007_000720.jpg SegmentationClass/2007_000720.png
289 | JPEGImages/2008_003774.jpg SegmentationClass/2008_003774.png
290 | JPEGImages/2007_009947.jpg SegmentationClass/2007_009947.png
291 | JPEGImages/2008_005600.jpg SegmentationClass/2008_005600.png
292 | JPEGImages/2011_000573.jpg SegmentationClass/2011_000573.png
293 | JPEGImages/2010_003269.jpg SegmentationClass/2010_003269.png
294 | JPEGImages/2007_005043.jpg SegmentationClass/2007_005043.png
295 | JPEGImages/2007_000121.jpg SegmentationClass/2007_000121.png
296 | JPEGImages/2009_003636.jpg SegmentationClass/2009_003636.png
297 | JPEGImages/2010_002440.jpg SegmentationClass/2010_002440.png
298 | JPEGImages/2010_003974.jpg SegmentationClass/2010_003974.png
299 | JPEGImages/2010_002973.jpg SegmentationClass/2010_002973.png
300 | JPEGImages/2011_002561.jpg SegmentationClass/2011_002561.png
301 | JPEGImages/2010_005643.jpg SegmentationClass/2010_005643.png
302 | JPEGImages/2008_000316.jpg SegmentationClass/2008_000316.png
303 | JPEGImages/2011_001895.jpg SegmentationClass/2011_001895.png
304 | JPEGImages/2009_001724.jpg SegmentationClass/2009_001724.png
305 | JPEGImages/2008_000505.jpg SegmentationClass/2008_000505.png
306 | JPEGImages/2008_003701.jpg SegmentationClass/2008_003701.png
307 | JPEGImages/2007_005951.jpg SegmentationClass/2007_005951.png
308 | JPEGImages/2009_004178.jpg SegmentationClass/2009_004178.png
309 | JPEGImages/2007_005314.jpg SegmentationClass/2007_005314.png
310 | JPEGImages/2007_003431.jpg SegmentationClass/2007_003431.png
311 | JPEGImages/2007_008526.jpg SegmentationClass/2007_008526.png
312 | JPEGImages/2009_000996.jpg SegmentationClass/2009_000996.png
313 | JPEGImages/2010_001842.jpg SegmentationClass/2010_001842.png
314 | JPEGImages/2007_007523.jpg SegmentationClass/2007_007523.png
315 | JPEGImages/2009_001972.jpg SegmentationClass/2009_001972.png
316 | JPEGImages/2010_003342.jpg SegmentationClass/2010_003342.png
317 | JPEGImages/2008_000861.jpg SegmentationClass/2008_000861.png
318 | JPEGImages/2007_006151.jpg SegmentationClass/2007_006151.png
319 | JPEGImages/2009_002423.jpg SegmentationClass/2009_002423.png
320 | JPEGImages/2011_001753.jpg SegmentationClass/2011_001753.png
321 | JPEGImages/2010_002499.jpg SegmentationClass/2010_002499.png
322 | JPEGImages/2007_005124.jpg SegmentationClass/2007_005124.png
323 | JPEGImages/2011_002585.jpg SegmentationClass/2011_002585.png
324 | JPEGImages/2011_001175.jpg SegmentationClass/2011_001175.png
325 | JPEGImages/2010_004283.jpg SegmentationClass/2010_004283.png
326 | JPEGImages/2009_005177.jpg SegmentationClass/2009_005177.png
327 | JPEGImages/2008_002697.jpg SegmentationClass/2008_002697.png
328 | JPEGImages/2007_002293.jpg SegmentationClass/2007_002293.png
329 | JPEGImages/2008_005266.jpg SegmentationClass/2008_005266.png
330 | JPEGImages/2007_006281.jpg SegmentationClass/2007_006281.png
331 | JPEGImages/2007_003118.jpg SegmentationClass/2007_003118.png
332 | JPEGImages/2009_002844.jpg SegmentationClass/2009_002844.png
333 | JPEGImages/2010_002720.jpg SegmentationClass/2010_002720.png
334 | JPEGImages/2007_000713.jpg SegmentationClass/2007_000713.png
335 | JPEGImages/2011_000652.jpg SegmentationClass/2011_000652.png
336 | JPEGImages/2010_004071.jpg SegmentationClass/2010_004071.png
337 | JPEGImages/2010_001399.jpg SegmentationClass/2010_001399.png
338 | JPEGImages/2008_007313.jpg SegmentationClass/2008_007313.png
339 | JPEGImages/2010_002778.jpg SegmentationClass/2010_002778.png
340 | JPEGImages/2007_007048.jpg SegmentationClass/2007_007048.png
341 | JPEGImages/2010_003170.jpg SegmentationClass/2010_003170.png
342 | JPEGImages/2008_000287.jpg SegmentationClass/2008_000287.png
343 | JPEGImages/2007_001027.jpg SegmentationClass/2007_001027.png
344 | JPEGImages/2007_005212.jpg SegmentationClass/2007_005212.png
345 | JPEGImages/2007_002105.jpg SegmentationClass/2007_002105.png
346 | JPEGImages/2009_000006.jpg SegmentationClass/2009_000006.png
347 | JPEGImages/2007_001073.jpg SegmentationClass/2007_001073.png
348 | JPEGImages/2008_005375.jpg SegmentationClass/2008_005375.png
349 | JPEGImages/2010_004669.jpg SegmentationClass/2010_004669.png
350 | JPEGImages/2011_002872.jpg SegmentationClass/2011_002872.png
351 | JPEGImages/2008_000584.jpg SegmentationClass/2008_000584.png
352 | JPEGImages/2009_004374.jpg SegmentationClass/2009_004374.png
353 | JPEGImages/2007_005878.jpg SegmentationClass/2007_005878.png
354 | JPEGImages/2008_000162.jpg SegmentationClass/2008_000162.png
355 | JPEGImages/2010_004960.jpg SegmentationClass/2010_004960.png
356 | JPEGImages/2009_003922.jpg SegmentationClass/2009_003922.png
357 | JPEGImages/2009_004990.jpg SegmentationClass/2009_004990.png
358 | JPEGImages/2008_002215.jpg SegmentationClass/2008_002215.png
359 | JPEGImages/2009_004446.jpg SegmentationClass/2009_004446.png
360 | JPEGImages/2007_003000.jpg SegmentationClass/2007_003000.png
