├── .gitignore
├── README.md
├── gen_pgt.py
├── infer_AuxAff.py
├── iter_learn.sh
├── main.py
├── mis
    ├── example.jpg
    ├── framework.jpg
    ├── framework2.jpg
    ├── sota_coco.png
    └── sota_voc.png
├── network
    ├── AuxSegNet.py
    ├── non_local.py
    └── resnet38d.py
├── prerequisite.txt
├── step
    ├── gen_pgt.py
    ├── infer_AuxAff.py
    └── train_AuxAff.py
├── tool
    ├── exutils.py
    ├── imutils.py
    ├── pyutils.py
    └── torchutils.py
├── train_AuxAff.py
└── voc12
    ├── cls_labels.npy
    ├── test.txt
    ├── train_aug_id.txt
    └── val_id.txt


/.gitignore:
--------------------------------------------------------------------------------
  1 | # Byte-compiled / optimized / DLL files
  2 | __pycache__/
  3 | *.py[cod]
  4 | *$py.class
  5 | 
  6 | # C extensions
  7 | *.so
  8 | 
  9 | # Distribution / packaging
 10 | .Python
 11 | build/
 12 | develop-eggs/
 13 | dist/
 14 | downloads/
 15 | eggs/
 16 | .eggs/
 17 | lib/
 18 | lib64/
 19 | parts/
 20 | sdist/
 21 | var/
 22 | wheels/
 23 | pip-wheel-metadata/
 24 | share/python-wheels/
 25 | *.egg-info/
 26 | .installed.cfg
 27 | *.egg
 28 | MANIFEST
 29 | 
 30 | # PyInstaller
 31 | #  Usually these files are written by a python script from a template
 32 | #  before PyInstaller builds the exe, so as to inject date/other infos into it.
 33 | *.manifest
 34 | *.spec
 35 | 
 36 | # Installer logs
 37 | pip-log.txt
 38 | pip-delete-this-directory.txt
 39 | 
 40 | # Unit test / coverage reports
 41 | htmlcov/
 42 | .tox/
 43 | .nox/
 44 | .coverage
 45 | .coverage.*
 46 | .cache
 47 | nosetests.xml
 48 | coverage.xml
 49 | *.cover
 50 | *.py,cover
 51 | .hypothesis/
 52 | .pytest_cache/
 53 | 
 54 | # Translations
 55 | *.mo
 56 | *.pot
 57 | 
 58 | # Django stuff:
 59 | *.log
 60 | local_settings.py
 61 | db.sqlite3
 62 | db.sqlite3-journal
 63 | 
 64 | # Flask stuff:
 65 | instance/
 66 | .webassets-cache
 67 | 
 68 | # Scrapy stuff:
 69 | .scrapy
 70 | 
 71 | # Sphinx documentation
 72 | docs/_build/
 73 | 
 74 | # PyBuilder
 75 | target/
 76 | 
 77 | # Jupyter Notebook
 78 | .ipynb_checkpoints
 79 | 
 80 | # IPython
 81 | profile_default/
 82 | ipython_config.py
 83 | 
 84 | # pyenv
 85 | .python-version
 86 | 
 87 | # pipenv
 88 | #   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
 89 | #   However, in case of collaboration, if having platform-specific dependencies or dependencies
 90 | #   having no cross-platform support, pipenv may install dependencies that don't work, or not
 91 | #   install all needed dependencies.
 92 | #Pipfile.lock
 93 | 
 94 | # PEP 582; used by e.g. github.com/David-OConnor/pyflow
 95 | __pypackages__/
 96 | 
 97 | # Celery stuff
 98 | celerybeat-schedule
 99 | celerybeat.pid
100 | 
101 | # SageMath parsed files
102 | *.sage.py
103 | 
104 | # Environments
105 | .env
106 | .venv
107 | env/
108 | venv/
109 | ENV/
110 | env.bak/
111 | venv.bak/
112 | 
113 | # Spyder project settings
114 | .spyderproject
115 | .spyproject
116 | 
117 | # Rope project settings
118 | .ropeproject
119 | 
120 | # mkdocs documentation
121 | /site
122 | 
123 | # mypy
124 | .mypy_cache/
125 | .dmypy.json
126 | dmypy.json
127 | 
128 | # Pyre type checker
129 | .pyre/
130 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | # AuxSegNet
 2 | 
 3 | The pytorch code for our ICCV 2021 paper [Leveraging Auxiliary Tasks with Affinity Learning for Weakly Supervised Semantic Segmentation](https://arxiv.org/abs/2107.11787).
 4 | 
 5 | <p align="left">
 6 |   <img src="mis/framework2.jpg" width="720" title="" >
 7 | </p>
 8 | 
 9 | <p align="left">
10 |   <img src="mis/example.jpg" width="720" title="" >
11 | </p>
12 | 
13 | ## Prerequisite
14 | - Ubuntu 18.04, with Python 3.6 and the following python dependencies.
15 | ```
16 | pip install -r prerequisite.txt
17 | ```
18 | - Download [the PASCAL VOC 2012 development kit](http://host.robots.ox.ac.uk/pascal/VOC/voc2012).
19 | 
20 | ## Usage
21 | 
22 | 
23 | #### 1. Prepare initial pseudo labels
24 | - Off-the-shelf saliency maps used as the initial saliency pseudo labels. [[DSS]](https://drive.google.com/open?id=1Ls2HBtg3jUiuk3WUuMtdUOVUFCgvE8IX)
25 | - Extract the class activation maps (CAM) from a pre-trained single-task classification network. [[ResNet38]](https://drive.google.com/file/d/1xESB7017zlZHqxEWuh1Rb89UhjTGIKOA/view?usp=sharing)
26 | - Generate the initial pseudo segmentation labels using the above saliency and CAM maps via [[heuristic fusion]](https://github.com/xulianuwa/AuxSegNet/blob/597506a4f44cca81d11c986217e5318361e8f65e/tool/imutils.py#L36).
27 | #### 2. Train the AuxSegNet
28 | 
29 | ```
30 | python train_AuxAff.py --img_path 'Path to the training images'\
31 |                        --seg_pgt_path 'Path to the pseudo segmentation labels' \
32 |                        --sal_pgt_path 'Path to the pseudo saliency labels' \
33 |                        --init_weights 'Path to the initialization weights' \
34 |                        --save_path 'Path to save the trained AuxSegNet model' 
35 | ```
36 | 
37 | 
38 | #### 3. Pseudo label updating
39 | ```
40 | python gen_pgt.py --weights 'path to the trained AuxSegNet weights'\   
41 |                   --img_path 'Path to the training images'\
42 |                   --SALpath 'Path to the pre-trained saliency maps' \
43 |                   --seg_pgt_path 'Path to save updated pseudo segmentation labels' \
44 |                   --sal_pgt_path 'Path to save updated pseudo saliency labels' 
45 | ```
46 | #### 4. Iterate Step 2 and 3
47 | 
48 | #### (Optional) Integrated iterative model learning and label updating (Step 2-4) 
49 | ```
50 | bash iter_learn.sh
51 | ```
52 | #### 5. Inference
53 | ```
54 | python infer_AuxAff.py --img_path 'Path to the training images'\
55 |                        --weights 'Path to the trained AuxSegNet weights'\
56 |                        --save_path 'Path to save the segmentation results'
57 | 
58 | ```
59 | ## Performance comparison with SOTA 
60 | 
61 | Segmentation results on the PASCAL VOC 2012 dataset
62 | 
63 | <p align="left">
64 |   <img src="mis/sota_voc.png" width="300" title="" >
65 | </p>
66 | 
67 | [comment]: <> (| Method         | Val &#40;mIoU&#41;    | Test &#40;mIoU&#41;    | )
68 | 
69 | [comment]: <> (| ------------- |:-------------:|:-----:|)
70 | 
71 | [comment]: <> (|ICD &#40;CVPR20&#41;|67.8|68.0|)
72 | 
73 | [comment]: <> (|Zhang et al. &#40;ECCV20&#41;|66.6|66.7|)
74 | 
75 | [comment]: <> (|Sun et al. &#40;ECCV20&#41;|66.2|66.9|)
76 | 
77 | [comment]: <> (| AuxSegNet &#40;ours&#41;     | **69.0** | **68.6** | )
78 | 
79 | Segmentation results on the MS COCO dataset
80 | <p align="left">
81 |   <img src="mis/sota_coco.png" width="300" title="" >
82 | </p>
83 | 
84 | 
85 | ## Citation
86 | Please consider citing our paper if the code is helpful in your research and development.
87 | ```
88 | @inproceedings{xu2021leveraging,
89 |   title={Leveraging Auxiliary Tasks with Affinity Learning for Weakly Supervised Semantic Segmentation},
90 |   author={Xu, Lian and Ouyang, Wanli and Bennamoun, Mohammed and Boussaid, Farid and Sohel, Ferdous and Xu, Dan},
91 |   booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
92 |   year={2021}
93 | }
94 | ```
95 | 


--------------------------------------------------------------------------------
/gen_pgt.py:
--------------------------------------------------------------------------------
 1 | import torch
 2 | import torch.nn.functional as F
 3 | from torch.backends import cudnn
 4 | 
 5 | cudnn.enabled = True
 6 | import importlib
 7 | import argparse
 8 | import os.path
 9 | from tqdm import tqdm
10 | 
11 | from tool.imutils import fusion_cam, put_palette, crf_for_sal, save_as_png
12 | 
13 | import cv2
14 | import numpy as np
15 | from pathlib import Path
16 | 
17 | 
18 | if __name__ == '__main__':
19 | 
20 |     parser = argparse.ArgumentParser()
21 |     parser.add_argument("--gpu_id", type=str, default='0', help='GPU_id')
22 |     parser.add_argument("--weights", default="", type=str)
23 |     parser.add_argument("--network", default="AuxSegNet", type=str)
24 |     parser.add_argument("--list_path", default="./voc12/train_aug_id.txt", type=str)
25 |     parser.add_argument("--num_classes", default=21, type=int)
26 | 
27 |     parser.add_argument("--sal_pgt_path", default=None, type=str)
28 |     parser.add_argument("--seg_pgt_path", default=None, type=str)
29 |     parser.add_argument("--SALpath", default=None, type=str)
30 |     parser.add_argument("--img_path", default="", type=str)
31 | 
32 | 
33 |     args = parser.parse_args()
34 | 
35 |     gpu_id = args.gpu_id
36 |     os.environ["CUDA_VISIBLE_DEVICES"] = gpu_id
37 | 
38 |     cls_labels_dict = np.load('voc12/cls_labels.npy', allow_pickle=True).item()
39 | 
40 |     Path(args.sal_pgt_path).mkdir(parents=True, exist_ok=True)
41 |     Path(args.seg_pgt_path).mkdir(parents=True, exist_ok=True)
42 |     
43 |     model = getattr(importlib.import_module('network.' + args.network), 'SegNet')(num_classes=args.num_classes)
44 | 
45 |     model.load_state_dict(torch.load(args.weights))
46 |     model.eval()
47 |     model.cuda()
48 | 
49 |     im_path = args.img_path
50 |     img_list = open(args.list_path).readlines()
51 | 
52 |     with torch.no_grad():
53 |         for idx in tqdm(range(len(img_list))):
54 |             i = img_list[idx]
55 |             img_temp = cv2.imread(os.path.join(im_path, i.strip() + '.jpg'))
56 |             img_temp = cv2.cvtColor(img_temp, cv2.COLOR_BGR2RGB).astype(np.float)
57 |             img_original = img_temp.astype(np.uint8)
58 |             img_temp[:, :, 0] = (img_temp[:, :, 0] / 255. - 0.485) / 0.229
59 |             img_temp[:, :, 1] = (img_temp[:, :, 1] / 255. - 0.456) / 0.224
60 |             img_temp[:, :, 2] = (img_temp[:, :, 2] / 255. - 0.406) / 0.225
61 | 
62 |             sal_temp_np = cv2.imread(os.path.join(args.SALpath, i.strip() + '.png'), 0) / 255.
63 |             sal_temp = torch.from_numpy(sal_temp_np[np.newaxis, :]).cuda()
64 | 
65 |             img_label = cls_labels_dict[i.strip()]
66 |             label_i = cls_labels_dict[i.strip()]
67 |             label_i = torch.from_numpy(label_i)
68 |             label_i = label_i.unsqueeze(0).cuda()
69 |             input = torch.from_numpy(img_temp[np.newaxis, :].transpose(0, 3, 1, 2)).float().cuda()
70 |             N, C, H, W = input.size()
71 | 
72 |             _, cam, b_mask, ref_bmask, init_seg, seg, ct_aff = model(x=input.cuda(), require_cam=True)
73 | 
74 |             cam = F.relu(cam.clone())
75 |             cam = cam / (torch.max(torch.max(cam, dim=2, keepdim=True)[0], dim=3, keepdim=True)[0] + 1e-5)
76 |             refined_cam = cam.view(cam.size(0), cam.size(1), -1).permute(0, 2, 1).contiguous()
77 |             refined_cam = torch.matmul(ct_aff, refined_cam)
78 |             refined_cam = refined_cam.permute(0, 2, 1).contiguous().view(*cam.size())
79 |             refined_cam = F.interpolate(refined_cam, size=(H,W), mode='bilinear', align_corners=False)
80 | 
81 |             ref_bmask = F.sigmoid(ref_bmask)
82 |             ref_sal = F.interpolate(ref_bmask, size=(H, W), mode='bilinear', align_corners=False)
83 |             ref_sal_np = ref_sal.detach().cpu().numpy().squeeze()
84 |             fused_sal = (ref_sal_np + sal_temp_np) / 2
85 |             fused_sal = crf_for_sal(img_original, fused_sal)
86 |             save_as_png(fused_sal.squeeze(), i.strip(), args.sal_pgt_path)
87 |             updated_sal = torch.from_numpy(fused_sal[np.newaxis, :]).cuda()
88 | 
89 |             pgt = fusion_cam(refined_cam, updated_sal, label_i, args, [i.strip()])
90 |             pgt_np = pgt.detach().cpu().numpy().squeeze()
91 |             save_name = os.path.join(args.seg_pgt_path, i.strip() + '.png')
92 |             put_palette(pgt_np, save_name)


--------------------------------------------------------------------------------
/infer_AuxAff.py:
--------------------------------------------------------------------------------
 1 | import numpy as np
 2 | import torch
 3 | from torch.backends import cudnn
 4 | cudnn.enabled = True
 5 | import importlib
 6 | from tool import imutils
 7 | import argparse
 8 | import cv2
 9 | import os.path
10 | import torch.nn.functional as F
11 | from tqdm import tqdm
12 | from pathlib import Path
13 | 
14 | def _crf_with_alpha(pred_prob, ori_img):
15 |     crf_score = imutils.crf_inference_inf(ori_img, pred_prob, labels=21)
16 |     return crf_score
17 | 
18 | def str2bool(v):
19 |     if v.lower() in ('yes','true','t','y','1','True'):
20 |         return True
21 |     elif v.lower() in ('no','false','f','n','0','False'):
22 |         return False
23 | 
24 | 
25 | if __name__ == '__main__':
26 | 
27 |     parser = argparse.ArgumentParser()
28 |     parser.add_argument('--gpu_id', type=str, default='0', help='GPU_id')
29 |     parser.add_argument("--weights", default="", type=str)
30 |     parser.add_argument("--network", default="AuxSegNet", type=str)
31 |     parser.add_argument("--save_path", default=None, type=str)
32 |     parser.add_argument("--list_path", default="./voc12/val_id.txt", type=str)
33 |     parser.add_argument("--img_path", default="", type=str)
34 |     parser.add_argument("--num_classes", default=21, type=int)
35 |     parser.add_argument("--use_crf", default=False, type=str2bool)
36 |     args = parser.parse_args()
37 | 
38 |     gpu_id = args.gpu_id
39 |     os.environ["CUDA_VISIBLE_DEVICES"] = gpu_id
40 | 
41 |     Path(args.save_path).mkdir(parents=True, exist_ok=True)
42 | 
43 |     model = getattr(importlib.import_module('network.' + args.network), 'SegNet')(num_classes=args.num_classes)
44 | 
45 |     model.load_state_dict(torch.load(args.weights))
46 |     model.eval()
47 |     model.cuda()
48 |     im_path = args.img_path
49 |     img_list = open(args.list_path).readlines()
50 | 
51 |     with torch.no_grad():
52 |         for idx in tqdm(range(len(img_list))):
53 |             i = img_list[idx]
54 | 
55 |             img_temp = cv2.imread(os.path.join(im_path, i.strip() + '.jpg'))
56 |             img_temp = cv2.cvtColor(img_temp, cv2.COLOR_BGR2RGB).astype(np.float)
57 |             img_original = img_temp.astype(np.uint8)
58 | 
59 |             img_temp[:, :, 0] = (img_temp[:, :, 0] / 255. - 0.485) / 0.229
60 |             img_temp[:, :, 1] = (img_temp[:, :, 1] / 255. - 0.456) / 0.224
61 |             img_temp[:, :, 2] = (img_temp[:, :, 2] / 255. - 0.406) / 0.225
62 | 
63 |             input = torch.from_numpy(img_temp[np.newaxis, :].transpose(0, 3, 1, 2)).float().cuda()
64 | 
65 |             N, C, H, W = input.size()
66 |             init_prob, prob = model(x=input, require_cls=False, require_sal=False)
67 | 
68 |             prob = F.interpolate(prob, (H, W), mode='bilinear', align_corners=False)
69 |             prob = F.softmax(prob, dim=1)
70 |             output = prob.cpu().data[0].numpy()
71 | 
72 |             if args.use_crf:
73 |                 crf_output = _crf_with_alpha(output, img_original)
74 |                 pred = np.argmax(crf_output, 0)
75 |             else:
76 |                 pred = np.argmax(output, axis=0)
77 | 
78 |             save_path = os.path.join(args.save_path, i.strip() + '.png')
79 |             cv2.imwrite(save_path, pred.astype(np.uint8))


--------------------------------------------------------------------------------
/iter_learn.sh:
--------------------------------------------------------------------------------
1 | python main.py --init_weights   \
2 |                --image_path   \
3 |                --init_salpgt_path  \
4 |                --init_segpgt_path  \
5 |                --model_path  \
6 |                --sal_pgt_path  \
7 |                --seg_pgt_path  \
8 |                --seg_output_path