361 | JPEGImages/2011_002947.jpg SegmentationClass/2011_002947.png
362 | JPEGImages/2010_004773.jpg SegmentationClass/2010_004773.png
363 | JPEGImages/2008_007242.jpg SegmentationClass/2008_007242.png
364 | JPEGImages/2011_001400.jpg SegmentationClass/2011_001400.png
365 | JPEGImages/2010_002097.jpg SegmentationClass/2010_002097.png
366 | JPEGImages/2007_000032.jpg SegmentationClass/2007_000032.png


--------------------------------------------------------------------------------
/partitions/pascal/92/labeled.txt:
--------------------------------------------------------------------------------
 1 | JPEGImages/2007_000733.jpg SegmentationClass/2007_000733.png
 2 | JPEGImages/2010_004493.jpg SegmentationClass/2010_004493.png
 3 | JPEGImages/2010_000466.jpg SegmentationClass/2010_000466.png
 4 | JPEGImages/2009_000894.jpg SegmentationClass/2009_000894.png
 5 | JPEGImages/2007_002234.jpg SegmentationClass/2007_002234.png
 6 | JPEGImages/2009_005145.jpg SegmentationClass/2009_005145.png
 7 | JPEGImages/2008_008511.jpg SegmentationClass/2008_008511.png
 8 | JPEGImages/2007_009327.jpg SegmentationClass/2007_009327.png
 9 | JPEGImages/2007_003431.jpg SegmentationClass/2007_003431.png
10 | JPEGImages/2009_005069.jpg SegmentationClass/2009_005069.png
11 | JPEGImages/2011_002585.jpg SegmentationClass/2011_002585.png
12 | JPEGImages/2009_004409.jpg SegmentationClass/2009_004409.png
13 | JPEGImages/2008_000645.jpg SegmentationClass/2008_000645.png
14 | JPEGImages/2007_005144.jpg SegmentationClass/2007_005144.png
15 | JPEGImages/2009_000895.jpg SegmentationClass/2009_000895.png
16 | JPEGImages/2010_003670.jpg SegmentationClass/2010_003670.png
17 | JPEGImages/2011_000243.jpg SegmentationClass/2011_000243.png
18 | JPEGImages/2010_000978.jpg SegmentationClass/2010_000978.png
19 | JPEGImages/2007_005797.jpg SegmentationClass/2007_005797.png
20 | JPEGImages/2011_001412.jpg SegmentationClass/2011_001412.png
21 | JPEGImages/2010_004074.jpg SegmentationClass/2010_004074.png
22 | JPEGImages/2009_001514.jpg SegmentationClass/2009_001514.png
23 | JPEGImages/2010_005217.jpg SegmentationClass/2010_005217.png
24 | JPEGImages/2010_005775.jpg SegmentationClass/2010_005775.png
25 | JPEGImages/2007_005043.jpg SegmentationClass/2007_005043.png
26 | JPEGImages/2008_000595.jpg SegmentationClass/2008_000595.png
27 | JPEGImages/2010_001933.jpg SegmentationClass/2010_001933.png
28 | JPEGImages/2011_003216.jpg SegmentationClass/2011_003216.png
29 | JPEGImages/2009_002052.jpg SegmentationClass/2009_002052.png
30 | JPEGImages/2008_001761.jpg SegmentationClass/2008_001761.png
31 | JPEGImages/2009_002362.jpg SegmentationClass/2009_002362.png
32 | JPEGImages/2010_004069.jpg SegmentationClass/2010_004069.png
33 | JPEGImages/2009_001270.jpg SegmentationClass/2009_001270.png
34 | JPEGImages/2008_006345.jpg SegmentationClass/2008_006345.png
35 | JPEGImages/2010_004900.jpg SegmentationClass/2010_004900.png
36 | JPEGImages/2009_003249.jpg SegmentationClass/2009_003249.png
37 | JPEGImages/2008_004014.jpg SegmentationClass/2008_004014.png
38 | JPEGImages/2010_004916.jpg SegmentationClass/2010_004916.png
39 | JPEGImages/2008_007242.jpg SegmentationClass/2008_007242.png
40 | JPEGImages/2009_004095.jpg SegmentationClass/2009_004095.png
41 | JPEGImages/2007_002142.jpg SegmentationClass/2007_002142.png
42 | JPEGImages/2011_000882.jpg SegmentationClass/2011_000882.png
43 | JPEGImages/2011_001015.jpg SegmentationClass/2011_001015.png
44 | JPEGImages/2010_002532.jpg SegmentationClass/2010_002532.png
45 | JPEGImages/2011_000069.jpg SegmentationClass/2011_000069.png
46 | JPEGImages/2008_000188.jpg SegmentationClass/2008_000188.png
47 | JPEGImages/2008_000105.jpg SegmentationClass/2008_000105.png
48 | JPEGImages/2009_004790.jpg SegmentationClass/2009_004790.png
49 | JPEGImages/2011_002488.jpg SegmentationClass/2011_002488.png
50 | JPEGImages/2009_001100.jpg SegmentationClass/2009_001100.png
51 | JPEGImages/2010_004560.jpg SegmentationClass/2010_004560.png
52 | JPEGImages/2010_004773.jpg SegmentationClass/2010_004773.png
53 | JPEGImages/2010_003093.jpg SegmentationClass/2010_003093.png
54 | JPEGImages/2007_009082.jpg SegmentationClass/2007_009082.png
55 | JPEGImages/2011_001536.jpg SegmentationClass/2011_001536.png
56 | JPEGImages/2008_000470.jpg SegmentationClass/2008_000470.png
57 | JPEGImages/2008_000284.jpg SegmentationClass/2008_000284.png
58 | JPEGImages/2010_005064.jpg SegmentationClass/2010_005064.png
59 | JPEGImages/2008_000336.jpg SegmentationClass/2008_000336.png
60 | JPEGImages/2008_008462.jpg SegmentationClass/2008_008462.png
61 | JPEGImages/2007_001960.jpg SegmentationClass/2007_001960.png
62 | JPEGImages/2007_004951.jpg SegmentationClass/2007_004951.png
63 | JPEGImages/2008_004112.jpg SegmentationClass/2008_004112.png
64 | JPEGImages/2009_001339.jpg SegmentationClass/2009_001339.png
65 | JPEGImages/2008_005074.jpg SegmentationClass/2008_005074.png
66 | JPEGImages/2010_003010.jpg SegmentationClass/2010_003010.png
67 | JPEGImages/2011_000122.jpg SegmentationClass/2011_000122.png
68 | JPEGImages/2010_001279.jpg SegmentationClass/2010_001279.png
69 | JPEGImages/2010_004363.jpg SegmentationClass/2010_004363.png
70 | JPEGImages/2011_002561.jpg SegmentationClass/2011_002561.png
71 | JPEGImages/2007_005989.jpg SegmentationClass/2007_005989.png
72 | JPEGImages/2010_003799.jpg SegmentationClass/2010_003799.png
73 | JPEGImages/2007_006004.jpg SegmentationClass/2007_006004.png
74 | JPEGImages/2007_009630.jpg SegmentationClass/2007_009630.png
75 | JPEGImages/2008_001159.jpg SegmentationClass/2008_001159.png
76 | JPEGImages/2007_007003.jpg SegmentationClass/2007_007003.png
77 | JPEGImages/2008_001610.jpg SegmentationClass/2008_001610.png
78 | JPEGImages/2007_007415.jpg SegmentationClass/2007_007415.png
79 | JPEGImages/2007_002273.jpg SegmentationClass/2007_002273.png
80 | JPEGImages/2010_003157.jpg SegmentationClass/2010_003157.png
81 | JPEGImages/2011_002107.jpg SegmentationClass/2011_002107.png
82 | JPEGImages/2010_005982.jpg SegmentationClass/2010_005982.png
83 | JPEGImages/2007_005878.jpg SegmentationClass/2007_005878.png
84 | JPEGImages/2011_000457.jpg SegmentationClass/2011_000457.png
85 | JPEGImages/2008_000832.jpg SegmentationClass/2008_000832.png
86 | JPEGImages/2008_003769.jpg SegmentationClass/2008_003769.png
87 | JPEGImages/2010_002047.jpg SegmentationClass/2010_002047.png
88 | JPEGImages/2009_002387.jpg SegmentationClass/2009_002387.png
89 | JPEGImages/2009_002117.jpg SegmentationClass/2009_002117.png
90 | JPEGImages/2011_002920.jpg SegmentationClass/2011_002920.png
91 | JPEGImages/2010_004760.jpg SegmentationClass/2010_004760.png
92 | JPEGImages/2008_004822.jpg SegmentationClass/2008_004822.png


--------------------------------------------------------------------------------
/tools/train.sh:
--------------------------------------------------------------------------------
 1 | #!/bin/bash
 2 | now=$(date +"%Y%m%d_%H%M%S")
 3 | 
 4 | config=configs/pascal.yaml
 5 | labeled_id_path=partitions/pascal/92/labeled.txt
 6 | unlabeled_id_path=partitions/pascal/92/unlabeled.txt
 7 | save_path=exp/pascal/92/corrmatch
 8 | #config=configs/cityscapes.yaml
 9 | #labeled_id_path=partitions/cityscapes/1_4/labeled.txt
10 | #unlabeled_id_path=partitions/cityscapes/1_4/unlabeled.txt
11 | #save_path=exp/cityscapes/1_4/corrmatch
12 | 
13 | mkdir -p $save_path
14 | 
15 | python -m torch.distributed.launch \
16 |     --nproc_per_node=$1 \
17 |     --master_addr=localhost \
18 |     --master_port=$2 \
19 |     corrmatch.py \
20 |     --config=$config --labeled-id-path $labeled_id_path --unlabeled-id-path $unlabeled_id_path \
21 |     --save-path $save_path --port $2 2>&1 | tee $save_path/$now.txt


--------------------------------------------------------------------------------
/tools/val.sh:
--------------------------------------------------------------------------------
 1 | #!/bin/bash
 2 | now=$(date +"%Y%m%d_%H%M%S")
 3 | 
 4 | config=configs/pascal.yaml
 5 | checkpoint_path=your/checkpoint/path
 6 | 
 7 | python -m torch.distributed.launch \
 8 |     --nproc_per_node=$1 \
 9 |     --master_addr=localhost \
10 |     --master_port=$2 \
11 |     evaluate.py \
12 |     --config=$config --checkpoint_path $checkpoint_path


--------------------------------------------------------------------------------
/tools/vis.sh:
--------------------------------------------------------------------------------
 1 | #!/bin/bash
 2 | now=$(date +"%Y%m%d_%H%M%S")
 3 | 
 4 | config=configs/pascal.yaml
 5 | labeled_id_path=partitions/pascal/92/labeled.txt
 6 | unlabeled_id_path=partitions/pascal/92/unlabeled.txt
 7 | save_path=exp/pascal/92/unimatch
 8 | #config=configs/cityscapes.yaml
 9 | #labeled_id_path=partitions/cityscapes/1_4/labeled.txt
10 | #unlabeled_id_path=partitions/cityscapes/1_4/unlabeled.txt
11 | #save_path=exp/cityscapes/1_4/unimatch
12 | 
13 | mkdir -p $save_path
14 | 
15 | python -m torch.distributed.launch \
16 |     --nproc_per_node=$1 \
17 |     --master_addr=localhost \
18 |     --master_port=$2 \
19 |     vis.py \
20 |     --config=$config --labeled-id-path $labeled_id_path --unlabeled-id-path $unlabeled_id_path \
21 |     --save-path $save_path --port $2 2>&1 | tee $save_path/$now.txt


--------------------------------------------------------------------------------
/tools/vis_mask.sh:
--------------------------------------------------------------------------------
 1 | #!/bin/bash
 2 | now=$(date +"%Y%m%d_%H%M%S")
 3 | 
 4 | config=configs/pascal.yaml
 5 | labeled_id_path=partitions/pascal/92/labeled.txt
 6 | unlabeled_id_path=partitions/pascal/92/unlabeled.txt
 7 | save_path=exp/pascal/92/unimatch
 8 | #config=configs/cityscapes.yaml