--------------------------------------------------------------------------------
/main.py:
--------------------------------------------------------------------------------
 1 | from torch.backends import cudnn
 2 | cudnn.enabled = True
 3 | import argparse
 4 | import os
 5 | import step.train_AuxAff, step.gen_pgt, step.infer_AuxAff
 6 | 
 7 | 
 8 | def str2bool(v):
 9 |     if v.lower() in ('yes','true','t','y','1','True'):
10 |         return True
11 |     elif v.lower() in ('no','false','f','n','0','False'):
12 |         return False
13 | 
14 | if __name__ == '__main__':
15 | 
16 |     parser = argparse.ArgumentParser()
17 |     parser.add_argument('--gpu_id', type=str, default='0', help='GPU_id')
18 | 
19 |     # Dataset parameters
20 |     parser.add_argument("--train_list", default="voc12/train_aug_id.txt", type=str)
21 |     parser.add_argument("--test_list", default="voc12/val_id.txt", type=str)
22 |     parser.add_argument("--image_path", default="", type=str)
23 |     parser.add_argument("--init_salpgt_path", default="", type=str, help='off-the-shelf saliency maps')
24 |     parser.add_argument("--init_segpgt_path", default="", type=str, help='initial semantic segmentation pseudo label maps')
25 |     parser.add_argument("--num_classes", default=21, type=int)
26 | 
27 |     # Parameters for training AuxSegNet
28 |     parser.add_argument("--network", default='AuxSegNet', type=str)
29 |     parser.add_argument("--batch_size", default=4, type=int)
30 |     parser.add_argument("--num_epochs", default=15, type=int)
31 |     parser.add_argument("--lr", default=0.0007, type=float)
32 |     parser.add_argument("--num_workers", default=16, type=int)
33 |     parser.add_argument("--wt_dec", default=1e-5, type=float)
34 |     parser.add_argument("--init_weights", default='', type=str)
35 | 
36 |     parser.add_argument("--session_name", default="AuxSegNet_", type=str)
37 |     parser.add_argument("--crop_size", default=321, type=int)
38 |     parser.add_argument('--print_intervals', type=int, default=50)
39 | 
40 |     parser.add_argument('--sal_loss_weight', type=float, default=1.0)
41 |     parser.add_argument('--cls_loss_weight', type=float, default=1.0)
42 |     parser.add_argument('--seg_loss_weight', type=float, default=1.0)
43 | 
44 |     parser.add_argument("--num_steps", default=4, type=int, help='number of steps for the iterative affinity learning')
45 | 
46 |     # Parameters for testing AuxSegNet
47 |     parser.add_argument("--use_crf", default=False, type=str2bool)
48 | 
49 |     # Output paths
50 |     parser.add_argument("--model_path", default=None, type=str, help='path to save trained AuxSegNet models')
51 |     parser.add_argument("--sal_pgt_path", default=None, type=str, help='path to save updated saliency pgt')
52 |     parser.add_argument("--seg_pgt_path", default=None, type=str, help='path to save updated semantic segmentation pgt')
53 |     parser.add_argument("--seg_output_path", default=None, type=str, help='path to save semantic segmentation output results')
54 | 
55 |     args = parser.parse_args()
56 | 
57 |     for s in range(args.num_steps):
58 |         if s > 0:
59 |             args.init_weights = os.path.join(args.model_path, args.session_name + 's' + str(s-1) + '.pth')
60 | 
61 |         print(f'Training the AuxSegNet at the {s}-th step')
62 |         step.train_AuxAff.run(args, step_index=s)
63 |         print(f'Pseudo label updating at the {s}-th step')
64 |         step.gen_pgt.run(args, step_index=s)
65 | 
66 |         if s == args.num_steps - 1:
67 |             print('Testing the AuxSegNet')
68 |             step.infer_AuxAff.run(args, step_index=s)
69 | 


--------------------------------------------------------------------------------
/mis/example.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/xulianuwa/AuxSegNet/9f40a5d349e2405b42a541a3534e51eb58b04077/mis/example.jpg


--------------------------------------------------------------------------------
/mis/framework.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/xulianuwa/AuxSegNet/9f40a5d349e2405b42a541a3534e51eb58b04077/mis/framework.jpg


--------------------------------------------------------------------------------
/mis/framework2.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/xulianuwa/AuxSegNet/9f40a5d349e2405b42a541a3534e51eb58b04077/mis/framework2.jpg


--------------------------------------------------------------------------------
/mis/sota_coco.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/xulianuwa/AuxSegNet/9f40a5d349e2405b42a541a3534e51eb58b04077/mis/sota_coco.png


--------------------------------------------------------------------------------
/mis/sota_voc.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/xulianuwa/AuxSegNet/9f40a5d349e2405b42a541a3534e51eb58b04077/mis/sota_voc.png


--------------------------------------------------------------------------------
/network/AuxSegNet.py:
--------------------------------------------------------------------------------
  1 | import torch
  2 | import torch.nn as nn
  3 | import torch.nn.functional as F
  4 | import network.resnet38d
  5 | from network.non_local import NLBlock
  6 | 
  7 | 
  8 | class Net(network.resnet38d.Net):
  9 |     def __init__(self, num_classes=21):
 10 |         super().__init__()
 11 | 
 12 |         self.dropout7 = torch.nn.Dropout2d(0.5)
 13 | 
 14 |         self.fc8 = nn.Conv2d(4096, num_classes - 1, (1, 1), bias=False)
 15 |         torch.nn.init.xavier_uniform_(self.fc8.weight)
 16 | 
 17 |         self.seg_conv1 = nn.Conv2d(4096, 128, (3, 3), padding=6, dilation=(6, 6), bias=True)
 18 |         torch.nn.init.xavier_uniform_(self.seg_conv1.weight)
 19 | 
 20 |         self.seg_conv2 = nn.Conv2d(128, 128, (3, 3), padding=12, dilation=(12, 12), bias=True)
 21 |         torch.nn.init.xavier_uniform_(self.seg_conv2.weight)
 22 | 
 23 |         self.seg_conv3 = nn.Conv2d(128, 128, (3, 3), padding=18, dilation=(18, 18), bias=True)
 24 |         torch.nn.init.xavier_uniform_(self.seg_conv3.weight)
 25 | 
 26 |         self.seg_conv4 = nn.Conv2d(128, num_classes, (1, 1), bias=False)
 27 |         torch.nn.init.xavier_uniform_(self.seg_conv4.weight)
 28 | 
 29 |         self.sal_conv1 = nn.Conv2d(4096, 128, (3, 3), padding=6, dilation=(6, 6), bias=True)
 30 |         torch.nn.init.xavier_uniform_(self.sal_conv1.weight)
 31 | 
 32 |         self.sal_conv2 = nn.Conv2d(128, 128, (3, 3), padding=12, dilation=(12, 12), bias=True)
 33 |         torch.nn.init.xavier_uniform_(self.sal_conv2.weight)
 34 | 
 35 |         self.sal_conv3 = nn.Conv2d(128, 128, (3, 3), padding=18, dilation=(18, 18), bias=True)
 36 |         torch.nn.init.xavier_uniform_(self.sal_conv3.weight)
 37 | 
 38 |         self.sal_conv4 = nn.Conv2d(128, 1, (1, 1), bias=False)
 39 |         torch.nn.init.xavier_uniform_(self.sal_conv4.weight)
 40 | 
 41 |         self.seg_att = NLBlock(in_channels=128, inter_channels=128)
 42 |         self.sal_att = NLBlock(in_channels=128, inter_channels=128)
 43 | 
 44 |         self.seg_g = nn.Conv2d(128, 128, (1, 1))
 45 |         self.seg_W = nn.Conv2d(128, 128, (1, 1))
 46 |         self.sal_g = nn.Conv2d(128, 128, (1, 1))
 47 |         self.sal_W = nn.Conv2d(128, 128, (1, 1))
 48 |         self.cross_att = nn.Sequential(nn.Conv2d(2, 2, (3, 3), padding=1, bias=False),
 49 |                                        nn.ReLU(),
 50 |                                        nn.Conv2d(2, 2, (1, 1)),
 51 |                                        nn.Softmax(dim=1))
 52 | 
 53 |     def forward(self, x, return_aff = False):
 54 |         batch_size = x.size(0)
 55 | 
 56 |         x_seg = super().forward(x)
 57 |         x_sal = x_seg.clone()
 58 |         x_cls = x_seg.clone()
 59 |         x_cam = x_seg.clone()
 60 | 
 61 |         x_cls = self.dropout7(x_cls)
 62 |         x_cls = F.avg_pool2d(
 63 |             x_cls, kernel_size=(x_cls.size(2), x_cls.size(3)), padding=0)
 64 |         x_cls = self.fc8(x_cls)
 65 |         x_cls = x_cls.view(x_cls.size(0), -1)
 66 |         cam = F.conv2d(x_cam, self.fc8.weight)
 67 | 
 68 |         x_seg = F.relu(self.seg_conv1(x_seg))
 69 |         x_seg = F.relu(self.seg_conv2(x_seg))
 70 |         seg_feats = x_seg
 71 | 
 72 |         x_sal = F.relu(self.sal_conv1(x_sal))
 73 |         x_sal = F.relu(self.sal_conv2(x_sal))
 74 |         sal_feats = x_sal
 75 | 
 76 |         seg_aff = self.seg_att(seg_feats)
 77 |         sal_aff = self.sal_att(sal_feats)
 78 | 
 79 |         g_seg_feats = self.seg_g(seg_feats)
 80 |         g_seg_feats = g_seg_feats.view(batch_size, seg_feats.size(1), -1)
 81 |         g_seg_feats = g_seg_feats.permute(0, 2, 1).contiguous()
 82 |         refined_seg_feats = torch.matmul(seg_aff, g_seg_feats)
 83 |         refined_seg_feats = refined_seg_feats.permute(0, 2, 1).contiguous()
 84 |         refined_seg_feats = refined_seg_feats.view(*seg_feats.size())
 85 |         W_seg_feats = self.seg_W(refined_seg_feats)
 86 |         seg_feats = seg_feats + W_seg_feats
 87 | 
 88 |         seg_feats = F.relu(self.seg_conv3(seg_feats))
 89 |         init_seg_mask = self.seg_conv4(seg_feats)
 90 | 
 91 |         g_sal_feats = self.sal_g(sal_feats)
 92 |         g_sal_feats = g_sal_feats.view(batch_size, sal_feats.size(1), -1).permute(0, 2, 1).contiguous()
 93 |         refined_sal_feats = torch.matmul(sal_aff, g_sal_feats)
 94 |         refined_sal_feats = refined_sal_feats.permute(0, 2, 1).contiguous().view(*sal_feats.size())
 95 |         sal_feats = sal_feats + self.sal_W(refined_sal_feats)
 96 | 
 97 |         sal_feats = F.relu(self.sal_conv3(sal_feats))
 98 |         init_sal_mask = self.sal_conv4(sal_feats)
 99 | 
100 |         if len(seg_aff.size()) != 4:
101 |             seg_aff = seg_aff.unsqueeze(1)
102 |             sal_aff = sal_aff.unsqueeze(1)
103 | 
104 |         cross_aff = torch.cat((seg_aff, sal_aff), dim=1)
105 |         cross_w = self.cross_att(cross_aff)
106 |         cross_aff = cross_aff[:, 0] * cross_w[:, 0] + cross_aff[:, 1] * cross_w[:, 1]
107 | 
108 |         sal_mask = init_sal_mask.view(batch_size, init_sal_mask.size(1), -1)
109 |         sal_mask = sal_mask.permute(0, 2, 1).contiguous()
110 |         refined_sal_mask = torch.matmul(cross_aff, sal_mask)
111 |         refined_sal_mask = refined_sal_mask.permute(0, 2, 1).contiguous().view(*init_sal_mask.size())
112 | 
113 |         seg_mask = init_seg_mask.view(batch_size, init_seg_mask.size(1), -1)
114 |         seg_mask = seg_mask.permute(0, 2, 1).contiguous()
115 |         refined_seg_mask = torch.matmul(cross_aff, seg_mask)
116 |         refined_seg_mask = refined_seg_mask.permute(0, 2, 1).contiguous().view(*init_seg_mask.size())
117 | 
118 |         if return_aff:
119 |             return x_cls, cam, init_sal_mask, refined_sal_mask, init_seg_mask, refined_seg_mask, cross_aff
120 |         else:
121 |             return x_cls, cam, init_sal_mask, refined_sal_mask, init_seg_mask, refined_seg_mask
122 | 
123 | 
124 | class SegNet(Net):
125 |     def __init__(self, num_classes=21):
126 |         super().__init__(num_classes=num_classes)
127 | 
128 |     def forward(self, x, require_seg=True, require_cls=True, require_sal=True, require_cam=False):
129 | 
130 |         if require_seg == True and require_cls == True and require_sal == True and require_cam == False:
131 |             x_cls, cam, init_sal, refined_sal, init_seg, refined_seg = super().forward(x)
132 |             return x_cls, init_sal, refined_sal, init_seg, refined_seg
133 | 
134 |         if require_seg == True and require_cls == True and require_sal == True and require_cam == True:
135 |             x_cls, cam, init_sal, refined_sal, init_seg, refined_seg, aff = super().forward(x, return_aff=True)
136 |             return x_cls, cam, init_sal, refined_sal, init_seg, refined_seg, aff
137 | 
138 |         if require_cls == False and require_seg == True and require_sal == False:
139 |             x_cls, cam, init_sal, refined_sal, init_seg, refined_seg = super().forward(x)
140 |             return init_seg, refined_seg
141 | 
142 |     def get_10x_lr_params(self):
143 |         for name, param in self.named_parameters():
144 |             if 'seg' in name or 'sal' in name or 'fc8' in name:
145 |                 yield param
146 | 
147 |     def get_1x_lr_params(self):
148 |         for name, param in self.named_parameters():
149 |             if 'seg' not in name and 'sal' not in name and 'fc8' not in name:
150 |                 yield param


--------------------------------------------------------------------------------
/network/non_local.py:
--------------------------------------------------------------------------------
 1 | from torch import nn
 2 | from torch.nn import functional as F
 3 | import torch
 4 | 
 5 | class NLBlock(nn.Module):
 6 |     def __init__(self, in_channels, inter_channels=None, bn_layer=True):
 7 |         super(NLBlock, self).__init__()
 8 |         self.in_channels = in_channels
 9 |         self.inter_channels = inter_channels
10 |         self.g = nn.Conv2d(in_channels=self.in_channels, out_channels=self.inter_channels, kernel_size=(1,1))
11 | 
12 |         if bn_layer:
13 |             self.W_z = nn.Sequential(
14 |                 nn.Conv2d(in_channels=self.inter_channels, out_channels=self.in_channels, kernel_size=(1,1)),
15 |                 nn.BatchNorm2d(self.in_channels)
16 |             )
17 |             nn.init.constant_(self.W_z[1].weight, 0)
18 |             nn.init.constant_(self.W_z[1].bias, 0)
19 |         else:
20 |             self.W_z = nn.Conv2d(in_channels=self.inter_channels, out_channels=self.in_channels, kernel_size=(1,1))
21 |             nn.init.constant_(self.W_z.weight, 0)
22 |             nn.init.constant_(self.W_z.bias, 0)
23 | 
24 |         self.theta = nn.Conv2d(in_channels=self.in_channels, out_channels=self.inter_channels, kernel_size=(1,1))
25 |         self.phi = nn.Conv2d(in_channels=self.in_channels, out_channels=self.inter_channels, kernel_size=(1,1))
26 | 
27 |     def forward(self, x):
28 |         batch_size = x.size(0)
29 |         theta_x = self.theta(x).view(batch_size, self.inter_channels, -1)
30 |         phi_x = self.phi(x).view(batch_size, self.inter_channels, -1)
31 |         theta_x = theta_x.permute(0, 2, 1)
32 |         f = torch.matmul(theta_x, phi_x)
33 |         f_div_C = F.softmax(f, dim=-1)
34 |         return f_div_C
35 | 