 9 | #labeled_id_path=partitions/cityscapes/1_4/labeled.txt
10 | #unlabeled_id_path=partitions/cityscapes/1_4/unlabeled.txt
11 | #save_path=exp/cityscapes/1_4/unimatch
12 | 
13 | mkdir -p $save_path
14 | 
15 | python -m torch.distributed.launch \
16 |     --nproc_per_node=$1 \
17 |     --master_addr=localhost \
18 |     --master_port=$2 \
19 |     vis_mask.py \
20 |     --config=$config --labeled-id-path $labeled_id_path --unlabeled-id-path $unlabeled_id_path \
21 |     --save-path $save_path --port $2 2>&1 | tee $save_path/$now.txt


--------------------------------------------------------------------------------
/util/dist_helper.py:
--------------------------------------------------------------------------------
 1 | import os
 2 | import subprocess
 3 | 
 4 | import torch
 5 | import torch.distributed as dist
 6 | 
 7 | 
 8 | def setup_distributed(backend="nccl", port=None):
 9 |     """AdaHessian Optimizer
10 |     Lifted from https://github.com/BIGBALLON/distribuuuu/blob/master/distribuuuu/utils.py
11 |     Originally licensed MIT, Copyright (c) 2020 Wei Li
12 |     """
13 |     num_gpus = torch.cuda.device_count()
14 | 
15 |     if "SLURM_JOB_ID" in os.environ:
16 |         rank = int(os.environ["SLURM_PROCID"])
17 |         world_size = int(os.environ["SLURM_NTASKS"])
18 |         node_list = os.environ["SLURM_NODELIST"]
19 |         addr = subprocess.getoutput(f"scontrol show hostname {node_list} | head -n1")
20 |         # specify master port
21 |         if port is not None:
22 |             os.environ["MASTER_PORT"] = str(port)
23 |         elif "MASTER_PORT" not in os.environ:
24 |             os.environ["MASTER_PORT"] = "10685"
25 |         if "MASTER_ADDR" not in os.environ:
26 |             os.environ["MASTER_ADDR"] = addr
27 |         os.environ["WORLD_SIZE"] = str(world_size)
28 |         os.environ["LOCAL_RANK"] = str(rank % num_gpus)
29 |         os.environ["RANK"] = str(rank)
30 |     else:
31 |         rank = int(os.environ["RANK"])
32 |         world_size = int(os.environ["WORLD_SIZE"])
33 | 
34 |     torch.cuda.set_device(rank % num_gpus)
35 | 
36 |     dist.init_process_group(
37 |         backend=backend,
38 |         world_size=world_size,
39 |         rank=rank,
40 |     )
41 |     return rank, world_size
42 | 


--------------------------------------------------------------------------------
/util/mesh_helper.py:
--------------------------------------------------------------------------------
 1 | import torch
 2 | import torch.nn.functional as F
 3 | from einops.layers.torch import Rearrange, Reduce
 4 | from einops import rearrange, reduce, asnumpy, parse_shape
 5 | 
 6 | a = torch.randn([4, 21, 65, 65])
 7 | b = torch.randn([4, 21, 65, 65])
 8 | c = torch.randn([4, 21, 65, 65])
 9 | h, w = a.shape[-2:]
10 | h_local, w_local = 3, 3
11 | for i in torch.arange(h - 2) + 1:
12 |     for j in torch.arange(w - 2) + 1:
13 |         a_local = a[:, :, i - 1:i + 2, j - 1:j + 2]
14 |         a_local = rearrange(a_local, 'n c h w -> n c (h w)')
15 |         b_local = b[:, :, i - 1:i + 2, j - 1:j + 2]
16 |         b_local = rearrange(b_local, 'n c h w -> n c (h w)')
17 |         corr_local = torch.matmul(a_local.transpose(1, 2), b_local) / torch.sqrt(torch.tensor(a_local.shape[1]).float())
18 |         print(corr_local.shape)
19 |         exit()
20 |         c_local = c[:, :, i - 1:i + 2, j - 1:j + 2]
21 |         c_local = rearrange(c_local, 'n c h w -> n c (h w)')
22 |         out_local = rearrange(torch.matmul(c_local, F.softmax(corr_local, dim=-1)), 'n c (h w) -> n c h w', h=h_local,
23 |                               w=w_local)
24 |         print(out_local.shape)
25 | 


--------------------------------------------------------------------------------
/util/ohem.py:
--------------------------------------------------------------------------------
 1 | import torch
 2 | from torch import nn
 3 | import torch.nn.functional as F
 4 | import numpy as np
 5 | 
 6 | 
 7 | # see https://github.com/charlesCXK/TorchSemiSeg/blob/main/furnace/seg_opr/loss_opr.py
 8 | class ProbOhemCrossEntropy2d(nn.Module):
 9 |     def __init__(self, ignore_index, reduction='mean', thresh=0.7, min_kept=256,
10 |                  down_ratio=1, use_weight=False):
11 |         super(ProbOhemCrossEntropy2d, self).__init__()
12 |         self.ignore_index = ignore_index
13 |         self.thresh = float(thresh)
14 |         self.min_kept = int(min_kept)
15 |         self.down_ratio = down_ratio
16 |         if use_weight:
17 |             weight = torch.FloatTensor(
18 |                 [0.8373, 0.918, 0.866, 1.0345, 1.0166, 0.9969, 0.9754, 1.0489,
19 |                  0.8786, 1.0023, 0.9539, 0.9843, 1.1116, 0.9037, 1.0865, 1.0955,
20 |                  1.0865, 1.1529, 1.0507])
21 |             self.criterion = torch.nn.CrossEntropyLoss(reduction=reduction,