--------------------------------------------------------------------------------
/network/resnet38d.py:
--------------------------------------------------------------------------------
  1 | import torch
  2 | from torch import nn
  3 | import numpy as np
  4 | 
  5 | import torch.nn.functional as F
  6 | 
  7 | class ResBlock(nn.Module):
  8 |     def __init__(self, in_channels, mid_channels, out_channels, stride=1, first_dilation=None, dilation=1):
  9 |         super(ResBlock, self).__init__()
 10 | 
 11 |         self.same_shape = (in_channels == out_channels and stride == 1)
 12 | 
 13 |         if first_dilation == None: first_dilation = dilation
 14 | 
 15 |         self.bn_branch2a = nn.BatchNorm2d(in_channels)
 16 | 
 17 |         self.conv_branch2a = nn.Conv2d(in_channels, mid_channels, 3, stride,
 18 |                                        padding=first_dilation, dilation=first_dilation, bias=False)
 19 | 
 20 |         self.bn_branch2b1 = nn.BatchNorm2d(mid_channels)
 21 | 
 22 |         self.conv_branch2b1 = nn.Conv2d(mid_channels, out_channels, 3, padding=dilation, dilation=dilation, bias=False)
 23 | 
 24 |         if not self.same_shape:
 25 |             self.conv_branch1 = nn.Conv2d(in_channels, out_channels, 1, stride, bias=False)
 26 | 
 27 |     def forward(self, x, get_x_bn_relu=False):
 28 | 
 29 |         branch2 = self.bn_branch2a(x)
 30 |         branch2 = F.relu(branch2)
 31 | 
 32 |         x_bn_relu = branch2
 33 | 
 34 |         if not self.same_shape:
 35 |             branch1 = self.conv_branch1(branch2)
 36 |         else:
 37 |             branch1 = x
 38 | 
 39 |         branch2 = self.conv_branch2a(branch2)
 40 |         branch2 = self.bn_branch2b1(branch2)
 41 |         branch2 = F.relu(branch2)
 42 |         branch2 = self.conv_branch2b1(branch2)
 43 | 
 44 |         x = branch1 + branch2
 45 | 
 46 |         if get_x_bn_relu:
 47 |             return x, x_bn_relu
 48 | 
 49 |         return x
 50 | 
 51 |     def __call__(self, x, get_x_bn_relu=False):
 52 |         return self.forward(x, get_x_bn_relu=get_x_bn_relu)
 53 | 
 54 | class ResBlock_bot(nn.Module):
 55 |     def __init__(self, in_channels, out_channels, stride=1, dilation=1, dropout=0.):
 56 |         super(ResBlock_bot, self).__init__()
 57 | 
 58 |         self.same_shape = (in_channels == out_channels and stride == 1)
 59 | 
 60 |         self.bn_branch2a = nn.BatchNorm2d(in_channels)
 61 |         self.conv_branch2a = nn.Conv2d(in_channels, out_channels//4, 1, stride, bias=False)
 62 | 
 63 |         self.bn_branch2b1 = nn.BatchNorm2d(out_channels//4)
 64 |         self.dropout_2b1 = torch.nn.Dropout2d(dropout)
 65 |         self.conv_branch2b1 = nn.Conv2d(out_channels//4, out_channels//2, 3, padding=dilation, dilation=dilation, bias=False)
 66 | 
 67 |         self.bn_branch2b2 = nn.BatchNorm2d(out_channels//2)
 68 |         self.dropout_2b2 = torch.nn.Dropout2d(dropout)
 69 |         self.conv_branch2b2 = nn.Conv2d(out_channels//2, out_channels, 1, bias=False)
 70 | 
 71 |         if not self.same_shape:
 72 |             self.conv_branch1 = nn.Conv2d(in_channels, out_channels, 1, stride, bias=False)
 73 | 
 74 |     def forward(self, x, get_x_bn_relu=False):
 75 | 
 76 |         branch2 = self.bn_branch2a(x)
 77 |         branch2 = F.relu(branch2)
 78 |         x_bn_relu = branch2
 79 | 
 80 |         branch1 = self.conv_branch1(branch2)
 81 | 
 82 |         branch2 = self.conv_branch2a(branch2)
 83 | 
 84 |         branch2 = self.bn_branch2b1(branch2)
 85 |         branch2 = F.relu(branch2)
 86 |         branch2 = self.dropout_2b1(branch2)
 87 |         branch2 = self.conv_branch2b1(branch2)
 88 | 
 89 |         branch2 = self.bn_branch2b2(branch2)
 90 |         branch2 = F.relu(branch2)
 91 |         branch2 = self.dropout_2b2(branch2)
 92 |         branch2 = self.conv_branch2b2(branch2)
 93 | 
 94 |         x = branch1 + branch2
 95 | 
 96 |         if get_x_bn_relu:
 97 |             return x, x_bn_relu
 98 | 
 99 |         return x
100 | 
101 |     def __call__(self, x, get_x_bn_relu=False):
102 |         return self.forward(x, get_x_bn_relu=get_x_bn_relu)
103 | 
104 | class Normalize():
105 |     def __init__(self, mean = (0.485, 0.456, 0.406), std = (0.229, 0.224, 0.225)):
106 | 
107 |         self.mean = mean
108 |         self.std = std
109 | 
110 |     def __call__(self, img):
111 |         imgarr = np.asarray(img)
112 |         proc_img = np.empty_like(imgarr, np.float32)
113 | 
114 |         proc_img[..., 0] = (imgarr[..., 0] / 255. - self.mean[0]) / self.std[0]
115 |         proc_img[..., 1] = (imgarr[..., 1] / 255. - self.mean[1]) / self.std[1]
116 |         proc_img[..., 2] = (imgarr[..., 2] / 255. - self.mean[2]) / self.std[2]
117 | 
118 |         return proc_img
119 | 
120 | class Net(nn.Module):
121 |     def __init__(self):
122 |         super(Net, self).__init__()
123 | 
124 |         self.conv1a = nn.Conv2d(3, 64, 3, padding=1, bias=False)
125 | 
126 |         self.b2 = ResBlock(64, 128, 128, stride=2)
127 |         self.b2_1 = ResBlock(128, 128, 128)
128 |         self.b2_2 = ResBlock(128, 128, 128)
129 | 
130 |         self.b3 = ResBlock(128, 256, 256, stride=2)
131 |         self.b3_1 = ResBlock(256, 256, 256)
132 |         self.b3_2 = ResBlock(256, 256, 256)
133 | 
134 |         self.b4 = ResBlock(256, 512, 512, stride=2)
135 |         self.b4_1 = ResBlock(512, 512, 512)
136 |         self.b4_2 = ResBlock(512, 512, 512)
137 |         self.b4_3 = ResBlock(512, 512, 512)
138 |         self.b4_4 = ResBlock(512, 512, 512)
139 |         self.b4_5 = ResBlock(512, 512, 512)
140 | 
141 |         self.b5 = ResBlock(512, 512, 1024, stride=1, first_dilation=1, dilation=2)
142 |         self.b5_1 = ResBlock(1024, 512, 1024, dilation=2)
143 |         self.b5_2 = ResBlock(1024, 512, 1024, dilation=2)
144 | 
145 |         self.b6 = ResBlock_bot(1024, 2048, stride=1, dilation=4, dropout=0.3)
146 | 
147 |         self.b7 = ResBlock_bot(2048, 4096, dilation=4, dropout=0.5)
148 | 
149 |         self.bn7 = nn.BatchNorm2d(4096)
150 | 
151 |         self.not_training = [self.conv1a]
152 | 
153 |         self.normalize = Normalize()
154 | 
155 |         return
156 | 
157 |     def forward(self, x):
158 |         return self.forward_as_dict(x)['conv6']
159 | 
160 |     def forward_as_dict(self, x):
161 | 
162 |         x = self.conv1a(x)
163 | 
164 |         x = self.b2(x)
165 |         x = self.b2_1(x)
166 |         x = self.b2_2(x)
167 | 
168 |         x = self.b3(x)
169 |         x = self.b3_1(x)
170 |         x = self.b3_2(x)
171 | 
172 |         x = self.b4(x)
173 |         x = self.b4_1(x)
174 |         x = self.b4_2(x)
175 |         x = self.b4_3(x)
176 |         x = self.b4_4(x)
177 |         x = self.b4_5(x)
178 | 
179 |         x, conv4 = self.b5(x, get_x_bn_relu=True)
180 |         x = self.b5_1(x)
181 |         x = self.b5_2(x)
182 | 
183 |         x, conv5 = self.b6(x, get_x_bn_relu=True)
184 | 
185 |         x = self.b7(x)
186 |         conv6 = F.relu(self.bn7(x))
187 | 
188 |         return dict({'conv4': conv4, 'conv5': conv5, 'conv6': conv6})
189 | 
190 | 
191 |     def train(self, mode=True):
192 | 
193 |         super().train(mode)
194 | 
195 |         for layer in self.not_training:
196 | 
197 |             if isinstance(layer, torch.nn.Conv2d):
198 |                 layer.weight.requires_grad = False
199 | 
200 |             elif isinstance(layer, torch.nn.Module):
201 |                 for c in layer.children():
202 |                     c.weight.requires_grad = False
203 |                     if c.bias is not None:
204 |                         c.bias.requires_grad = False
205 | 
206 |         for layer in self.modules():
207 | 
208 |             if isinstance(layer, torch.nn.BatchNorm2d):
209 |                 layer.eval()
210 |                 layer.bias.requires_grad = False
211 |                 layer.weight.requires_grad = False
212 | 
213 |         return
214 | 
215 | def convert_mxnet_to_torch(filename):
216 |     import mxnet
217 | 
218 |     save_dict = mxnet.nd.load(filename)
219 | 
220 |     renamed_dict = dict()
221 | 
222 |     bn_param_mx_pt = {'beta': 'bias', 'gamma': 'weight', 'mean': 'running_mean', 'var': 'running_var'}
223 | 
224 |     for k, v in save_dict.items():
225 | 
226 |         v = torch.from_numpy(v.asnumpy())
227 |         toks = k.split('_')
228 | 
229 |         if 'conv1a' in toks[0]:
230 |             renamed_dict['conv1a.weight'] = v
231 | 
232 |         elif 'linear1000' in toks[0]:
233 |             pass
234 | 
235 |         elif 'branch' in toks[1]:
236 | 
237 |             pt_name = []
238 | 
239 |             if toks[0][-1] != 'a':
240 |                 pt_name.append('b' + toks[0][-3] + '_' + toks[0][-1])
241 |             else:
242 |                 pt_name.append('b' + toks[0][-2])
243 | 
244 |             if 'res' in toks[0]:
245 |                 layer_type = 'conv'
246 |                 last_name = 'weight'
247 | 
248 |             else:  # 'bn' in toks[0]:
249 |                 layer_type = 'bn'
250 |                 last_name = bn_param_mx_pt[toks[-1]]
251 | 
252 |             pt_name.append(layer_type + '_' + toks[1])
253 | 
254 |             pt_name.append(last_name)
255 | 
256 |             torch_name = '.'.join(pt_name)
257 |             renamed_dict[torch_name] = v
258 | 
259 |         else:
260 |             last_name = bn_param_mx_pt[toks[-1]]
261 |             renamed_dict['bn7.' + last_name] = v
262 | 
263 |     return renamed_dict
264 | 
265 | 


--------------------------------------------------------------------------------
/prerequisite.txt:
--------------------------------------------------------------------------------
1 | numpy
2 | opencv-python
3 | tqdm
4 | torch
5 | torchvision
6 | Pillow
7 | pydensecrf


--------------------------------------------------------------------------------
/step/gen_pgt.py:
--------------------------------------------------------------------------------
 1 | import torch
 2 | import torch.nn.functional as F
 3 | from torch.backends import cudnn
 4 | cudnn.enabled = True
 5 | import importlib
 6 | import os.path
 7 | from tqdm import tqdm
 8 | from tool.imutils import fusion_cam, put_palette, crf_for_sal, save_as_png
 9 | import cv2
10 | import numpy as np
11 | import sys
12 | sys.path.append("..")
13 | 
14 | def run(args, step_index):
15 | 
16 |     sal_pgt_path = os.path.join(args.sal_pgt_path, str(step_index))
17 |     os.makedirs(sal_pgt_path, exist_ok=True)
18 | 
19 |     seg_pgt_path = os.path.join(args.seg_pgt_path, str(step_index))
20 |     os.makedirs(seg_pgt_path, exist_ok=True)
21 |     model_path = os.path.join(args.model_path, args.session_name + 's' + str(step_index) + '.pth')
22 | 
23 |     gpu_id = args.gpu_id
24 |     os.environ["CUDA_VISIBLE_DEVICES"] = gpu_id
25 |     cls_labels_dict = np.load('voc12/cls_labels.npy', allow_pickle=True).item()
26 | 
27 |     model = getattr(importlib.import_module('network.' + args.network), 'SegNet')(num_classes=args.num_classes)
28 | 
29 |     model.load_state_dict(torch.load(model_path))
30 |     model.eval()
31 |     model.cuda()
32 | 
33 |     im_path = args.image_path
34 |     img_list = open(args.train_list).readlines()
35 | 
36 |     with torch.no_grad():
37 |         for idx in tqdm(range(len(img_list))):
38 |             i = img_list[idx]
39 |             img_temp = cv2.imread(os.path.join(im_path, i.strip() + '.jpg'))
40 |             img_temp = cv2.cvtColor(img_temp, cv2.COLOR_BGR2RGB).astype(np.float)
41 |             img_original = img_temp.astype(np.uint8)
42 |             img_temp[:, :, 0] = (img_temp[:, :, 0] / 255. - 0.485) / 0.229
43 |             img_temp[:, :, 1] = (img_temp[:, :, 1] / 255. - 0.456) / 0.224
44 |             img_temp[:, :, 2] = (img_temp[:, :, 2] / 255. - 0.406) / 0.225
45 | 
46 |             sal_temp_np = cv2.imread(os.path.join(args.init_salpgt_path, i.strip() + '.png'), 0) / 255.
47 |             label_i = cls_labels_dict[i.strip()]
48 |             label_i = torch.from_numpy(label_i)
49 |             label_i = label_i.unsqueeze(0).cuda()
50 |             input = torch.from_numpy(img_temp[np.newaxis, :].transpose(0, 3, 1, 2)).float().cuda()
51 |             N, C, H, W = input.size()
52 | 
53 |             _, cam, b_mask, ref_bmask, init_seg, seg, ct_aff = model(x=input.cuda(), require_cam=True)
54 | 
55 |             cam = F.relu(cam.clone())
56 |             cam = cam / (torch.max(torch.max(cam, dim=2, keepdim=True)[0], dim=3, keepdim=True)[0] + 1e-5)
57 |             refined_cam = cam.view(cam.size(0), cam.size(1), -1).permute(0, 2, 1).contiguous()
58 |             refined_cam = torch.matmul(ct_aff, refined_cam)
59 |             refined_cam = refined_cam.permute(0, 2, 1).contiguous().view(*cam.size())
60 |             refined_cam = F.interpolate(refined_cam, size=(H,W), mode='bilinear', align_corners=False)
61 | 
62 |             ref_bmask = F.sigmoid(ref_bmask)
63 |             ref_sal = F.interpolate(ref_bmask, size=(H, W), mode='bilinear', align_corners=False)
64 |             ref_sal_np = ref_sal.detach().cpu().numpy().squeeze()
65 |             fused_sal = (ref_sal_np + sal_temp_np) / 2
66 |             fused_sal = crf_for_sal(img_original, fused_sal)
67 |             save_as_png(fused_sal.squeeze(), i.strip(), sal_pgt_path)
68 |             updated_sal = torch.from_numpy(fused_sal[np.newaxis, :]).cuda()
69 | 
70 |             pgt = fusion_cam(refined_cam, updated_sal, label_i, args, [i.strip()])
71 |             pgt_np = pgt.detach().cpu().numpy().squeeze()
72 |             save_name = os.path.join(seg_pgt_path, i.strip() + '.png')
73 |             put_palette(pgt_np, save_name)


--------------------------------------------------------------------------------
/step/infer_AuxAff.py:
--------------------------------------------------------------------------------
 1 | import numpy as np
 2 | import torch
 3 | from torch.backends import cudnn
 4 | cudnn.enabled = True
 5 | import importlib
 6 | from tool import imutils
 7 | import cv2
 8 | import os.path
 9 | import torch.nn.functional as F
10 | from tqdm import tqdm
11 | 
12 | def _crf_with_alpha(pred_prob, ori_img):
13 |     crf_score = imutils.crf_inference_inf(ori_img, pred_prob, labels=21)
14 |     return crf_score
15 | 
16 | 
17 | def run(args, step_index):
18 | 
19 |     seg_output_path = os.path.join(args.seg_output_path, str(step_index))
20 |     os.makedirs(seg_output_path)
21 |     gpu_id = args.gpu_id
22 |     os.environ["CUDA_VISIBLE_DEVICES"] = gpu_id
23 |     model = getattr(importlib.import_module('network.' + args.network), 'SegNet')(num_classes=args.num_classes)
24 | 
25 |     model_path = os.path.join(args.model_path, args.session_name + 's' + str(step_index) + '.pth')
26 |     model.load_state_dict(torch.load(model_path))
27 |     model.eval()
28 |     model.cuda()
29 |     im_path = args.image_path
30 |     img_list = open(args.test_list).readlines()
31 | 
32 |     with torch.no_grad():
33 |         for idx in tqdm(range(len(img_list))):
34 |             i = img_list[idx]
35 | 
36 |             img_temp = cv2.imread(os.path.join(im_path, i.strip() + '.jpg'))
37 |             img_temp = cv2.cvtColor(img_temp, cv2.COLOR_BGR2RGB).astype(np.float)
38 |             img_original = img_temp.astype(np.uint8)
39 | 
40 |             img_temp[:, :, 0] = (img_temp[:, :, 0] / 255. - 0.485) / 0.229
41 |             img_temp[:, :, 1] = (img_temp[:, :, 1] / 255. - 0.456) / 0.224
42 |             img_temp[:, :, 2] = (img_temp[:, :, 2] / 255. - 0.406) / 0.225
43 | 
44 |             input = torch.from_numpy(img_temp[np.newaxis, :].transpose(0, 3, 1, 2)).float().cuda()
45 | 
46 |             N, C, H, W = input.size()
47 |             init_prob, prob = model(x=input, require_cls=False, require_sal=False)
48 | 
49 |             prob = F.interpolate(prob, (H, W), mode='bilinear', align_corners=False)
50 |             prob = F.softmax(prob, dim=1)
51 |             output = prob.cpu().data[0].numpy()
52 | 
53 |             if args.use_crf:
54 |                 crf_output = _crf_with_alpha(output, img_original)
55 |                 pred = np.argmax(crf_output, 0)
56 |             else:
57 |                 pred = np.argmax(output, axis=0)
58 | 
59 |             save_path = os.path.join(seg_output_path, i.strip() + '.png')
60 |             cv2.imwrite(save_path, pred.astype(np.uint8))


--------------------------------------------------------------------------------
/step/train_AuxAff.py:
--------------------------------------------------------------------------------
  1 | from torch.backends import cudnn
  2 | cudnn.enabled = True
  3 | from tool import pyutils, torchutils
  4 | import importlib
  5 | import tool.exutils as exutils
  6 | import torch.nn.functional as F
  7 | import torch
  8 | import os
  9 | import numpy as np
 10 | 
 11 | 
 12 | def run(args, step_index):
 13 |     gpu_id = args.gpu_id
 14 |     os.environ["CUDA_VISIBLE_DEVICES"] = gpu_id
 15 | 
 16 |     save_path = args.model_path
 17 |     os.makedirs(save_path, exist_ok=True)
 18 | 
 19 |     pyutils.Logger(os.path.join(args.model_path, args.session_name + '.log'))
 20 | 
 21 |     criterion = torch.nn.CrossEntropyLoss(weight=None, ignore_index=255, reduction='elementwise_mean').cuda()
 22 | 
 23 |     model = getattr(importlib.import_module('network.' + args.network), 'SegNet')(num_classes=args.num_classes)
 24 | 
 25 | 
 26 |     weights_dict = torch.load(args.init_weights)
 27 |     model.load_state_dict(weights_dict, strict=False)
 28 | 
 29 |     img_list = exutils.read_file(args.train_list)
 30 |     train_size = len(img_list)
 31 |     max_step = (train_size // args.batch_size) * args.num_epochs
 32 | 
 33 |     data_list = []
 34 |     for i in range(200):
 35 |         np.random.shuffle(img_list)
 36 |         data_list.extend(img_list)
 37 | 
 38 |     optimizer = torchutils.PolyOptimizer_cls([
 39 |         {'params': model.get_1x_lr_params(), 'lr': args.lr},
 40 |         {'params': model.get_10x_lr_params(), 'lr': 10 * args.lr}
 41 |     ], lr=args.lr, weight_decay=args.wt_dec, max_step=max_step)
 42 | 
 43 |     model = torch.nn.DataParallel(model).cuda()
 44 |     model.train()
 45 | 
 46 |     avg_meter = pyutils.AverageMeter('loss')
 47 | 
 48 |     timer = pyutils.Timer("Session started: ")
 49 | 
 50 |     data_gen = exutils.chunker(data_list, args.batch_size)
 51 | 
 52 |     for iter in range(max_step + 1):
 53 |         chunk = data_gen.__next__()
 54 |         images, ori_images, sal_images, seg_labels, label, img_names = exutils.get_data_from_chunk(chunk, args, step_index)
 55 | 
 56 |         b, _, w, h = ori_images.shape
 57 |         label = label.cuda(non_blocking=True)
 58 |         seg_labels = seg_labels.long().cuda()
 59 |         sal_images = sal_images.cuda()
 60 |         images = images.cuda()
 61 | 
 62 |         x_cls, init_sal, refined_sal, init_seg, refined_seg = model(x=images)
 63 | 
 64 |         ##################################### image classification ###########################################
 65 |         cls_loss = F.multilabel_soft_margin_loss(x_cls, label)
 66 | 
 67 |         ##################################### saliency detection ###########################################
 68 |         init_sal = F.interpolate(init_sal, size=(w, h), mode='bilinear', align_corners=False)
 69 |         sal_labels = sal_images.unsqueeze(1)
 70 |         init_sal_loss = F.binary_cross_entropy_with_logits(init_sal, sal_labels)
 71 | 
 72 |         refined_sal = F.interpolate(refined_sal, size=(w, h), mode='bilinear', align_corners=False)
 73 |         refined_sal_loss = F.binary_cross_entropy_with_logits(refined_sal, sal_labels)
 74 | 
 75 |         ##################################### segmentation ###########################################
 76 |         init_seg = F.interpolate(init_seg, size=(w, h), mode='bilinear', align_corners=False)
 77 |         init_seg_loss = criterion(init_seg, seg_labels)
 78 | 
 79 |         refined_seg = F.interpolate(refined_seg, size=(w, h), mode='bilinear', align_corners=False)
 80 |         refined_seg_loss = criterion(refined_seg, seg_labels)
 81 | 
 82 | 
 83 |         loss = args.cls_loss_weight * cls_loss + args.sal_loss_weight * init_sal_loss + \
 84 |                args.sal_loss_weight * refined_sal_loss + args.seg_loss_weight * init_seg_loss + args.seg_loss_weight * refined_seg_loss
 85 | 
 86 |         avg_meter.add({'loss': loss.item()})
 87 | 
 88 |         optimizer.zero_grad()
 89 |         loss.backward()
 90 |         optimizer.step()
 91 | 
 92 |         if (optimizer.global_step - 1) % args.print_intervals == 0:
 93 |             timer.update_progress(optimizer.global_step / max_step)
 94 | 
 95 |             print('Iter:%5d/%5d' % (optimizer.global_step - 1, max_step),
 96 |                   'Loss:%.4f' % (avg_meter.pop('loss')),
 97 |                   'imps:%.1f' % ((iter + 1) * args.batch_size / timer.get_stage_elapsed()),
 98 |                   'Fin:%s' % (timer.str_est_finish()),
 99 |                   'lr: %.5f' % (optimizer.param_groups[0]['lr']), flush=True)
100 | 
101 |     torch.save(model.module.state_dict(), os.path.join(save_path, args.session_name + 's' + str(step_index) +'.pth'))


--------------------------------------------------------------------------------
/tool/exutils.py:
--------------------------------------------------------------------------------
  1 | import numpy as np
  2 | import torch
  3 | import cv2
  4 | import random
  5 | import os
  6 | import PIL.Image as Image
  7 | 
  8 | 
  9 | def read_file(path_to_file):
 10 |     with open(path_to_file) as f:
 11 |         img_list = []
 12 |         for line in f:
 13 |             img_list.append(line.strip())
 14 |     return img_list
 15 | 
 16 | 
 17 | def chunker(seq, size):
 18 |     return (seq[pos:pos + size] for pos in range(0, len(seq), size))
 19 | 
 20 | 
 21 | def flip(I, flip_p):
 22 |     if flip_p > 0.5:
 23 |         return np.fliplr(I)
 24 |     else:
 25 |         return I
 26 | 
 27 | 
 28 | def scale_im(img_temp, scale):
 29 |     new_dims = (int(img_temp.shape[1] * scale), int(img_temp.shape[0] * scale))
 30 |     return cv2.resize(img_temp, new_dims).astype(float)
 31 | 
 32 | 
 33 | def scale_gt(img_temp, scale):
 34 |     new_dims = (int(img_temp.shape[1] * scale), int(img_temp.shape[0] * scale))
 35 |     return cv2.resize(img_temp, new_dims, interpolation=cv2.INTER_NEAREST).astype(float)
 36 | 
 37 | 
 38 | def load_image_label_list_from_npy(img_name_list):
 39 |     cls_labels_dict = np.load('voc12/cls_labels.npy', allow_pickle=True).item()
 40 | 
 41 |     return [cls_labels_dict[img_name] for img_name in img_name_list]
 42 | 
 43 | 
 44 | def RandomCrop(imgarr, cropsize):
 45 |     h, w, c = imgarr.shape
 46 | 
 47 |     ch = min(cropsize, h)
 48 |     cw = min(cropsize, w)
 49 | 
 50 |     w_space = w - cropsize
 51 |     h_space = h - cropsize
 52 | 
 53 |     if w_space > 0:
 54 |         cont_left = 0
 55 |         img_left = random.randrange(w_space + 1)
 56 |     else:
 57 |         cont_left = random.randrange(-w_space + 1)
 58 |         img_left = 0
 59 | 
 60 |     if h_space > 0:
 61 |         cont_top = 0
 62 |         img_top = random.randrange(h_space + 1)
 63 |     else:
 64 |         cont_top = random.randrange(-h_space + 1)
 65 |         img_top = 0
 66 | 
 67 |     img_container = np.zeros((cropsize, cropsize, imgarr.shape[-1]), np.float32)
 68 | 
 69 |     cropping = np.zeros((cropsize, cropsize), np.bool)
 70 | 
 71 |     img_container[cont_top:cont_top + ch, cont_left:cont_left + cw] = \
 72 |         imgarr[img_top:img_top + ch, img_left:img_left + cw]
 73 |     cropping[cont_top:cont_top + ch, cont_left:cont_left + cw] = 1
 74 | 
 75 |     return img_container, cropping
 76 | 
 77 | 
 78 | 
 79 | def crop(img_temp, dim, new_p=True, h_p=0, w_p=0):
 80 |     h = img_temp.shape[0]
 81 |     w = img_temp.shape[1]
 82 |     trig_h = trig_w = False
 83 |     if (h > dim):
 84 |         if (new_p):
 85 |             h_p = int(random.uniform(0, 1) * (h - dim))
 86 |         if len(img_temp.shape) == 2:
 87 |             img_temp = img_temp[h_p:h_p + dim, :]
 88 |         else:
 89 |             img_temp = img_temp[h_p:h_p + dim, :, :]
 90 |     elif (h < dim):
 91 |         trig_h = True
 92 |     if (w > dim):
 93 |         if (new_p):
 94 |             w_p = int(random.uniform(0, 1) * (w - dim))
 95 |         if len(img_temp.shape) == 2:
 96 |             img_temp = img_temp[:, w_p:w_p + dim]
 97 |         else:
 98 |             img_temp = img_temp[:, w_p:w_p + dim, :]
 99 |     elif (w < dim):
100 |         trig_w = True
101 |     if (trig_h or trig_w):
102 |         if len(img_temp.shape) == 2:
103 |             pad = np.zeros((dim, dim))
104 |             pad[:img_temp.shape[0], :img_temp.shape[1]] = img_temp
105 |         else:
106 |             pad = np.zeros((dim, dim, 3))
107 |             pad[:img_temp.shape[0], :img_temp.shape[1], :] = img_temp
108 |         return (pad, h_p, w_p)
109 |     else:
110 |         return (img_temp, h_p, w_p)
111 | 
112 | 
113 | def get_data_from_chunk(chunk, args):
114 |     img_path = args.img_path
115 | 
116 |     scale = np.random.uniform(0.7, 1.3)
117 |     dim = args.crop_size
118 |     images = np.zeros((dim, dim, 3, len(chunk)))
119 |     ori_images = np.zeros((dim, dim, 3, len(chunk)), dtype=np.uint8)
120 |     sal_images = np.zeros((dim, dim, len(chunk)))
121 |     gt_images = np.zeros((dim, dim, len(chunk)))
122 |     labels = load_image_label_list_from_npy(chunk)
123 |     labels = torch.from_numpy(np.array(labels))
124 | 
125 |     img_names = []
126 | 
127 |     for i, pieces in enumerate(chunk):
128 |         piece = pieces.replace('.jpg','').strip()
129 |         img_names.append(piece)
130 |         flip_p = np.random.uniform(0, 1)
131 |         img_temp = cv2.imread(os.path.join(img_path, piece + '.jpg'))
132 |         img_temp = cv2.cvtColor(img_temp, cv2.COLOR_BGR2RGB).astype(np.float)
133 |         sal_temp = cv2.imread(os.path.join(args.sal_pgt_path, piece + '.png'), 0) / 255.
134 | 
135 |         gt_temp = np.asarray(Image.open(os.path.join(args.seg_pgt_path, piece + '.png')))
136 |         img_temp = scale_im(img_temp, scale)
137 |         sal_temp = scale_im(sal_temp, scale)
138 |         gt_temp = scale_gt(gt_temp, scale)
139 |         img_temp = flip(img_temp, flip_p)
140 |         sal_temp = flip(sal_temp, flip_p)
141 |         gt_temp = flip(gt_temp, flip_p)
142 | 
143 |         img_temp[:, :, 0] = (img_temp[:, :, 0] / 255. - 0.485) / 0.229
144 |         img_temp[:, :, 1] = (img_temp[:, :, 1] / 255. - 0.456) / 0.224
145 |         img_temp[:, :, 2] = (img_temp[:, :, 2] / 255. - 0.406) / 0.225
146 | 
147 |         img_temp, img_temp_h_p, img_temp_w_p = crop(img_temp, dim)
148 |         gt_temp = crop(gt_temp, dim, False, img_temp_h_p, img_temp_w_p)[0]
149 |         gt_images[:, :, i] = gt_temp[:, :]
150 |         sal_temp = crop(sal_temp, dim, False, img_temp_h_p, img_temp_w_p)[0]
151 |         sal_temp[sal_temp > 0.5] = 1.0
152 |         sal_temp[sal_temp <= 0.5] = 0.0
153 |         sal_images[:, :, i] = sal_temp[:, :]
154 | 
155 |         ori_temp = np.zeros_like(img_temp)
156 |         ori_temp[:, :, 0] = (img_temp[:, :, 0] * 0.229 + 0.485) * 255.
157 |         ori_temp[:, :, 1] = (img_temp[:, :, 1] * 0.224 + 0.456) * 255.
158 |         ori_temp[:, :, 2] = (img_temp[:, :, 2] * 0.225 + 0.406) * 255.
159 |         ori_images[:, :, :, i] = ori_temp.astype(np.uint8)
160 | 
161 |         images[:, :, :, i] = img_temp
162 | 
163 |     images = images.transpose((3, 2, 0, 1))
164 |     ori_images = ori_images.transpose((3, 2, 0, 1))
165 |     gt_images = gt_images.transpose((2, 0, 1))
166 |     gt_images = torch.from_numpy(gt_images).float()
167 |     sal_images = sal_images.transpose((2, 0, 1))
168 |     sal_images = torch.from_numpy(sal_images).float()
169 |     images = torch.from_numpy(images).float()
170 |     return images, ori_images, sal_images, gt_images, labels, img_names
171 | 


--------------------------------------------------------------------------------
/tool/imutils.py:
--------------------------------------------------------------------------------
  1 | import numpy as np
  2 | import pydensecrf.densecrf as dcrf
  3 | import torch
  4 | import PIL.Image as Image
  5 | import os
  6 | import cv2
  7 | 
  8 | 
  9 | palette = [0, 0, 0, 128, 0, 0, 0, 128, 0, 128, 128, 0, 0, 0, 128, 128, 0, 128, 0, 128, 128, 128, 128, 128,
 10 |            64, 0, 0, 192, 0, 0, 64, 128, 0, 192, 128, 0, 64, 0, 128, 192, 0, 128, 64, 128, 128, 192, 128, 128,
 11 |            0, 64, 0, 128, 64, 0, 0, 192, 0, 128, 192, 0, 0, 64, 128, 128, 64, 128, 0, 192, 128, 128, 192, 128,
 12 |            64, 64, 0, 192, 64, 0, 64, 192, 0, 192, 192, 0]
 13 | 
 14 | def put_palette(seg_label, out_name):
 15 |     out = seg_label.astype(np.uint8)
 16 |     out = Image.fromarray(out, mode='P')
 17 |     out.putpalette(palette)
 18 |     out.save(out_name)
 19 | 
 20 | def show_cam_on_image(img, mask, img_name, save_path):
 21 |     img = np.float32(img) / 255.
 22 |     heatmap = cv2.applyColorMap(np.uint8(255 * mask), cv2.COLORMAP_JET)
 23 |     heatmap = np.float32(heatmap) / 255
 24 |     cam = heatmap + img
 25 |     cam = cam / np.max(cam)
 26 |     cam = np.uint8(255 * cam)
 27 |     cv2.imwrite(os.path.join(save_path, img_name + ".jpg"), cam)
 28 | 
 29 | def save_as_png(mask, save_name, save_path):
 30 |     out_name = save_name + '.png'
 31 |     out = mask * 255
 32 |     out = Image.fromarray(out.astype(np.uint8), mode='P')
 33 |     out.save(os.path.join(save_path, out_name))
 34 | 
 35 | 
 36 | def fusion_cam(cam, sal_images, label, args, name='toy'):
 37 |     if len(sal_images.size()) == 4:
 38 |         sal_images = sal_images.squeeze(1)
 39 | 
 40 |     cam = cam * label.unsqueeze(-1).unsqueeze(-1).cuda()
 41 |     cam_max = torch.max(cam, dim=1)[0]
 42 | 
 43 |     cam_object_region = cam_max > 0.2
 44 |     sal_object_region = sal_images > 0.06
 45 |     object_region = sal_object_region * cam_object_region
 46 |     not_sure_region1 = sal_object_region * (~cam_object_region)
 47 |     not_sure_region2 = (~sal_object_region) * (cam_object_region)
 48 |     not_sure_region = not_sure_region1 | not_sure_region2
 49 | 
 50 |     gt = torch.argmax(cam, dim=1) + 1
 51 |     gt = gt.float() * object_region.float()
 52 | 
 53 |     for b in range(gt.size(0)):
 54 |         gt[b, not_sure_region[b]] = 255
 55 |         seg_name = os.path.join(args.seg_pgt_path, name[b] + '.png')
 56 |         output_i = gt[b].detach().cpu().numpy()
 57 |         put_palette(output_i, seg_name)
 58 |     return gt.long()
 59 | 
 60 | def crf_inference_inf(img, probs, t=10, scale_factor=1, labels=21):
 61 |     import pydensecrf.densecrf as dcrf
 62 |     from pydensecrf.utils import unary_from_softmax
 63 | 
 64 |     h, w = img.shape[:2]
 65 |     n_labels = labels
 66 | 
 67 |     d = dcrf.DenseCRF2D(w, h, n_labels)
 68 | 
 69 |     unary = unary_from_softmax(probs)
 70 |     unary = np.ascontiguousarray(unary)
 71 | 
 72 |     img_c = np.ascontiguousarray(img)
 73 | 
 74 |     d.setUnaryEnergy(unary)
 75 |     d.addPairwiseGaussian(sxy=4/scale_factor, compat=3)
 76 |     d.addPairwiseBilateral(sxy=83/scale_factor, srgb=5, rgbim=np.copy(img_c), compat=3)
 77 |     Q = d.inference(t)
 78 | 
 79 |     return np.array(Q).reshape((n_labels, h, w))
 80 | 
 81 | def sigmoid(x):
 82 |     return 1 / (1 + np.exp(-x))
 83 | 
 84 | def crf_for_sal(img, anno, EPSILON=1e-8, tau=1.05):
 85 |     img = np.array(img)
 86 |     d = dcrf.DenseCRF2D(img.shape[1], img.shape[0], 2)
 87 |     n_energy = -np.log((1.0 - anno + EPSILON)) / (tau * sigmoid(1 - anno))
 88 |     p_energy = -np.log(anno + EPSILON) / (tau * sigmoid(anno))
 89 |     U = np.zeros((2, img.shape[0] * img.shape[1]), dtype='float32')
 90 |     U[0, :] = n_energy.flatten()
 91 |     U[1, :] = p_energy.flatten()
 92 |     d.setUnaryEnergy(U)
 93 | 
 94 |     d.addPairwiseGaussian(sxy=3, compat=3)
 95 |     d.addPairwiseBilateral(sxy=60, srgb=5, rgbim=np.ascontiguousarray(np.copy(img)), compat=5)
 96 | 
 97 |     # Do the inference
 98 |     infer = np.array(d.inference(1)).astype('float32')
 99 |     res = np.expand_dims(infer[1, :].reshape(img.shape[:2]), 0)
100 | 
101 |     return res