22 |                                                        weight=weight,
23 |                                                        ignore_index=ignore_index)
24 |         else:
25 |             self.criterion = torch.nn.CrossEntropyLoss(reduction=reduction,
26 |                                                        ignore_index=ignore_index)
27 | 
28 |     def forward(self, pred, target):
29 |         b, c, h, w = pred.size()
30 |         target = target.view(-1)
31 |         valid_mask = target.ne(self.ignore_index)
32 |         target = target * valid_mask.long()
33 |         num_valid = valid_mask.sum()
34 | 
35 |         prob = F.softmax(pred, dim=1)
36 |         prob = (prob.transpose(0, 1)).reshape(c, -1)
37 | 
38 |         if self.min_kept > num_valid:
39 |             pass
40 |         elif num_valid > 0:
41 |             prob = prob.masked_fill_(~valid_mask, 1)
42 |             mask_prob = prob[
43 |                 target, torch.arange(len(target), dtype=torch.long)]
44 |             threshold = self.thresh
45 |             if self.min_kept > 0:
46 |                 index = mask_prob.argsort()
47 |                 threshold_index = index[min(len(index), self.min_kept) - 1]
48 |                 if mask_prob[threshold_index] > self.thresh:
49 |                     threshold = mask_prob[threshold_index]
50 |                 kept_mask = mask_prob.le(threshold)
51 |                 target = target * kept_mask.long()
52 |                 valid_mask = valid_mask * kept_mask
53 | 
54 |         target = target.masked_fill_(~valid_mask, self.ignore_index)
55 |         target = target.view(b, h, w)
56 | 
57 |         return self.criterion(pred, target)
58 | 


--------------------------------------------------------------------------------
/util/thresh_helper.py:
--------------------------------------------------------------------------------
 1 | import torch
 2 | import torch.distributed as dist
 3 | 
 4 | 
 5 | class ThreshController:
 6 |     def __init__(self, nclass, momentum, thresh_init=0.85):
 7 | 
 8 |         self.thresh_global = torch.tensor(thresh_init).cuda()
 9 |         self.momentum = momentum
10 |         self.nclass = nclass
11 |         self.gpu_num = dist.get_world_size()
12 | 
13 |     def new_global_mask_pooling(self, pred, ignore_mask=None):
14 |         return_dict = {}
15 |         n, c, h, w = pred.shape
16 |         pred_gather = torch.zeros([n * self.gpu_num, c, h, w]).cuda()
17 |         dist.all_gather_into_tensor(pred_gather, pred)
18 |         pred = pred_gather
19 |         if ignore_mask is not None:
20 |             ignore_mask_gather = torch.zeros([n * self.gpu_num, h, w]).cuda().long()
21 |             dist.all_gather_into_tensor(ignore_mask_gather, ignore_mask)
22 |             ignore_mask = ignore_mask_gather
23 |         mask_pred = torch.argmax(pred, dim=1)
24 |         pred_softmax = pred.softmax(dim=1)
25 |         pred_conf = pred_softmax.max(dim=1)[0]
26 |         unique_cls = torch.unique(mask_pred)
27 |         cls_num = len(unique_cls)
28 |         new_global = 0.0
29 |         for cls in unique_cls:
30 |             cls_map = (mask_pred == cls)
31 |             if ignore_mask is not None:
32 |                 cls_map *= (ignore_mask != 255)
33 |             if cls_map.sum() == 0:
34 |                 cls_num -= 1
35 |                 continue
36 |             pred_conf_cls_all = pred_conf[cls_map]
37 |             cls_max_conf = pred_conf_cls_all.max()
38 |             new_global += cls_max_conf
39 |         if cls_num > 0:
40 |             return_dict['new_global'] = new_global / cls_num
41 |         else:
42 |             return_dict['new_global'] = None
43 | 
44 |         return return_dict
45 | 
46 |     def thresh_update(self, pred, ignore_mask=None, update_g=False):
47 |         thresh = self.new_global_mask_pooling(pred, ignore_mask)
48 |         if update_g and thresh['new_global'] is not None:
49 |             self.thresh_global = self.momentum * self.thresh_global + (1 - self.momentum) * thresh['new_global']
50 | 
51 |     def get_thresh_global(self):
52 |         return self.thresh_global
53 | 


--------------------------------------------------------------------------------
/util/utils.py:
--------------------------------------------------------------------------------
  1 | import numpy as np
  2 | import logging
  3 | import os
  4 | 
  5 | 
  6 | def count_params(model):
  7 |     param_num = sum(p.numel() for p in model.parameters())
  8 |     return param_num / 1e6
  9 | 
 10 | 
 11 | def color_map(dataset='pascal'):
 12 |     cmap = np.zeros((256, 3), dtype='uint8')
 13 | 
 14 |     if dataset == 'pascal' or dataset == 'coco':
 15 |         def bitget(byteval, idx):
 16 |             return (byteval & (1 << idx)) != 0
 17 | 
 18 |         for i in range(256):
 19 |             r = g = b = 0