--------------------------------------------------------------------------------
/tool/pyutils.py:
--------------------------------------------------------------------------------
  1 | import time
  2 | import sys
  3 | 
  4 | class Logger(object):
  5 |     def __init__(self, outfile):
  6 |         self.terminal = sys.stdout
  7 |         self.log = open(outfile, "w")
  8 |         sys.stdout = self
  9 | 
 10 |     def write(self, message):
 11 |         self.terminal.write(message)
 12 |         self.log.write(message)
 13 | 
 14 |     def flush(self):
 15 |         self.terminal.flush()
 16 | 
 17 | 
 18 | class AverageMeter:
 19 |     def __init__(self, *keys):
 20 |         self.__data = dict()
 21 |         for k in keys:
 22 |             self.__data[k] = [0.0, 0]
 23 | 
 24 |     def add(self, dict):
 25 |         for k, v in dict.items():
 26 |             self.__data[k][0] += v
 27 |             self.__data[k][1] += 1
 28 | 
 29 |     def get(self, *keys):
 30 |         if len(keys) == 1:
 31 |             return self.__data[keys[0]][0] / self.__data[keys[0]][1]
 32 |         else:
 33 |             v_list = [self.__data[k][0] / self.__data[k][1] for k in keys]
 34 |             return tuple(v_list)
 35 | 
 36 |     def pop(self, key=None):
 37 |         if key is None:
 38 |             for k in self.__data.keys():
 39 |                 self.__data[k] = [0.0, 0]
 40 |         else:
 41 |             v = self.get(key)
 42 |             self.__data[key] = [0.0, 0]
 43 |             return v
 44 | 
 45 | 
 46 | class Timer:
 47 |     def __init__(self, starting_msg = None):
 48 |         self.start = time.time()
 49 |         self.stage_start = self.start
 50 | 
 51 |         if starting_msg is not None:
 52 |             print(starting_msg, time.ctime(time.time()))
 53 | 
 54 | 
 55 |     def update_progress(self, progress):
 56 |         self.elapsed = time.time() - self.start
 57 |         self.est_total = self.elapsed / progress
 58 |         self.est_remaining = self.est_total - self.elapsed
 59 |         self.est_finish = int(self.start + self.est_total)
 60 | 
 61 | 
 62 |     def str_est_finish(self):
 63 |         return str(time.ctime(self.est_finish))
 64 | 
 65 |     def get_stage_elapsed(self):
 66 |         return time.time() - self.stage_start
 67 | 
 68 |     def reset_stage(self):
 69 |         self.stage_start = time.time()
 70 | 
 71 | 
 72 | from multiprocessing.pool import ThreadPool
 73 | 
 74 | class BatchThreader:
 75 | 
 76 |     def __init__(self, func, args_list, batch_size, prefetch_size=4, processes=12):
 77 |         self.batch_size = batch_size
 78 |         self.prefetch_size = prefetch_size
 79 | 
 80 |         self.pool = ThreadPool(processes=processes)
 81 |         self.async_result = []
 82 | 
 83 |         self.func = func
 84 |         self.left_args_list = args_list
 85 |         self.n_tasks = len(args_list)
 86 | 
 87 |         # initial work
 88 |         self.__start_works(self.__get_n_pending_works())
 89 | 
 90 | 
 91 |     def __start_works(self, times):
 92 |         for _ in range(times):
 93 |             args = self.left_args_list.pop(0)
 94 |             self.async_result.append(
 95 |                 self.pool.apply_async(self.func, args))
 96 | 
 97 | 
 98 |     def __get_n_pending_works(self):
 99 |         return min((self.prefetch_size + 1) * self.batch_size - len(self.async_result)
100 |                    , len(self.left_args_list))
101 | 
102 | 
103 | 
104 |     def pop_results(self):
105 | 
106 |         n_inwork = len(self.async_result)
107 | 
108 |         n_fetch = min(n_inwork, self.batch_size)
109 |         rtn = [self.async_result.pop(0).get()
110 |                 for _ in range(n_fetch)]
111 | 
112 |         to_fill = self.__get_n_pending_works()
113 |         if to_fill == 0:
114 |             self.pool.close()
115 |         else:
116 |             self.__start_works(to_fill)
117 | 
118 |         return rtn
119 | 
120 | 
121 | 
122 | 
123 | 
124 | 
125 | 


--------------------------------------------------------------------------------
/tool/torchutils.py:
--------------------------------------------------------------------------------
 1 | 
 2 | import torch
 3 | 
 4 | class PolyOptimizer(torch.optim.SGD):
 5 | 
 6 |     def __init__(self, params, lr, weight_decay, max_step, momentum=0.9):
 7 |         super().__init__(params, lr, weight_decay)
 8 | 
 9 |         self.global_step = 0
10 |         self.max_step = max_step
11 |         self.momentum = momentum
12 | 
13 |         self.__initial_lr = [group['lr'] for group in self.param_groups]
14 | 
15 | 
16 |     def step(self, closure=None):
17 |         if self.global_step < (0.5*self.max_step):
18 |             lr_mult = (1 - self.global_step / (0.5*self.max_step)) ** self.momentum
19 | 
20 |             for i in range(len(self.param_groups)):
21 |                 self.param_groups[i]['lr'] = self.__initial_lr[i] * lr_mult
22 | 
23 |         elif self.global_step < self.max_step:
24 |              lr_mult = (1 - (self.global_step-(0.5*self.max_step)) / (self.max_step-(0.5*self.max_step))) ** self.momentum
25 | 
26 |              for i in range(len(self.param_groups)):
27 |                  self.param_groups[i]['lr'] = 0.0007 * lr_mult
28 | 
29 |         super().step(closure)
30 | 
31 |         self.global_step += 1
32 | 
33 | 
34 | class PolyOptimizer_cls(torch.optim.SGD):
35 | 
36 |     def __init__(self, params, lr, weight_decay, max_step, momentum=0.9):
37 |         super().__init__(params, lr, weight_decay)
38 | 
39 |         self.global_step = 0
40 |         self.max_step = max_step
41 |         self.momentum = momentum
42 | 
43 |         self.__initial_lr = [group['lr'] for group in self.param_groups]
44 | 
45 | 
46 |     def step(self, closure=None):
47 | 
48 |         if self.global_step < self.max_step:
49 |             lr_mult = (1 - self.global_step / self.max_step) ** self.momentum
50 | 
51 |             for i in range(len(self.param_groups)):
52 |                 if i == 4:
53 |                     self.param_groups[i]['lr'] = self.__initial_lr[i]
54 |                 else:
55 |                     self.param_groups[i]['lr'] = self.__initial_lr[i] * lr_mult
56 | 
57 |         super().step(closure)
58 | 
59 |         self.global_step += 1
60 | 


--------------------------------------------------------------------------------
/train_AuxAff.py:
--------------------------------------------------------------------------------
  1 | from torch.backends import cudnn
  2 | 
  3 | cudnn.enabled = True
  4 | from tool import pyutils, torchutils
  5 | import argparse
  6 | import importlib
  7 | import tool.exutils as exutils
  8 | 
  9 | import torch.nn.functional as F
 10 | from pathlib import Path
 11 | 
 12 | import torch
 13 | import os
 14 | import numpy as np
 15 | 
 16 | 
 17 | if __name__ == '__main__':
 18 | 
 19 |     parser = argparse.ArgumentParser()
 20 |     parser.add_argument('--gpu_ids', type=str, default='0', help='GPU_id')
 21 | 
 22 |     parser.add_argument("--list_path", default="voc12/train_aug_id.txt", type=str)
 23 |     parser.add_argument("--img_path", default="", type=str)
 24 |     parser.add_argument("--save_path", default=None, type=str)
 25 |     parser.add_argument("--sal_pgt_path", default=None, type=str)
 26 |     parser.add_argument("--seg_pgt_path", default=None, type=str)
 27 | 
 28 |     parser.add_argument("--batch_size", default=4, type=int)
 29 |     parser.add_argument("--num_classes", default=21, type=int)
 30 |     parser.add_argument("--num_epochs", default=15, type=int)
 31 |     parser.add_argument("--network", default='AuxSegNet', type=str)
 32 |     parser.add_argument("--lr", default=0.0007, type=float)
 33 |     parser.add_argument("--num_workers", default=16, type=int)
 34 |     parser.add_argument("--wt_dec", default=1e-5, type=float)
 35 |     parser.add_argument("--init_weights", default='', type=str)
 36 | 
 37 |     parser.add_argument("--session_name", default="AuxSegNet_", type=str)
 38 |     parser.add_argument("--crop_size", default=321, type=int)
 39 | 
 40 |     parser.add_argument('--print_intervals', type=int, default=50)
 41 | 
 42 |     parser.add_argument('--sal_loss_weight', type=float, default=1.0)
 43 |     parser.add_argument('--cls_loss_weight', type=float, default=1.0)
 44 |     parser.add_argument('--seg_loss_weight', type=float, default=1.0)
 45 | 
 46 |     args = parser.parse_args()
 47 | 
 48 |     gpu_id = args.gpu_ids
 49 |     os.environ["CUDA_VISIBLE_DEVICES"] = gpu_id
 50 | 
 51 |     save_path = args.save_path
 52 |     Path(args.save_path).mkdir(parents=True, exist_ok=True)
 53 | 
 54 |     pyutils.Logger(os.path.join(args.save_path, args.session_name + '.log'))
 55 | 
 56 |     criterion = torch.nn.CrossEntropyLoss(weight=None, ignore_index=255, reduction='elementwise_mean').cuda()
 57 | 
 58 |     model = getattr(importlib.import_module('network.' + args.network), 'SegNet')(num_classes=args.num_classes)
 59 | 
 60 | 
 61 |     weights_dict = torch.load(args.init_weights)
 62 |     model.load_state_dict(weights_dict, strict=False)
 63 | 
 64 | 
 65 |     img_list = exutils.read_file(args.list_path)
 66 |     train_size = len(img_list)
 67 |     max_step = (train_size // args.batch_size) * args.num_epochs
 68 | 
 69 |     data_list = []
 70 |     for i in range(200):
 71 |         np.random.shuffle(img_list)
 72 |         data_list.extend(img_list)
 73 | 
 74 |     optimizer = torchutils.PolyOptimizer_cls([
 75 |         {'params': model.get_1x_lr_params(), 'lr': args.lr},
 76 |         {'params': model.get_10x_lr_params(), 'lr': 10 * args.lr}
 77 |     ], lr=args.lr, weight_decay=args.wt_dec, max_step=max_step)
 78 | 
 79 |     model = torch.nn.DataParallel(model).cuda()
 80 |     model.train()
 81 | 
 82 |     avg_meter = pyutils.AverageMeter('loss')
 83 | 
 84 |     timer = pyutils.Timer("Session started: ")
 85 | 
 86 |     data_gen = exutils.chunker(data_list, args.batch_size)
 87 | 
 88 |     for iter in range(max_step + 1):
 89 |         chunk = data_gen.__next__()
 90 |         img_list = chunk
 91 | 
 92 |         images, ori_images, sal_images, seg_labels, label, img_names = exutils.get_data_from_chunk(chunk, args)
 93 | 
 94 |         b, _, w, h = ori_images.shape
 95 |         label = label.cuda(non_blocking=True)
 96 |         seg_labels = seg_labels.long().cuda()
 97 |         sal_images = sal_images.cuda()
 98 |         images = images.cuda()
 99 | 
100 |         x_cls, init_sal, refined_sal, init_seg, refined_seg = model(x=images)
101 | 
102 |         ##################################### image classification ###########################################
103 |         cls_loss = F.multilabel_soft_margin_loss(x_cls, label)
104 | 
105 |         ##################################### saliency detection ###########################################
106 |         init_sal = F.interpolate(init_sal, size=(w, h), mode='bilinear', align_corners=False)
107 |         sal_labels = sal_images.unsqueeze(1)
108 |         init_sal_loss = F.binary_cross_entropy_with_logits(init_sal, sal_labels)
109 | 
110 |         refined_sal = F.interpolate(refined_sal, size=(w, h), mode='bilinear', align_corners=False)
111 |         refined_sal_loss = F.binary_cross_entropy_with_logits(refined_sal, sal_labels)
112 | 
113 |         ##################################### segmentation ###########################################
114 |         init_seg = F.interpolate(init_seg, size=(w, h), mode='bilinear', align_corners=False)
115 |         init_seg_loss = criterion(init_seg, seg_labels)
116 | 
117 |         refined_seg = F.interpolate(refined_seg, size=(w, h), mode='bilinear', align_corners=False)
118 |         refined_seg_loss = criterion(refined_seg, seg_labels)
119 | 
120 | 
121 |         loss = args.cls_loss_weight * cls_loss + args.sal_loss_weight * init_sal_loss + \
122 |                args.sal_loss_weight * refined_sal_loss + args.seg_loss_weight * init_seg_loss + args.seg_loss_weight * refined_seg_loss
123 | 
124 |         avg_meter.add({'loss': loss.item()})
125 | 
126 |         optimizer.zero_grad()
127 |         loss.backward()
128 |         optimizer.step()
129 | 
130 |         if (optimizer.global_step - 1) % args.print_intervals == 0:
131 |             timer.update_progress(optimizer.global_step / max_step)
132 | 
133 |             print('Iter:%5d/%5d' % (optimizer.global_step - 1, max_step),
134 |                   'Loss:%.4f' % (avg_meter.pop('loss')),
135 |                   'imps:%.1f' % ((iter + 1) * args.batch_size / timer.get_stage_elapsed()),
136 |                   'Fin:%s' % (timer.str_est_finish()),
137 |                   'lr: %.5f' % (optimizer.param_groups[0]['lr']), flush=True)
138 | 
139 |     torch.save(model.module.state_dict(), os.path.join(save_path, args.session_name + 'final.pth'))


--------------------------------------------------------------------------------
/voc12/cls_labels.npy:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/xulianuwa/AuxSegNet/9f40a5d349e2405b42a541a3534e51eb58b04077/voc12/cls_labels.npy


--------------------------------------------------------------------------------
/voc12/test.txt:
--------------------------------------------------------------------------------
   1 | /JPEGImages/2008_000006.jpg
   2 | /JPEGImages/2008_000011.jpg
   3 | /JPEGImages/2008_000012.jpg
   4 | /JPEGImages/2008_000018.jpg
   5 | /JPEGImages/2008_000024.jpg
   6 | /JPEGImages/2008_000030.jpg
   7 | /JPEGImages/2008_000031.jpg
   8 | /JPEGImages/2008_000046.jpg
   9 | /JPEGImages/2008_000047.jpg
  10 | /JPEGImages/2008_000048.jpg
  11 | /JPEGImages/2008_000057.jpg
  12 | /JPEGImages/2008_000058.jpg
  13 | /JPEGImages/2008_000068.jpg
  14 | /JPEGImages/2008_000072.jpg
  15 | /JPEGImages/2008_000079.jpg
  16 | /JPEGImages/2008_000081.jpg
  17 | /JPEGImages/2008_000083.jpg
  18 | /JPEGImages/2008_000088.jpg
  19 | /JPEGImages/2008_000094.jpg
  20 | /JPEGImages/2008_000101.jpg
  21 | /JPEGImages/2008_000104.jpg
  22 | /JPEGImages/2008_000106.jpg
  23 | /JPEGImages/2008_000108.jpg
  24 | /JPEGImages/2008_000110.jpg
  25 | /JPEGImages/2008_000111.jpg
  26 | /JPEGImages/2008_000126.jpg
  27 | /JPEGImages/2008_000127.jpg
  28 | /JPEGImages/2008_000129.jpg
  29 | /JPEGImages/2008_000130.jpg
  30 | /JPEGImages/2008_000135.jpg
  31 | /JPEGImages/2008_000150.jpg
  32 | /JPEGImages/2008_000152.jpg
  33 | /JPEGImages/2008_000156.jpg
  34 | /JPEGImages/2008_000159.jpg
  35 | /JPEGImages/2008_000160.jpg
  36 | /JPEGImages/2008_000161.jpg
  37 | /JPEGImages/2008_000166.jpg
  38 | /JPEGImages/2008_000167.jpg
  39 | /JPEGImages/2008_000168.jpg
  40 | /JPEGImages/2008_000169.jpg
  41 | /JPEGImages/2008_000171.jpg
  42 | /JPEGImages/2008_000175.jpg
  43 | /JPEGImages/2008_000178.jpg
  44 | /JPEGImages/2008_000186.jpg
  45 | /JPEGImages/2008_000198.jpg
  46 | /JPEGImages/2008_000206.jpg
  47 | /JPEGImages/2008_000208.jpg
  48 | /JPEGImages/2008_000209.jpg
  49 | /JPEGImages/2008_000211.jpg
  50 | /JPEGImages/2008_000220.jpg
  51 | /JPEGImages/2008_000224.jpg
  52 | /JPEGImages/2008_000230.jpg
  53 | /JPEGImages/2008_000240.jpg
  54 | /JPEGImages/2008_000248.jpg
  55 | /JPEGImages/2008_000249.jpg
  56 | /JPEGImages/2008_000250.jpg
  57 | /JPEGImages/2008_000256.jpg
  58 | /JPEGImages/2008_000279.jpg
  59 | /JPEGImages/2008_000282.jpg
  60 | /JPEGImages/2008_000285.jpg
  61 | /JPEGImages/2008_000286.jpg
  62 | /JPEGImages/2008_000296.jpg
  63 | /JPEGImages/2008_000300.jpg
  64 | /JPEGImages/2008_000322.jpg
  65 | /JPEGImages/2008_000324.jpg
  66 | /JPEGImages/2008_000337.jpg
  67 | /JPEGImages/2008_000366.jpg
  68 | /JPEGImages/2008_000369.jpg
  69 | /JPEGImages/2008_000377.jpg
  70 | /JPEGImages/2008_000384.jpg
  71 | /JPEGImages/2008_000390.jpg
  72 | /JPEGImages/2008_000404.jpg
  73 | /JPEGImages/2008_000411.jpg
  74 | /JPEGImages/2008_000434.jpg
  75 | /JPEGImages/2008_000440.jpg
  76 | /JPEGImages/2008_000460.jpg
  77 | /JPEGImages/2008_000467.jpg
  78 | /JPEGImages/2008_000478.jpg
  79 | /JPEGImages/2008_000485.jpg
  80 | /JPEGImages/2008_000487.jpg
  81 | /JPEGImages/2008_000490.jpg
  82 | /JPEGImages/2008_000503.jpg
  83 | /JPEGImages/2008_000504.jpg
  84 | /JPEGImages/2008_000507.jpg
  85 | /JPEGImages/2008_000513.jpg
  86 | /JPEGImages/2008_000523.jpg
  87 | /JPEGImages/2008_000529.jpg
  88 | /JPEGImages/2008_000556.jpg
  89 | /JPEGImages/2008_000565.jpg
  90 | /JPEGImages/2008_000580.jpg
  91 | /JPEGImages/2008_000590.jpg
  92 | /JPEGImages/2008_000596.jpg
  93 | /JPEGImages/2008_000597.jpg
  94 | /JPEGImages/2008_000600.jpg
  95 | /JPEGImages/2008_000603.jpg
  96 | /JPEGImages/2008_000604.jpg
  97 | /JPEGImages/2008_000612.jpg
  98 | /JPEGImages/2008_000617.jpg
  99 | /JPEGImages/2008_000621.jpg
 100 | /JPEGImages/2008_000627.jpg
 101 | /JPEGImages/2008_000633.jpg
 102 | /JPEGImages/2008_000643.jpg
 103 | /JPEGImages/2008_000644.jpg
 104 | /JPEGImages/2008_000649.jpg
 105 | /JPEGImages/2008_000651.jpg
 106 | /JPEGImages/2008_000664.jpg
 107 | /JPEGImages/2008_000665.jpg
 108 | /JPEGImages/2008_000680.jpg
 109 | /JPEGImages/2008_000681.jpg
 110 | /JPEGImages/2008_000684.jpg
 111 | /JPEGImages/2008_000685.jpg
 112 | /JPEGImages/2008_000688.jpg
 113 | /JPEGImages/2008_000693.jpg
 114 | /JPEGImages/2008_000698.jpg
 115 | /JPEGImages/2008_000707.jpg
 116 | /JPEGImages/2008_000709.jpg
 117 | /JPEGImages/2008_000712.jpg
 118 | /JPEGImages/2008_000747.jpg
 119 | /JPEGImages/2008_000751.jpg
 120 | /JPEGImages/2008_000754.jpg
 121 | /JPEGImages/2008_000762.jpg
 122 | /JPEGImages/2008_000767.jpg
 123 | /JPEGImages/2008_000768.jpg
 124 | /JPEGImages/2008_000773.jpg
 125 | /JPEGImages/2008_000774.jpg
 126 | /JPEGImages/2008_000779.jpg
 127 | /JPEGImages/2008_000797.jpg
 128 | /JPEGImages/2008_000813.jpg
 129 | /JPEGImages/2008_000816.jpg
 130 | /JPEGImages/2008_000846.jpg
 131 | /JPEGImages/2008_000866.jpg
 132 | /JPEGImages/2008_000871.jpg
 133 | /JPEGImages/2008_000872.jpg
 134 | /JPEGImages/2008_000891.jpg
 135 | /JPEGImages/2008_000892.jpg
 136 | /JPEGImages/2008_000894.jpg
 137 | /JPEGImages/2008_000896.jpg
 138 | /JPEGImages/2008_000898.jpg
 139 | /JPEGImages/2008_000909.jpg
 140 | /JPEGImages/2008_000913.jpg
 141 | /JPEGImages/2008_000920.jpg
 142 | /JPEGImages/2008_000933.jpg
 143 | /JPEGImages/2008_000935.jpg
 144 | /JPEGImages/2008_000937.jpg
 145 | /JPEGImages/2008_000938.jpg
 146 | /JPEGImages/2008_000954.jpg
 147 | /JPEGImages/2008_000958.jpg
 148 | /JPEGImages/2008_000963.jpg
 149 | /JPEGImages/2008_000967.jpg
 150 | /JPEGImages/2008_000974.jpg
 151 | /JPEGImages/2008_000986.jpg
 152 | /JPEGImages/2008_000994.jpg
 153 | /JPEGImages/2008_000995.jpg
 154 | /JPEGImages/2008_001008.jpg
 155 | /JPEGImages/2008_001010.jpg
 156 | /JPEGImages/2008_001014.jpg
 157 | /JPEGImages/2008_001016.jpg
 158 | /JPEGImages/2008_001025.jpg
 159 | /JPEGImages/2008_001029.jpg
 160 | /JPEGImages/2008_001037.jpg
 161 | /JPEGImages/2008_001059.jpg
 162 | /JPEGImages/2008_001061.jpg
 163 | /JPEGImages/2008_001072.jpg
 164 | /JPEGImages/2008_001124.jpg
 165 | /JPEGImages/2008_001126.jpg
 166 | /JPEGImages/2008_001131.jpg
 167 | /JPEGImages/2008_001138.jpg
 168 | /JPEGImages/2008_001144.jpg
 169 | /JPEGImages/2008_001151.jpg
 170 | /JPEGImages/2008_001156.jpg
 171 | /JPEGImages/2008_001179.jpg
 172 | /JPEGImages/2008_001181.jpg
 173 | /JPEGImages/2008_001184.jpg
 174 | /JPEGImages/2008_001186.jpg
 175 | /JPEGImages/2008_001197.jpg
 176 | /JPEGImages/2008_001207.jpg
 177 | /JPEGImages/2008_001212.jpg
 178 | /JPEGImages/2008_001233.jpg
 179 | /JPEGImages/2008_001234.jpg
 180 | /JPEGImages/2008_001258.jpg
 181 | /JPEGImages/2008_001268.jpg
 182 | /JPEGImages/2008_001279.jpg
 183 | /JPEGImages/2008_001281.jpg
 184 | /JPEGImages/2008_001288.jpg
 185 | /JPEGImages/2008_001291.jpg
 186 | /JPEGImages/2008_001298.jpg
 187 | /JPEGImages/2008_001309.jpg
 188 | /JPEGImages/2008_001315.jpg
 189 | /JPEGImages/2008_001316.jpg
 190 | /JPEGImages/2008_001319.jpg
 191 | /JPEGImages/2008_001327.jpg
 192 | /JPEGImages/2008_001328.jpg
 193 | /JPEGImages/2008_001332.jpg
 194 | /JPEGImages/2008_001341.jpg
 195 | /JPEGImages/2008_001347.jpg
 196 | /JPEGImages/2008_001355.jpg
 197 | /JPEGImages/2008_001378.jpg
 198 | /JPEGImages/2008_001386.jpg
 199 | /JPEGImages/2008_001400.jpg
 200 | /JPEGImages/2008_001409.jpg
 201 | /JPEGImages/2008_001411.jpg
 202 | /JPEGImages/2008_001416.jpg
 203 | /JPEGImages/2008_001418.jpg
 204 | /JPEGImages/2008_001435.jpg
 205 | /JPEGImages/2008_001459.jpg
 206 | /JPEGImages/2008_001469.jpg
 207 | /JPEGImages/2008_001474.jpg
 208 | /JPEGImages/2008_001477.jpg
 209 | /JPEGImages/2008_001483.jpg
 210 | /JPEGImages/2008_001484.jpg
 211 | /JPEGImages/2008_001485.jpg
 212 | /JPEGImages/2008_001496.jpg
 213 | /JPEGImages/2008_001507.jpg
 214 | /JPEGImages/2008_001511.jpg
 215 | /JPEGImages/2008_001519.jpg
 216 | /JPEGImages/2008_001557.jpg
 217 | /JPEGImages/2008_001567.jpg
 218 | /JPEGImages/2008_001570.jpg
 219 | /JPEGImages/2008_001571.jpg
 220 | /JPEGImages/2008_001572.jpg
 221 | /JPEGImages/2008_001579.jpg
 222 | /JPEGImages/2008_001587.jpg
 223 | /JPEGImages/2008_001608.jpg
 224 | /JPEGImages/2008_001611.jpg
 225 | /JPEGImages/2008_001614.jpg
 226 | /JPEGImages/2008_001621.jpg
 227 | /JPEGImages/2008_001639.jpg
 228 | /JPEGImages/2008_001658.jpg
 229 | /JPEGImages/2008_001678.jpg
 230 | /JPEGImages/2008_001700.jpg
 231 | /JPEGImages/2008_001713.jpg
 232 | /JPEGImages/2008_001720.jpg
 233 | /JPEGImages/2008_001755.jpg
 234 | /JPEGImages/2008_001779.jpg
 235 | /JPEGImages/2008_001785.jpg
 236 | /JPEGImages/2008_001793.jpg
 237 | /JPEGImages/2008_001794.jpg
 238 | /JPEGImages/2008_001803.jpg
 239 | /JPEGImages/2008_001818.jpg
 240 | /JPEGImages/2008_001848.jpg
 241 | /JPEGImages/2008_001855.jpg
 242 | /JPEGImages/2008_001857.jpg
 243 | /JPEGImages/2008_001861.jpg
 244 | /JPEGImages/2008_001875.jpg
 245 | /JPEGImages/2008_001878.jpg
 246 | /JPEGImages/2008_001886.jpg
 247 | /JPEGImages/2008_001897.jpg
 248 | /JPEGImages/2008_001916.jpg
 249 | /JPEGImages/2008_001925.jpg
 250 | /JPEGImages/2008_001949.jpg
 251 | /JPEGImages/2008_001953.jpg
 252 | /JPEGImages/2008_001972.jpg
 253 | /JPEGImages/2008_001999.jpg
 254 | /JPEGImages/2008_002027.jpg
 255 | /JPEGImages/2008_002040.jpg
 256 | /JPEGImages/2008_002057.jpg
 257 | /JPEGImages/2008_002070.jpg
 258 | /JPEGImages/2008_002075.jpg
 259 | /JPEGImages/2008_002095.jpg
 260 | /JPEGImages/2008_002104.jpg
 261 | /JPEGImages/2008_002105.jpg
 262 | /JPEGImages/2008_002106.jpg
 263 | /JPEGImages/2008_002136.jpg
 264 | /JPEGImages/2008_002137.jpg
 265 | /JPEGImages/2008_002147.jpg
 266 | /JPEGImages/2008_002149.jpg
 267 | /JPEGImages/2008_002163.jpg
 268 | /JPEGImages/2008_002173.jpg
 269 | /JPEGImages/2008_002174.jpg
 270 | /JPEGImages/2008_002184.jpg
 271 | /JPEGImages/2008_002186.jpg
 272 | /JPEGImages/2008_002188.jpg
 273 | /JPEGImages/2008_002190.jpg
 274 | /JPEGImages/2008_002203.jpg
 275 | /JPEGImages/2008_002211.jpg
 276 | /JPEGImages/2008_002217.jpg
 277 | /JPEGImages/2008_002228.jpg
 278 | /JPEGImages/2008_002233.jpg
 279 | /JPEGImages/2008_002246.jpg
 280 | /JPEGImages/2008_002257.jpg
 281 | /JPEGImages/2008_002261.jpg
 282 | /JPEGImages/2008_002285.jpg
 283 | /JPEGImages/2008_002287.jpg
 284 | /JPEGImages/2008_002295.jpg
 285 | /JPEGImages/2008_002303.jpg
 286 | /JPEGImages/2008_002306.jpg
 287 | /JPEGImages/2008_002309.jpg
 288 | /JPEGImages/2008_002310.jpg
 289 | /JPEGImages/2008_002318.jpg
 290 | /JPEGImages/2008_002320.jpg
 291 | /JPEGImages/2008_002332.jpg
 292 | /JPEGImages/2008_002337.jpg
 293 | /JPEGImages/2008_002345.jpg
 294 | /JPEGImages/2008_002348.jpg
 295 | /JPEGImages/2008_002352.jpg
 296 | /JPEGImages/2008_002360.jpg
 297 | /JPEGImages/2008_002381.jpg
 298 | /JPEGImages/2008_002387.jpg
 299 | /JPEGImages/2008_002388.jpg
 300 | /JPEGImages/2008_002393.jpg
 301 | /JPEGImages/2008_002406.jpg
 302 | /JPEGImages/2008_002440.jpg
 303 | /JPEGImages/2008_002455.jpg
 304 | /JPEGImages/2008_002460.jpg
 305 | /JPEGImages/2008_002462.jpg
 306 | /JPEGImages/2008_002480.jpg
 307 | /JPEGImages/2008_002518.jpg
 308 | /JPEGImages/2008_002525.jpg
 309 | /JPEGImages/2008_002535.jpg
 310 | /JPEGImages/2008_002544.jpg
 311 | /JPEGImages/2008_002553.jpg
 312 | /JPEGImages/2008_002569.jpg
 313 | /JPEGImages/2008_002572.jpg
 314 | /JPEGImages/2008_002587.jpg
 315 | /JPEGImages/2008_002635.jpg
 316 | /JPEGImages/2008_002655.jpg
 317 | /JPEGImages/2008_002695.jpg
 318 | /JPEGImages/2008_002702.jpg
 319 | /JPEGImages/2008_002706.jpg
 320 | /JPEGImages/2008_002707.jpg
 321 | /JPEGImages/2008_002722.jpg
 322 | /JPEGImages/2008_002745.jpg
 323 | /JPEGImages/2008_002757.jpg
 324 | /JPEGImages/2008_002779.jpg
 325 | /JPEGImages/2008_002805.jpg
 326 | /JPEGImages/2008_002871.jpg
 327 | /JPEGImages/2008_002895.jpg
 328 | /JPEGImages/2008_002905.jpg
 329 | /JPEGImages/2008_002923.jpg
 330 | /JPEGImages/2008_002927.jpg
 331 | /JPEGImages/2008_002939.jpg
 332 | /JPEGImages/2008_002941.jpg
 333 | /JPEGImages/2008_002962.jpg
 334 | /JPEGImages/2008_002975.jpg
 335 | /JPEGImages/2008_003000.jpg
 336 | /JPEGImages/2008_003031.jpg
 337 | /JPEGImages/2008_003038.jpg
 338 | /JPEGImages/2008_003042.jpg
 339 | /JPEGImages/2008_003069.jpg
 340 | /JPEGImages/2008_003070.jpg
 341 | /JPEGImages/2008_003115.jpg
 342 | /JPEGImages/2008_003116.jpg
 343 | /JPEGImages/2008_003130.jpg
 344 | /JPEGImages/2008_003137.jpg
 345 | /JPEGImages/2008_003138.jpg
 346 | /JPEGImages/2008_003139.jpg
 347 | /JPEGImages/2008_003165.jpg
 348 | /JPEGImages/2008_003171.jpg
 349 | /JPEGImages/2008_003176.jpg
 350 | /JPEGImages/2008_003192.jpg
 351 | /JPEGImages/2008_003194.jpg
 352 | /JPEGImages/2008_003195.jpg
 353 | /JPEGImages/2008_003198.jpg
 354 | /JPEGImages/2008_003227.jpg
 355 | /JPEGImages/2008_003247.jpg
 356 | /JPEGImages/2008_003262.jpg
 357 | /JPEGImages/2008_003298.jpg
 358 | /JPEGImages/2008_003299.jpg
 359 | /JPEGImages/2008_003307.jpg
 360 | /JPEGImages/2008_003337.jpg
 361 | /JPEGImages/2008_003353.jpg
 362 | /JPEGImages/2008_003355.jpg
 363 | /JPEGImages/2008_003363.jpg
 364 | /JPEGImages/2008_003383.jpg
 365 | /JPEGImages/2008_003389.jpg
 366 | /JPEGImages/2008_003392.jpg
 367 | /JPEGImages/2008_003399.jpg
 368 | /JPEGImages/2008_003436.jpg
 369 | /JPEGImages/2008_003457.jpg
 370 | /JPEGImages/2008_003465.jpg
 371 | /JPEGImages/2008_003481.jpg
 372 | /JPEGImages/2008_003539.jpg
 373 | /JPEGImages/2008_003548.jpg
 374 | /JPEGImages/2008_003550.jpg
 375 | /JPEGImages/2008_003567.jpg
 376 | /JPEGImages/2008_003568.jpg
 377 | /JPEGImages/2008_003606.jpg
 378 | /JPEGImages/2008_003615.jpg
 379 | /JPEGImages/2008_003654.jpg
 380 | /JPEGImages/2008_003670.jpg
 381 | /JPEGImages/2008_003700.jpg
 382 | /JPEGImages/2008_003705.jpg
 383 | /JPEGImages/2008_003727.jpg