 20 |             c = i
 21 |             for j in range(8):
 22 |                 r = r | (bitget(c, 0) << 7-j)
 23 |                 g = g | (bitget(c, 1) << 7-j)
 24 |                 b = b | (bitget(c, 2) << 7-j)
 25 |                 c = c >> 3
 26 | 
 27 |             cmap[i] = np.array([r, g, b])
 28 | 
 29 |     elif dataset == 'cityscapes':
 30 |         cmap[0] = np.array([128, 64, 128])
 31 |         cmap[1] = np.array([244, 35, 232])
 32 |         cmap[2] = np.array([70, 70, 70])
 33 |         cmap[3] = np.array([102, 102, 156])
 34 |         cmap[4] = np.array([190, 153, 153])
 35 |         cmap[5] = np.array([153, 153, 153])
 36 |         cmap[6] = np.array([250, 170, 30])
 37 |         cmap[7] = np.array([220, 220, 0])
 38 |         cmap[8] = np.array([107, 142, 35])
 39 |         cmap[9] = np.array([152, 251, 152])
 40 |         cmap[10] = np.array([70, 130, 180])
 41 |         cmap[11] = np.array([220, 20, 60])
 42 |         cmap[12] = np.array([255,  0,  0])
 43 |         cmap[13] = np.array([0,  0, 142])
 44 |         cmap[14] = np.array([0,  0, 70])
 45 |         cmap[15] = np.array([0, 60, 100])
 46 |         cmap[16] = np.array([0, 80, 100])
 47 |         cmap[17] = np.array([0,  0, 230])
 48 |         cmap[18] = np.array([119, 11, 32])
 49 | 
 50 |     return cmap
 51 | 
 52 | 
 53 | class AverageMeter(object):
 54 |     """Computes and stores the average and current value"""
 55 | 
 56 |     def __init__(self, length=0):
 57 |         self.length = length
 58 |         self.reset()
 59 | 
 60 |     def reset(self):
 61 |         if self.length > 0:
 62 |             self.history = []
 63 |         else:
 64 |             self.count = 0
 65 |             self.sum = 0.0
 66 |         self.val = 0.0
 67 |         self.avg = 0.0
 68 | 
 69 |     def update(self, val, num=1):
 70 |         if self.length > 0:
 71 |             # currently assert num==1 to avoid bad usage, refine when there are some explict requirements
 72 |             assert num == 1
 73 |             self.history.append(val)
 74 |             if len(self.history) > self.length:
 75 |                 del self.history[0]
 76 | 
 77 |             self.val = self.history[-1]
 78 |             self.avg = np.mean(self.history)
 79 |         else:
 80 |             self.val = val
 81 |             self.sum += val * num
 82 |             self.count += num
 83 |             self.avg = self.sum / self.count
 84 | 
 85 | 
 86 | def intersectionAndUnion(output, target, K, ignore_index=255):
 87 |     # 'K' classes, output and target sizes are N or N * L or N * H * W, each value in range 0 to K - 1.
 88 |     assert output.ndim in [1, 2, 3]
 89 |     assert output.shape == target.shape
 90 |     output = output.reshape(output.size).copy()
 91 |     target = target.reshape(target.size)
 92 |     output[np.where(target == ignore_index)[0]] = ignore_index
 93 |     intersection = output[np.where(output == target)[0]]
 94 |     area_intersection, _ = np.histogram(intersection, bins=np.arange(K + 1))
 95 |     area_output, _ = np.histogram(output, bins=np.arange(K + 1))
 96 |     area_target, _ = np.histogram(target, bins=np.arange(K + 1))
 97 |     area_union = area_output + area_target - area_intersection
 98 |     return area_intersection, area_union, area_target
 99 | 
100 | 
101 | logs = set()
102 | 
103 | 
104 | def init_log(name, level=logging.INFO):
105 |     if (name, level) in logs:
106 |         return
107 |     logs.add((name, level))
108 |     logger = logging.getLogger(name)
109 |     logger.setLevel(level)
110 |     ch = logging.StreamHandler()
111 |     ch.setLevel(level)
112 |     if "SLURM_PROCID" in os.environ:
113 |         rank = int(os.environ["SLURM_PROCID"])
114 |         logger.addFilter(lambda record: rank == 0)
115 |     else:
116 |         rank = 0
117 |     format_str = "[%(asctime)s][%(levelname)8s] %(message)s"
118 |     formatter = logging.Formatter(format_str)
119 |     ch.setFormatter(formatter)
120 |     logger.addHandler(ch)
121 |     return logger
122 | 


--------------------------------------------------------------------------------
/vis.py:
--------------------------------------------------------------------------------
  1 | import argparse
  2 | import os
  3 | 
  4 | import torch
  5 | import torch.distributed as dist
  6 | import torch.nn.functional as F
  7 | from torch.utils.data import DataLoader
  8 | import yaml
  9 | import matplotlib
 10 | 
 11 | matplotlib.use('agg')
 12 | import matplotlib.pyplot as plt
 13 | import seaborn as sns
 14 | from einops import rearrange
 15 | 
 16 | from dataset.semi import SemiDataset
 17 | from model.semseg.deeplabv3plus_vis import DeepLabV3Plus
 18 | from util.dist_helper import setup_distributed
 19 | 
 20 | parser = argparse.ArgumentParser(description='Semi-Supervised Semantic Segmentation')