 384 | /JPEGImages/2008_003731.jpg
 385 | /JPEGImages/2008_003734.jpg
 386 | /JPEGImages/2008_003760.jpg
 387 | /JPEGImages/2008_003804.jpg
 388 | /JPEGImages/2008_003807.jpg
 389 | /JPEGImages/2008_003810.jpg
 390 | /JPEGImages/2008_003822.jpg
 391 | /JPEGImages/2008_003833.jpg
 392 | /JPEGImages/2008_003877.jpg
 393 | /JPEGImages/2008_003879.jpg
 394 | /JPEGImages/2008_003895.jpg
 395 | /JPEGImages/2008_003901.jpg
 396 | /JPEGImages/2008_003903.jpg
 397 | /JPEGImages/2008_003911.jpg
 398 | /JPEGImages/2008_003919.jpg
 399 | /JPEGImages/2008_003927.jpg
 400 | /JPEGImages/2008_003937.jpg
 401 | /JPEGImages/2008_003946.jpg
 402 | /JPEGImages/2008_003950.jpg
 403 | /JPEGImages/2008_003955.jpg
 404 | /JPEGImages/2008_003981.jpg
 405 | /JPEGImages/2008_003991.jpg
 406 | /JPEGImages/2008_004009.jpg
 407 | /JPEGImages/2008_004039.jpg
 408 | /JPEGImages/2008_004052.jpg
 409 | /JPEGImages/2008_004063.jpg
 410 | /JPEGImages/2008_004070.jpg
 411 | /JPEGImages/2008_004078.jpg
 412 | /JPEGImages/2008_004104.jpg
 413 | /JPEGImages/2008_004139.jpg
 414 | /JPEGImages/2008_004177.jpg
 415 | /JPEGImages/2008_004181.jpg
 416 | /JPEGImages/2008_004200.jpg
 417 | /JPEGImages/2008_004219.jpg
 418 | /JPEGImages/2008_004236.jpg
 419 | /JPEGImages/2008_004250.jpg
 420 | /JPEGImages/2008_004266.jpg
 421 | /JPEGImages/2008_004299.jpg
 422 | /JPEGImages/2008_004320.jpg
 423 | /JPEGImages/2008_004334.jpg
 424 | /JPEGImages/2008_004343.jpg
 425 | /JPEGImages/2008_004349.jpg
 426 | /JPEGImages/2008_004366.jpg
 427 | /JPEGImages/2008_004386.jpg
 428 | /JPEGImages/2008_004401.jpg
 429 | /JPEGImages/2008_004423.jpg
 430 | /JPEGImages/2008_004448.jpg
 431 | /JPEGImages/2008_004481.jpg
 432 | /JPEGImages/2008_004516.jpg
 433 | /JPEGImages/2008_004536.jpg
 434 | /JPEGImages/2008_004582.jpg
 435 | /JPEGImages/2008_004609.jpg
 436 | /JPEGImages/2008_004638.jpg
 437 | /JPEGImages/2008_004642.jpg
 438 | /JPEGImages/2008_004644.jpg
 439 | /JPEGImages/2008_004669.jpg
 440 | /JPEGImages/2008_004673.jpg
 441 | /JPEGImages/2008_004691.jpg
 442 | /JPEGImages/2008_004693.jpg
 443 | /JPEGImages/2008_004709.jpg
 444 | /JPEGImages/2008_004715.jpg
 445 | /JPEGImages/2008_004757.jpg
 446 | /JPEGImages/2008_004775.jpg
 447 | /JPEGImages/2008_004782.jpg
 448 | /JPEGImages/2008_004785.jpg
 449 | /JPEGImages/2008_004798.jpg
 450 | /JPEGImages/2008_004848.jpg
 451 | /JPEGImages/2008_004861.jpg
 452 | /JPEGImages/2008_004870.jpg
 453 | /JPEGImages/2008_004877.jpg
 454 | /JPEGImages/2008_004884.jpg
 455 | /JPEGImages/2008_004891.jpg
 456 | /JPEGImages/2008_004901.jpg
 457 | /JPEGImages/2008_004919.jpg
 458 | /JPEGImages/2008_005058.jpg
 459 | /JPEGImages/2008_005069.jpg
 460 | /JPEGImages/2008_005086.jpg
 461 | /JPEGImages/2008_005087.jpg
 462 | /JPEGImages/2008_005112.jpg
 463 | /JPEGImages/2008_005113.jpg
 464 | /JPEGImages/2008_005118.jpg
 465 | /JPEGImages/2008_005128.jpg
 466 | /JPEGImages/2008_005129.jpg
 467 | /JPEGImages/2008_005153.jpg
 468 | /JPEGImages/2008_005161.jpg
 469 | /JPEGImages/2008_005162.jpg
 470 | /JPEGImages/2008_005165.jpg
 471 | /JPEGImages/2008_005187.jpg
 472 | /JPEGImages/2008_005227.jpg
 473 | /JPEGImages/2008_005308.jpg
 474 | /JPEGImages/2008_005318.jpg
 475 | /JPEGImages/2008_005320.jpg
 476 | /JPEGImages/2008_005351.jpg
 477 | /JPEGImages/2008_005372.jpg
 478 | /JPEGImages/2008_005383.jpg
 479 | /JPEGImages/2008_005391.jpg
 480 | /JPEGImages/2008_005407.jpg
 481 | /JPEGImages/2008_005420.jpg
 482 | /JPEGImages/2008_005440.jpg
 483 | /JPEGImages/2008_005487.jpg
 484 | /JPEGImages/2008_005493.jpg
 485 | /JPEGImages/2008_005520.jpg
 486 | /JPEGImages/2008_005551.jpg
 487 | /JPEGImages/2008_005556.jpg
 488 | /JPEGImages/2008_005576.jpg
 489 | /JPEGImages/2008_005578.jpg
 490 | /JPEGImages/2008_005594.jpg
 491 | /JPEGImages/2008_005619.jpg
 492 | /JPEGImages/2008_005629.jpg
 493 | /JPEGImages/2008_005644.jpg
 494 | /JPEGImages/2008_005645.jpg
 495 | /JPEGImages/2008_005651.jpg
 496 | /JPEGImages/2008_005661.jpg
 497 | /JPEGImages/2008_005662.jpg
 498 | /JPEGImages/2008_005667.jpg
 499 | /JPEGImages/2008_005694.jpg
 500 | /JPEGImages/2008_005697.jpg
 501 | /JPEGImages/2008_005709.jpg
 502 | /JPEGImages/2008_005710.jpg
 503 | /JPEGImages/2008_005733.jpg
 504 | /JPEGImages/2008_005749.jpg
 505 | /JPEGImages/2008_005753.jpg
 506 | /JPEGImages/2008_005771.jpg
 507 | /JPEGImages/2008_005781.jpg
 508 | /JPEGImages/2008_005793.jpg
 509 | /JPEGImages/2008_005802.jpg
 510 | /JPEGImages/2008_005833.jpg
 511 | /JPEGImages/2008_005844.jpg
 512 | /JPEGImages/2008_005908.jpg
 513 | /JPEGImages/2008_005931.jpg
 514 | /JPEGImages/2008_005952.jpg
 515 | /JPEGImages/2008_006016.jpg
 516 | /JPEGImages/2008_006030.jpg
 517 | /JPEGImages/2008_006033.jpg
 518 | /JPEGImages/2008_006054.jpg
 519 | /JPEGImages/2008_006073.jpg
 520 | /JPEGImages/2008_006091.jpg
 521 | /JPEGImages/2008_006142.jpg
 522 | /JPEGImages/2008_006150.jpg
 523 | /JPEGImages/2008_006206.jpg
 524 | /JPEGImages/2008_006217.jpg
 525 | /JPEGImages/2008_006264.jpg
 526 | /JPEGImages/2008_006283.jpg
 527 | /JPEGImages/2008_006308.jpg
 528 | /JPEGImages/2008_006313.jpg
 529 | /JPEGImages/2008_006333.jpg
 530 | /JPEGImages/2008_006343.jpg
 531 | /JPEGImages/2008_006381.jpg
 532 | /JPEGImages/2008_006391.jpg
 533 | /JPEGImages/2008_006423.jpg
 534 | /JPEGImages/2008_006428.jpg
 535 | /JPEGImages/2008_006440.jpg
 536 | /JPEGImages/2008_006444.jpg
 537 | /JPEGImages/2008_006473.jpg
 538 | /JPEGImages/2008_006505.jpg
 539 | /JPEGImages/2008_006531.jpg
 540 | /JPEGImages/2008_006560.jpg
 541 | /JPEGImages/2008_006571.jpg
 542 | /JPEGImages/2008_006582.jpg
 543 | /JPEGImages/2008_006594.jpg
 544 | /JPEGImages/2008_006601.jpg
 545 | /JPEGImages/2008_006633.jpg
 546 | /JPEGImages/2008_006653.jpg
 547 | /JPEGImages/2008_006678.jpg
 548 | /JPEGImages/2008_006755.jpg
 549 | /JPEGImages/2008_006772.jpg
 550 | /JPEGImages/2008_006788.jpg
 551 | /JPEGImages/2008_006799.jpg
 552 | /JPEGImages/2008_006809.jpg
 553 | /JPEGImages/2008_006838.jpg
 554 | /JPEGImages/2008_006845.jpg
 555 | /JPEGImages/2008_006852.jpg
 556 | /JPEGImages/2008_006894.jpg
 557 | /JPEGImages/2008_006905.jpg
 558 | /JPEGImages/2008_006947.jpg
 559 | /JPEGImages/2008_006983.jpg
 560 | /JPEGImages/2008_007049.jpg
 561 | /JPEGImages/2008_007065.jpg
 562 | /JPEGImages/2008_007068.jpg
 563 | /JPEGImages/2008_007111.jpg
 564 | /JPEGImages/2008_007148.jpg
 565 | /JPEGImages/2008_007159.jpg
 566 | /JPEGImages/2008_007193.jpg
 567 | /JPEGImages/2008_007228.jpg
 568 | /JPEGImages/2008_007235.jpg
 569 | /JPEGImages/2008_007249.jpg
 570 | /JPEGImages/2008_007255.jpg
 571 | /JPEGImages/2008_007268.jpg
 572 | /JPEGImages/2008_007275.jpg
 573 | /JPEGImages/2008_007292.jpg
 574 | /JPEGImages/2008_007299.jpg
 575 | /JPEGImages/2008_007306.jpg
 576 | /JPEGImages/2008_007316.jpg
 577 | /JPEGImages/2008_007400.jpg
 578 | /JPEGImages/2008_007401.jpg
 579 | /JPEGImages/2008_007419.jpg
 580 | /JPEGImages/2008_007437.jpg
 581 | /JPEGImages/2008_007483.jpg
 582 | /JPEGImages/2008_007487.jpg
 583 | /JPEGImages/2008_007520.jpg
 584 | /JPEGImages/2008_007551.jpg
 585 | /JPEGImages/2008_007603.jpg
 586 | /JPEGImages/2008_007616.jpg
 587 | /JPEGImages/2008_007654.jpg
 588 | /JPEGImages/2008_007663.jpg
 589 | /JPEGImages/2008_007708.jpg
 590 | /JPEGImages/2008_007795.jpg
 591 | /JPEGImages/2008_007801.jpg
 592 | /JPEGImages/2008_007859.jpg
 593 | /JPEGImages/2008_007903.jpg
 594 | /JPEGImages/2008_007920.jpg
 595 | /JPEGImages/2008_007926.jpg
 596 | /JPEGImages/2008_008014.jpg
 597 | /JPEGImages/2008_008017.jpg
 598 | /JPEGImages/2008_008060.jpg
 599 | /JPEGImages/2008_008077.jpg
 600 | /JPEGImages/2008_008107.jpg
 601 | /JPEGImages/2008_008108.jpg
 602 | /JPEGImages/2008_008119.jpg
 603 | /JPEGImages/2008_008126.jpg
 604 | /JPEGImages/2008_008133.jpg
 605 | /JPEGImages/2008_008144.jpg
 606 | /JPEGImages/2008_008216.jpg
 607 | /JPEGImages/2008_008244.jpg
 608 | /JPEGImages/2008_008248.jpg
 609 | /JPEGImages/2008_008250.jpg
 610 | /JPEGImages/2008_008260.jpg
 611 | /JPEGImages/2008_008277.jpg
 612 | /JPEGImages/2008_008280.jpg
 613 | /JPEGImages/2008_008290.jpg
 614 | /JPEGImages/2008_008304.jpg
 615 | /JPEGImages/2008_008340.jpg
 616 | /JPEGImages/2008_008371.jpg
 617 | /JPEGImages/2008_008390.jpg
 618 | /JPEGImages/2008_008397.jpg
 619 | /JPEGImages/2008_008409.jpg
 620 | /JPEGImages/2008_008412.jpg
 621 | /JPEGImages/2008_008419.jpg
 622 | /JPEGImages/2008_008454.jpg
 623 | /JPEGImages/2008_008491.jpg
 624 | /JPEGImages/2008_008498.jpg
 625 | /JPEGImages/2008_008565.jpg
 626 | /JPEGImages/2008_008599.jpg
 627 | /JPEGImages/2008_008603.jpg
 628 | /JPEGImages/2008_008631.jpg
 629 | /JPEGImages/2008_008634.jpg
 630 | /JPEGImages/2008_008640.jpg
 631 | /JPEGImages/2008_008646.jpg
 632 | /JPEGImages/2008_008660.jpg
 633 | /JPEGImages/2008_008663.jpg
 634 | /JPEGImages/2008_008664.jpg
 635 | /JPEGImages/2008_008709.jpg
 636 | /JPEGImages/2008_008720.jpg
 637 | /JPEGImages/2008_008747.jpg
 638 | /JPEGImages/2008_008768.jpg
 639 | /JPEGImages/2009_000004.jpg
 640 | /JPEGImages/2009_000019.jpg
 641 | /JPEGImages/2009_000024.jpg
 642 | /JPEGImages/2009_000025.jpg
 643 | /JPEGImages/2009_000053.jpg
 644 | /JPEGImages/2009_000076.jpg
 645 | /JPEGImages/2009_000107.jpg
 646 | /JPEGImages/2009_000110.jpg
 647 | /JPEGImages/2009_000115.jpg
 648 | /JPEGImages/2009_000117.jpg
 649 | /JPEGImages/2009_000175.jpg
 650 | /JPEGImages/2009_000220.jpg
 651 | /JPEGImages/2009_000259.jpg
 652 | /JPEGImages/2009_000275.jpg
 653 | /JPEGImages/2009_000314.jpg
 654 | /JPEGImages/2009_000368.jpg
 655 | /JPEGImages/2009_000373.jpg
 656 | /JPEGImages/2009_000384.jpg
 657 | /JPEGImages/2009_000388.jpg
 658 | /JPEGImages/2009_000423.jpg
 659 | /JPEGImages/2009_000433.jpg
 660 | /JPEGImages/2009_000434.jpg
 661 | /JPEGImages/2009_000458.jpg
 662 | /JPEGImages/2009_000475.jpg
 663 | /JPEGImages/2009_000481.jpg
 664 | /JPEGImages/2009_000495.jpg
 665 | /JPEGImages/2009_000514.jpg
 666 | /JPEGImages/2009_000555.jpg
 667 | /JPEGImages/2009_000556.jpg
 668 | /JPEGImages/2009_000561.jpg
 669 | /JPEGImages/2009_000571.jpg
 670 | /JPEGImages/2009_000581.jpg
 671 | /JPEGImages/2009_000605.jpg
 672 | /JPEGImages/2009_000609.jpg
 673 | /JPEGImages/2009_000644.jpg
 674 | /JPEGImages/2009_000654.jpg
 675 | /JPEGImages/2009_000671.jpg
 676 | /JPEGImages/2009_000733.jpg
 677 | /JPEGImages/2009_000740.jpg
 678 | /JPEGImages/2009_000766.jpg
 679 | /JPEGImages/2009_000775.jpg
 680 | /JPEGImages/2009_000776.jpg
 681 | /JPEGImages/2009_000795.jpg
 682 | /JPEGImages/2009_000850.jpg
 683 | /JPEGImages/2009_000881.jpg
 684 | /JPEGImages/2009_000900.jpg
 685 | /JPEGImages/2009_000914.jpg
 686 | /JPEGImages/2009_000941.jpg
 687 | /JPEGImages/2009_000977.jpg
 688 | /JPEGImages/2009_000984.jpg
 689 | /JPEGImages/2009_000986.jpg
 690 | /JPEGImages/2009_001005.jpg
 691 | /JPEGImages/2009_001015.jpg
 692 | /JPEGImages/2009_001058.jpg
 693 | /JPEGImages/2009_001072.jpg
 694 | /JPEGImages/2009_001087.jpg
 695 | /JPEGImages/2009_001092.jpg
 696 | /JPEGImages/2009_001109.jpg
 697 | /JPEGImages/2009_001114.jpg
 698 | /JPEGImages/2009_001115.jpg
 699 | /JPEGImages/2009_001141.jpg
 700 | /JPEGImages/2009_001174.jpg
 701 | /JPEGImages/2009_001175.jpg
 702 | /JPEGImages/2009_001182.jpg
 703 | /JPEGImages/2009_001222.jpg
 704 | /JPEGImages/2009_001228.jpg
 705 | /JPEGImages/2009_001246.jpg
 706 | /JPEGImages/2009_001262.jpg
 707 | /JPEGImages/2009_001274.jpg
 708 | /JPEGImages/2009_001284.jpg
 709 | /JPEGImages/2009_001297.jpg
 710 | /JPEGImages/2009_001331.jpg
 711 | /JPEGImages/2009_001336.jpg
 712 | /JPEGImages/2009_001337.jpg
 713 | /JPEGImages/2009_001379.jpg
 714 | /JPEGImages/2009_001392.jpg
 715 | /JPEGImages/2009_001451.jpg
 716 | /JPEGImages/2009_001485.jpg
 717 | /JPEGImages/2009_001488.jpg
 718 | /JPEGImages/2009_001497.jpg
 719 | /JPEGImages/2009_001504.jpg
 720 | /JPEGImages/2009_001506.jpg
 721 | /JPEGImages/2009_001573.jpg
 722 | /JPEGImages/2009_001576.jpg
 723 | /JPEGImages/2009_001603.jpg
 724 | /JPEGImages/2009_001613.jpg
 725 | /JPEGImages/2009_001652.jpg
 726 | /JPEGImages/2009_001661.jpg
 727 | /JPEGImages/2009_001668.jpg
 728 | /JPEGImages/2009_001680.jpg
 729 | /JPEGImages/2009_001688.jpg
 730 | /JPEGImages/2009_001697.jpg
 731 | /JPEGImages/2009_001729.jpg
 732 | /JPEGImages/2009_001771.jpg
 733 | /JPEGImages/2009_001785.jpg
 734 | /JPEGImages/2009_001793.jpg
 735 | /JPEGImages/2009_001814.jpg
 736 | /JPEGImages/2009_001866.jpg
 737 | /JPEGImages/2009_001872.jpg
 738 | /JPEGImages/2009_001880.jpg
 739 | /JPEGImages/2009_001883.jpg
 740 | /JPEGImages/2009_001891.jpg
 741 | /JPEGImages/2009_001913.jpg
 742 | /JPEGImages/2009_001938.jpg
 743 | /JPEGImages/2009_001946.jpg
 744 | /JPEGImages/2009_001953.jpg
 745 | /JPEGImages/2009_001969.jpg
 746 | /JPEGImages/2009_001978.jpg
 747 | /JPEGImages/2009_001995.jpg
 748 | /JPEGImages/2009_002007.jpg
 749 | /JPEGImages/2009_002036.jpg
 750 | /JPEGImages/2009_002041.jpg
 751 | /JPEGImages/2009_002049.jpg
 752 | /JPEGImages/2009_002051.jpg
 753 | /JPEGImages/2009_002062.jpg
 754 | /JPEGImages/2009_002063.jpg
 755 | /JPEGImages/2009_002067.jpg
 756 | /JPEGImages/2009_002085.jpg
 757 | /JPEGImages/2009_002092.jpg
 758 | /JPEGImages/2009_002114.jpg
 759 | /JPEGImages/2009_002115.jpg
 760 | /JPEGImages/2009_002142.jpg
 761 | /JPEGImages/2009_002148.jpg
 762 | /JPEGImages/2009_002157.jpg
 763 | /JPEGImages/2009_002181.jpg
 764 | /JPEGImages/2009_002220.jpg
 765 | /JPEGImages/2009_002284.jpg
 766 | /JPEGImages/2009_002287.jpg
 767 | /JPEGImages/2009_002300.jpg
 768 | /JPEGImages/2009_002310.jpg
 769 | /JPEGImages/2009_002315.jpg
 770 | /JPEGImages/2009_002334.jpg
 771 | /JPEGImages/2009_002337.jpg