 21 | parser.add_argument('--config', type=str, required=True)
 22 | parser.add_argument('--labeled-id-path', type=str, required=True)
 23 | parser.add_argument('--unlabeled-id-path', type=str, required=True)
 24 | parser.add_argument('--save-path', type=str, required=True)
 25 | parser.add_argument('--local_rank', default=0, type=int)
 26 | parser.add_argument('--port', default=None, type=int)
 27 | args = parser.parse_args()
 28 | 
 29 | 
 30 | def corr_compute(feat):
 31 |     Q = rearrange(feat, 'n c h w -> n c (h w)')
 32 |     K = rearrange(feat, 'n c h w -> n c (h w)')
 33 |     corr_map = torch.matmul(Q.transpose(1, 2), K) / torch.sqrt(torch.tensor(Q.shape[1]).float())
 34 |     corr_map = F.softmax(corr_map, dim=-1)
 35 |     return corr_map
 36 | 
 37 | 
 38 | def corr2heatmap_save(corr_map_i, pixel_index):
 39 |     temp_map = rearrange(corr_map_i[pixel_index], '(h w) -> 1 1 h w', h=c4_feats_i.shape[-2], w=c4_feats_i.shape[-1])
 40 |     temp_map = F.interpolate(temp_map, (h, w), mode='bilinear')
 41 |     temp_map = rearrange(temp_map, '1 1 h w -> h w')
 42 |     range_ = torch.max(temp_map) - torch.min(temp_map)
 43 |     temp_map = (- torch.min(temp_map) + temp_map) / range_
 44 |     plt.figure(figsize=(w / 50, h / 50))
 45 |     heat_map = sns.heatmap(temp_map.cpu().numpy(), cbar=False)
 46 |     heat_map = heat_map.get_figure()
 47 |     plt.axis('off')
 48 | 
 49 |     heat_map.savefig('temp/{}/{}_{}_corr.png'.format(file_name, int(pixel_index / c4_feats_i.shape[-1]) * int(
 50 |         h / c4_feats_i.shape[-2]), int(pixel_index % c4_feats_i.shape[-1]) * int(w / c4_feats_i.shape[-1])),
 51 |                      pad_inches=0, bbox_inches='tight')
 52 |     plt.clf()
 53 |     plt.close()
 54 |     del heat_map, temp_map
 55 | 
 56 | 
 57 | rank, word_size = setup_distributed(port=args.port)
 58 | 
 59 | cfg = yaml.load(open('configs/pascal.yaml', "r"), Loader=yaml.Loader)
 60 | 
 61 | model = DeepLabV3Plus(cfg)
 62 | model.load_state_dict(torch.load('Your/checkpoint/path'))
 63 | model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model)
 64 | model.cuda()
 65 | 
 66 | local_rank = int(os.environ["LOCAL_RANK"])
 67 | model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[local_rank],
 68 |                                                   output_device=local_rank, find_unused_parameters=False)
 69 | 
 70 | valset = SemiDataset(cfg['dataset'], cfg['data_root'], 'val')
 71 | valsampler = torch.utils.data.distributed.DistributedSampler(valset, shuffle=False)
 72 | valloader = DataLoader(valset, batch_size=1, pin_memory=True, num_workers=4,
 73 |                        drop_last=False, sampler=valsampler)
 74 | 
 75 | model.eval()
 76 | 
 77 | with torch.no_grad():
 78 |     for img, mask, ids, img_ori in valloader:
 79 |         dist.barrier()
 80 |         img = img.cuda()
 81 |         b, _, h, w = img.shape
 82 |         res = model(img, use_corr=True)
 83 |         pred = res['out']
 84 |         pred_mask = pred.argmax(dim=1)
 85 |         pred_conf = pred.softmax(dim=1).max(dim=1)[0]
 86 |         pred_conf_fliter = (pred_conf <= 0.95)
 87 |         mask_fliter = pred_mask.clone()
 88 |         mask_fliter[pred_conf_fliter] = 255
 89 |         corr_map = res['corr_map']
 90 |         c4_feats = res['c4']
 91 | 
 92 |         for i in range(pred_mask.shape[0]):
 93 |             file_name = ids[i].split(' ')[0].split('/')[1].split('.')[0]
 94 |             if not os.path.exists('temp/{}'.format(file_name)):
 95 |                 os.mkdir('temp/{}'.format(file_name))
 96 |             print(file_name)
 97 |             mask_pred_i = pred_mask[i]
 98 |             mask_i = mask[i]
 99 |             mask_filter_i = mask_fliter[i]
100 |             corr_map_i = corr_map[i]
101 |             c4_feats_i = c4_feats[i]
102 |             for pixel_index in range(corr_map_i.shape[0]):
103 |                 corr2heatmap_save(corr_map_i, pixel_index)
104 | 


--------------------------------------------------------------------------------
/vis_mask.py:
--------------------------------------------------------------------------------
  1 | import argparse
  2 | import os
  3 | 
  4 | import torch
  5 | import torch.distributed as dist
  6 | import torch.nn.functional as F
  7 | from torch.utils.data import DataLoader
  8 | import yaml
  9 | import matplotlib
 10 | 
 11 | matplotlib.use('agg')
 12 | import matplotlib.pyplot as plt
 13 | from einops import rearrange
 14 | import numpy as np
 15 | from PIL import Image
 16 | 
 17 | from dataset.semi import SemiDataset
 18 | from model.semseg.deeplabv3plus_vis import DeepLabV3Plus
 19 | from util.dist_helper import setup_distributed
 20 | 
 21 | parser = argparse.ArgumentParser(description='Semi-Supervised Semantic Segmentation')
 22 | parser.add_argument('--config', type=str, required=True)
 23 | parser.add_argument('--labeled-id-path', type=str, required=True)
 24 | parser.add_argument('--unlabeled-id-path', type=str, required=True)
 25 | parser.add_argument('--save-path', type=str, required=True)
 26 | parser.add_argument('--local_rank', default=0, type=int)
 27 | parser.add_argument('--port', default=None, type=int)
 28 | args = parser.parse_args()
 29 | 
 30 | def color_map(dataset='pascal'):
 31 |     cmap = np.zeros((256, 3), dtype='uint8')
 32 | 
 33 |     if dataset == 'pascal' or dataset == 'coco':
 34 |         def bitget(byteval, idx):
 35 |             return (byteval & (1 << idx)) != 0
 36 | 
 37 |         for i in range(256):
 38 |             r = g = b = 0
 39 |             c = i
 40 |             for j in range(8):
 41 |                 r = r | (bitget(c, 0) << 7-j)
 42 |                 g = g | (bitget(c, 1) << 7-j)
 43 |                 b = b | (bitget(c, 2) << 7-j)
 44 |                 c = c >> 3
 45 | 
 46 |             cmap[i] = np.array([r, g, b])
 47 | 
 48 |     elif dataset == 'cityscapes':
 49 |         cmap[0] = np.array([128, 64, 128])
 50 |         cmap[1] = np.array([244, 35, 232])
 51 |         cmap[2] = np.array([70, 70, 70])
 52 |         cmap[3] = np.array([102, 102, 156])
 53 |         cmap[4] = np.array([190, 153, 153])
 54 |         cmap[5] = np.array([153, 153, 153])
 55 |         cmap[6] = np.array([250, 170, 30])
 56 |         cmap[7] = np.array([220, 220, 0])
 57 |         cmap[8] = np.array([107, 142, 35])
 58 |         cmap[9] = np.array([152, 251, 152])
 59 |         cmap[10] = np.array([70, 130, 180])
 60 |         cmap[11] = np.array([220, 20, 60])
 61 |         cmap[12] = np.array([255,  0,  0])
 62 |         cmap[13] = np.array([0,  0, 142])
 63 |         cmap[14] = np.array([0,  0, 70])
 64 |         cmap[15] = np.array([0, 60, 100])
 65 |         cmap[16] = np.array([0, 80, 100])
 66 |         cmap[17] = np.array([0,  0, 230])
 67 |         cmap[18] = np.array([119, 11, 32])
 68 | 
 69 |     return cmap
 70 | 
 71 | 
 72 | rank, word_size = setup_distributed(port=args.port)
 73 | 
 74 | cfg = yaml.load(open('configs/pascal.yaml', "r"), Loader=yaml.Loader)
 75 | 
 76 | model = DeepLabV3Plus(cfg)
 77 | model.load_state_dict(torch.load('Your/checkpoint/path'))
 78 | model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model)
 79 | model.cuda()
 80 | 
 81 | local_rank = int(os.environ["LOCAL_RANK"])
 82 | model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[local_rank],
 83 |                                                   output_device=local_rank, find_unused_parameters=False)
 84 | 
 85 | valset = SemiDataset(cfg['dataset'], cfg['data_root'], 'val')
 86 | valsampler = torch.utils.data.distributed.DistributedSampler(valset, shuffle=False)
 87 | valloader = DataLoader(valset, batch_size=1, pin_memory=True, num_workers=4,
 88 |                        drop_last=False, sampler=valsampler)
 89 | 
 90 | model.eval()
 91 | if local_rank == 0:
 92 |     if not os.path.exists('visual'):
 93 |         os.mkdir('visual')
 94 | 
 95 | with torch.no_grad():
 96 |     for img, mask, ids, img_ori in valloader:
 97 |         dist.barrier()
 98 | 
 99 |         img = img.cuda()
100 |         b, _, h, w = img.shape
101 |         res = model(img, use_corr=True)
102 |         pred = res['out']
103 |         pred_mask = pred.argmax(dim=1)
104 |         pred_conf = pred.softmax(dim=1).max(dim=1)[0]
105 |         # take 0.95 as an example
106 |         pred_conf_fliter = (pred_conf <= 0.95)
107 |         mask_fliter = pred_mask.clone()
108 |         mask_fliter[pred_conf_fliter] = 255
109 |         for i in range(pred_mask.shape[0]):
110 |             file_name = ids[i].split(' ')[0].split('/')[1].split('.')[0]
111 |             if not os.path.exists('visual/{}'.format(file_name)):
112 |                 os.mkdir('visual/{}'.format(file_name))
113 |             print(file_name)
114 |             mask_pred_i = pred_mask[i]
115 |             mask_i = mask[i]
116 |             mask_filter_i = mask_fliter[i]
117 |             mask_i = Image.fromarray(mask_i.cpu().numpy().astype(np.uint8), mode='P')
118 |             mask_pred_i = Image.fromarray(mask_pred_i.cpu().numpy().astype(np.uint8), mode='P')
119 |             mask_filter_i = Image.fromarray(mask_filter_i.cpu().numpy().astype(np.uint8), mode='P')
120 |             platte = color_map()
121 |             mask_i.putpalette(platte)
122 |             mask_pred_i.putpalette(platte)
123 |             mask_filter_i.putpalette(platte)
124 |             mask_i.save('visual/{}/mask_gt.png'.format(file_name))
125 |             mask_pred_i.save('visual/{}/mask_pred.png'.format(file_name))
126 |             mask_filter_i.save('visual/{}/mask_filter.png'.format(file_name))
127 | 
128 | 


--------------------------------------------------------------------------------