 772 | /JPEGImages/2009_002354.jpg
 773 | /JPEGImages/2009_002357.jpg
 774 | /JPEGImages/2009_002411.jpg
 775 | /JPEGImages/2009_002426.jpg
 776 | /JPEGImages/2009_002458.jpg
 777 | /JPEGImages/2009_002459.jpg
 778 | /JPEGImages/2009_002461.jpg
 779 | /JPEGImages/2009_002466.jpg
 780 | /JPEGImages/2009_002481.jpg
 781 | /JPEGImages/2009_002483.jpg
 782 | /JPEGImages/2009_002503.jpg
 783 | /JPEGImages/2009_002581.jpg
 784 | /JPEGImages/2009_002583.jpg
 785 | /JPEGImages/2009_002589.jpg
 786 | /JPEGImages/2009_002600.jpg
 787 | /JPEGImages/2009_002601.jpg
 788 | /JPEGImages/2009_002602.jpg
 789 | /JPEGImages/2009_002641.jpg
 790 | /JPEGImages/2009_002646.jpg
 791 | /JPEGImages/2009_002656.jpg
 792 | /JPEGImages/2009_002666.jpg
 793 | /JPEGImages/2009_002720.jpg
 794 | /JPEGImages/2009_002767.jpg
 795 | /JPEGImages/2009_002768.jpg
 796 | /JPEGImages/2009_002794.jpg
 797 | /JPEGImages/2009_002821.jpg
 798 | /JPEGImages/2009_002825.jpg
 799 | /JPEGImages/2009_002839.jpg
 800 | /JPEGImages/2009_002840.jpg
 801 | /JPEGImages/2009_002859.jpg
 802 | /JPEGImages/2009_002860.jpg
 803 | /JPEGImages/2009_002881.jpg
 804 | /JPEGImages/2009_002889.jpg
 805 | /JPEGImages/2009_002892.jpg
 806 | /JPEGImages/2009_002895.jpg
 807 | /JPEGImages/2009_002896.jpg
 808 | /JPEGImages/2009_002900.jpg
 809 | /JPEGImages/2009_002924.jpg
 810 | /JPEGImages/2009_002966.jpg
 811 | /JPEGImages/2009_002973.jpg
 812 | /JPEGImages/2009_002981.jpg
 813 | /JPEGImages/2009_003004.jpg
 814 | /JPEGImages/2009_003021.jpg
 815 | /JPEGImages/2009_003028.jpg
 816 | /JPEGImages/2009_003037.jpg
 817 | /JPEGImages/2009_003038.jpg
 818 | /JPEGImages/2009_003055.jpg
 819 | /JPEGImages/2009_003085.jpg
 820 | /JPEGImages/2009_003100.jpg
 821 | /JPEGImages/2009_003106.jpg
 822 | /JPEGImages/2009_003117.jpg
 823 | /JPEGImages/2009_003139.jpg
 824 | /JPEGImages/2009_003170.jpg
 825 | /JPEGImages/2009_003179.jpg
 826 | /JPEGImages/2009_003184.jpg
 827 | /JPEGImages/2009_003186.jpg
 828 | /JPEGImages/2009_003190.jpg
 829 | /JPEGImages/2009_003221.jpg
 830 | /JPEGImages/2009_003236.jpg
 831 | /JPEGImages/2009_003242.jpg
 832 | /JPEGImages/2009_003244.jpg
 833 | /JPEGImages/2009_003260.jpg
 834 | /JPEGImages/2009_003264.jpg
 835 | /JPEGImages/2009_003274.jpg
 836 | /JPEGImages/2009_003283.jpg
 837 | /JPEGImages/2009_003296.jpg
 838 | /JPEGImages/2009_003332.jpg
 839 | /JPEGImages/2009_003341.jpg
 840 | /JPEGImages/2009_003354.jpg
 841 | /JPEGImages/2009_003370.jpg
 842 | /JPEGImages/2009_003371.jpg
 843 | /JPEGImages/2009_003374.jpg
 844 | /JPEGImages/2009_003391.jpg
 845 | /JPEGImages/2009_003393.jpg
 846 | /JPEGImages/2009_003404.jpg
 847 | /JPEGImages/2009_003405.jpg
 848 | /JPEGImages/2009_003414.jpg
 849 | /JPEGImages/2009_003428.jpg
 850 | /JPEGImages/2009_003470.jpg
 851 | /JPEGImages/2009_003474.jpg
 852 | /JPEGImages/2009_003532.jpg
 853 | /JPEGImages/2009_003536.jpg
 854 | /JPEGImages/2009_003578.jpg
 855 | /JPEGImages/2009_003580.jpg
 856 | /JPEGImages/2009_003620.jpg
 857 | /JPEGImages/2009_003621.jpg
 858 | /JPEGImages/2009_003680.jpg
 859 | /JPEGImages/2009_003699.jpg
 860 | /JPEGImages/2009_003727.jpg
 861 | /JPEGImages/2009_003737.jpg
 862 | /JPEGImages/2009_003780.jpg
 863 | /JPEGImages/2009_003811.jpg
 864 | /JPEGImages/2009_003824.jpg
 865 | /JPEGImages/2009_003831.jpg
 866 | /JPEGImages/2009_003844.jpg
 867 | /JPEGImages/2009_003850.jpg
 868 | /JPEGImages/2009_003851.jpg
 869 | /JPEGImages/2009_003864.jpg
 870 | /JPEGImages/2009_003868.jpg
 871 | /JPEGImages/2009_003869.jpg
 872 | /JPEGImages/2009_003893.jpg
 873 | /JPEGImages/2009_003909.jpg
 874 | /JPEGImages/2009_003924.jpg
 875 | /JPEGImages/2009_003925.jpg
 876 | /JPEGImages/2009_003960.jpg
 877 | /JPEGImages/2009_003979.jpg
 878 | /JPEGImages/2009_003990.jpg
 879 | /JPEGImages/2009_003997.jpg
 880 | /JPEGImages/2009_004006.jpg
 881 | /JPEGImages/2009_004010.jpg
 882 | /JPEGImages/2009_004066.jpg
 883 | /JPEGImages/2009_004077.jpg
 884 | /JPEGImages/2009_004081.jpg
 885 | /JPEGImages/2009_004097.jpg
 886 | /JPEGImages/2009_004098.jpg
 887 | /JPEGImages/2009_004136.jpg
 888 | /JPEGImages/2009_004216.jpg
 889 | /JPEGImages/2009_004220.jpg
 890 | /JPEGImages/2009_004266.jpg
 891 | /JPEGImages/2009_004269.jpg
 892 | /JPEGImages/2009_004286.jpg
 893 | /JPEGImages/2009_004296.jpg
 894 | /JPEGImages/2009_004321.jpg
 895 | /JPEGImages/2009_004342.jpg
 896 | /JPEGImages/2009_004343.jpg
 897 | /JPEGImages/2009_004344.jpg
 898 | /JPEGImages/2009_004385.jpg
 899 | /JPEGImages/2009_004408.jpg
 900 | /JPEGImages/2009_004420.jpg
 901 | /JPEGImages/2009_004441.jpg
 902 | /JPEGImages/2009_004447.jpg
 903 | /JPEGImages/2009_004461.jpg
 904 | /JPEGImages/2009_004467.jpg
 905 | /JPEGImages/2009_004485.jpg
 906 | /JPEGImages/2009_004488.jpg
 907 | /JPEGImages/2009_004516.jpg
 908 | /JPEGImages/2009_004521.jpg
 909 | /JPEGImages/2009_004544.jpg
 910 | /JPEGImages/2009_004596.jpg
 911 | /JPEGImages/2009_004613.jpg
 912 | /JPEGImages/2009_004615.jpg
 913 | /JPEGImages/2009_004618.jpg
 914 | /JPEGImages/2009_004621.jpg
 915 | /JPEGImages/2009_004646.jpg
 916 | /JPEGImages/2009_004659.jpg
 917 | /JPEGImages/2009_004663.jpg
 918 | /JPEGImages/2009_004666.jpg
 919 | /JPEGImages/2009_004691.jpg
 920 | /JPEGImages/2009_004715.jpg
 921 | /JPEGImages/2009_004726.jpg
 922 | /JPEGImages/2009_004753.jpg
 923 | /JPEGImages/2009_004776.jpg
 924 | /JPEGImages/2009_004811.jpg
 925 | /JPEGImages/2009_004814.jpg
 926 | /JPEGImages/2009_004818.jpg
 927 | /JPEGImages/2009_004835.jpg
 928 | /JPEGImages/2009_004863.jpg
 929 | /JPEGImages/2009_004894.jpg
 930 | /JPEGImages/2009_004909.jpg
 931 | /JPEGImages/2009_004928.jpg
 932 | /JPEGImages/2009_004937.jpg
 933 | /JPEGImages/2009_004954.jpg
 934 | /JPEGImages/2009_004966.jpg
 935 | /JPEGImages/2009_004970.jpg
 936 | /JPEGImages/2009_004976.jpg
 937 | /JPEGImages/2009_005004.jpg
 938 | /JPEGImages/2009_005011.jpg
 939 | /JPEGImages/2009_005053.jpg
 940 | /JPEGImages/2009_005072.jpg
 941 | /JPEGImages/2009_005115.jpg
 942 | /JPEGImages/2009_005146.jpg
 943 | /JPEGImages/2009_005151.jpg
 944 | /JPEGImages/2009_005164.jpg
 945 | /JPEGImages/2009_005179.jpg
 946 | /JPEGImages/2009_005224.jpg
 947 | /JPEGImages/2009_005243.jpg
 948 | /JPEGImages/2009_005249.jpg
 949 | /JPEGImages/2009_005252.jpg
 950 | /JPEGImages/2009_005254.jpg
 951 | /JPEGImages/2009_005258.jpg
 952 | /JPEGImages/2009_005264.jpg
 953 | /JPEGImages/2009_005266.jpg
 954 | /JPEGImages/2009_005276.jpg
 955 | /JPEGImages/2009_005290.jpg
 956 | /JPEGImages/2009_005295.jpg
 957 | /JPEGImages/2010_000004.jpg
 958 | /JPEGImages/2010_000005.jpg
 959 | /JPEGImages/2010_000006.jpg
 960 | /JPEGImages/2010_000032.jpg
 961 | /JPEGImages/2010_000062.jpg
 962 | /JPEGImages/2010_000093.jpg
 963 | /JPEGImages/2010_000094.jpg
 964 | /JPEGImages/2010_000161.jpg
 965 | /JPEGImages/2010_000176.jpg
 966 | /JPEGImages/2010_000223.jpg
 967 | /JPEGImages/2010_000226.jpg
 968 | /JPEGImages/2010_000236.jpg
 969 | /JPEGImages/2010_000239.jpg
 970 | /JPEGImages/2010_000287.jpg
 971 | /JPEGImages/2010_000300.jpg
 972 | /JPEGImages/2010_000301.jpg
 973 | /JPEGImages/2010_000328.jpg
 974 | /JPEGImages/2010_000378.jpg
 975 | /JPEGImages/2010_000405.jpg
 976 | /JPEGImages/2010_000407.jpg
 977 | /JPEGImages/2010_000472.jpg
 978 | /JPEGImages/2010_000479.jpg
 979 | /JPEGImages/2010_000491.jpg
 980 | /JPEGImages/2010_000533.jpg
 981 | /JPEGImages/2010_000535.jpg
 982 | /JPEGImages/2010_000542.jpg
 983 | /JPEGImages/2010_000554.jpg
 984 | /JPEGImages/2010_000580.jpg
 985 | /JPEGImages/2010_000594.jpg
 986 | /JPEGImages/2010_000596.jpg
 987 | /JPEGImages/2010_000599.jpg
 988 | /JPEGImages/2010_000606.jpg
 989 | /JPEGImages/2010_000615.jpg
 990 | /JPEGImages/2010_000654.jpg
 991 | /JPEGImages/2010_000659.jpg
 992 | /JPEGImages/2010_000693.jpg
 993 | /JPEGImages/2010_000698.jpg
 994 | /JPEGImages/2010_000730.jpg
 995 | /JPEGImages/2010_000734.jpg
 996 | /JPEGImages/2010_000741.jpg
 997 | /JPEGImages/2010_000755.jpg
 998 | /JPEGImages/2010_000768.jpg
 999 | /JPEGImages/2010_000794.jpg
1000 | /JPEGImages/2010_000813.jpg
1001 | /JPEGImages/2010_000817.jpg
1002 | /JPEGImages/2010_000834.jpg
1003 | /JPEGImages/2010_000839.jpg
1004 | /JPEGImages/2010_000848.jpg
1005 | /JPEGImages/2010_000881.jpg
1006 | /JPEGImages/2010_000888.jpg
1007 | /JPEGImages/2010_000900.jpg
1008 | /JPEGImages/2010_000903.jpg
1009 | /JPEGImages/2010_000924.jpg
1010 | /JPEGImages/2010_000946.jpg
1011 | /JPEGImages/2010_000953.jpg
1012 | /JPEGImages/2010_000957.jpg
1013 | /JPEGImages/2010_000967.jpg
1014 | /JPEGImages/2010_000992.jpg
1015 | /JPEGImages/2010_000998.jpg
1016 | /JPEGImages/2010_001053.jpg
1017 | /JPEGImages/2010_001067.jpg
1018 | /JPEGImages/2010_001114.jpg
1019 | /JPEGImages/2010_001132.jpg
1020 | /JPEGImages/2010_001138.jpg
1021 | /JPEGImages/2010_001169.jpg
1022 | /JPEGImages/2010_001171.jpg
1023 | /JPEGImages/2010_001228.jpg
1024 | /JPEGImages/2010_001260.jpg
1025 | /JPEGImages/2010_001268.jpg
1026 | /JPEGImages/2010_001280.jpg
1027 | /JPEGImages/2010_001298.jpg
1028 | /JPEGImages/2010_001302.jpg
1029 | /JPEGImages/2010_001308.jpg
1030 | /JPEGImages/2010_001324.jpg
1031 | /JPEGImages/2010_001332.jpg
1032 | /JPEGImages/2010_001335.jpg
1033 | /JPEGImages/2010_001345.jpg
1034 | /JPEGImages/2010_001346.jpg
1035 | /JPEGImages/2010_001349.jpg
1036 | /JPEGImages/2010_001373.jpg
1037 | /JPEGImages/2010_001381.jpg
1038 | /JPEGImages/2010_001392.jpg
1039 | /JPEGImages/2010_001396.jpg
1040 | /JPEGImages/2010_001420.jpg
1041 | /JPEGImages/2010_001500.jpg
1042 | /JPEGImages/2010_001506.jpg
1043 | /JPEGImages/2010_001521.jpg
1044 | /JPEGImages/2010_001532.jpg
1045 | /JPEGImages/2010_001558.jpg
1046 | /JPEGImages/2010_001598.jpg
1047 | /JPEGImages/2010_001611.jpg
1048 | /JPEGImages/2010_001631.jpg
1049 | /JPEGImages/2010_001639.jpg
1050 | /JPEGImages/2010_001651.jpg
1051 | /JPEGImages/2010_001663.jpg
1052 | /JPEGImages/2010_001664.jpg
1053 | /JPEGImages/2010_001728.jpg
1054 | /JPEGImages/2010_001778.jpg
1055 | /JPEGImages/2010_001861.jpg
1056 | /JPEGImages/2010_001874.jpg
1057 | /JPEGImages/2010_001900.jpg
1058 | /JPEGImages/2010_001905.jpg
1059 | /JPEGImages/2010_001969.jpg
1060 | /JPEGImages/2010_002008.jpg
1061 | /JPEGImages/2010_002014.jpg
1062 | /JPEGImages/2010_002049.jpg
1063 | /JPEGImages/2010_002052.jpg
1064 | /JPEGImages/2010_002091.jpg
1065 | /JPEGImages/2010_002115.jpg
1066 | /JPEGImages/2010_002119.jpg
1067 | /JPEGImages/2010_002134.jpg
1068 | /JPEGImages/2010_002156.jpg
1069 | /JPEGImages/2010_002160.jpg
1070 | /JPEGImages/2010_002186.jpg
1071 | /JPEGImages/2010_002210.jpg
1072 | /JPEGImages/2010_002241.jpg
1073 | /JPEGImages/2010_002252.jpg
1074 | /JPEGImages/2010_002258.jpg
1075 | /JPEGImages/2010_002262.jpg
1076 | /JPEGImages/2010_002273.jpg
1077 | /JPEGImages/2010_002290.jpg
1078 | /JPEGImages/2010_002292.jpg
1079 | /JPEGImages/2010_002347.jpg
1080 | /JPEGImages/2010_002358.jpg
1081 | /JPEGImages/2010_002360.jpg
1082 | /JPEGImages/2010_002367.jpg
1083 | /JPEGImages/2010_002416.jpg
1084 | /JPEGImages/2010_002451.jpg
1085 | /JPEGImages/2010_002481.jpg
1086 | /JPEGImages/2010_002490.jpg
1087 | /JPEGImages/2010_002495.jpg
1088 | /JPEGImages/2010_002588.jpg
1089 | /JPEGImages/2010_002607.jpg
1090 | /JPEGImages/2010_002609.jpg
1091 | /JPEGImages/2010_002610.jpg
1092 | /JPEGImages/2010_002641.jpg
1093 | /JPEGImages/2010_002685.jpg
1094 | /JPEGImages/2010_002699.jpg
1095 | /JPEGImages/2010_002719.jpg
1096 | /JPEGImages/2010_002735.jpg
1097 | /JPEGImages/2010_002751.jpg
1098 | /JPEGImages/2010_002804.jpg
1099 | /JPEGImages/2010_002835.jpg
1100 | /JPEGImages/2010_002852.jpg
1101 | /JPEGImages/2010_002885.jpg
1102 | /JPEGImages/2010_002889.jpg
1103 | /JPEGImages/2010_002904.jpg
1104 | /JPEGImages/2010_002908.jpg
1105 | /JPEGImages/2010_002916.jpg
1106 | /JPEGImages/2010_002974.jpg
1107 | /JPEGImages/2010_002977.jpg
1108 | /JPEGImages/2010_003005.jpg
1109 | /JPEGImages/2010_003021.jpg
1110 | /JPEGImages/2010_003030.jpg
1111 | /JPEGImages/2010_003038.jpg
1112 | /JPEGImages/2010_003046.jpg
1113 | /JPEGImages/2010_003052.jpg
1114 | /JPEGImages/2010_003089.jpg
1115 | /JPEGImages/2010_003110.jpg
1116 | /JPEGImages/2010_003118.jpg
1117 | /JPEGImages/2010_003171.jpg
1118 | /JPEGImages/2010_003217.jpg
1119 | /JPEGImages/2010_003221.jpg
1120 | /JPEGImages/2010_003228.jpg
1121 | /JPEGImages/2010_003243.jpg
1122 | /JPEGImages/2010_003271.jpg
1123 | /JPEGImages/2010_003295.jpg
1124 | /JPEGImages/2010_003306.jpg
1125 | /JPEGImages/2010_003324.jpg
1126 | /JPEGImages/2010_003363.jpg
1127 | /JPEGImages/2010_003382.jpg
1128 | /JPEGImages/2010_003388.jpg
1129 | /JPEGImages/2010_003389.jpg
1130 | /JPEGImages/2010_003392.jpg
1131 | /JPEGImages/2010_003430.jpg
1132 | /JPEGImages/2010_003442.jpg
1133 | /JPEGImages/2010_003459.jpg
1134 | /JPEGImages/2010_003485.jpg
1135 | /JPEGImages/2010_003486.jpg
1136 | /JPEGImages/2010_003500.jpg
1137 | /JPEGImages/2010_003523.jpg
1138 | /JPEGImages/2010_003542.jpg
1139 | /JPEGImages/2010_003552.jpg
1140 | /JPEGImages/2010_003570.jpg
1141 | /JPEGImages/2010_003572.jpg
1142 | /JPEGImages/2010_003586.jpg
1143 | /JPEGImages/2010_003615.jpg
1144 | /JPEGImages/2010_003623.jpg
1145 | /JPEGImages/2010_003657.jpg
1146 | /JPEGImages/2010_003666.jpg
1147 | /JPEGImages/2010_003705.jpg
1148 | /JPEGImages/2010_003710.jpg
1149 | /JPEGImages/2010_003720.jpg
1150 | /JPEGImages/2010_003733.jpg
1151 | /JPEGImages/2010_003750.jpg
1152 | /JPEGImages/2010_003767.jpg
1153 | /JPEGImages/2010_003802.jpg
1154 | /JPEGImages/2010_003809.jpg
1155 | /JPEGImages/2010_003830.jpg
1156 | /JPEGImages/2010_003832.jpg
1157 | /JPEGImages/2010_003836.jpg
1158 | /JPEGImages/2010_003838.jpg
1159 | /JPEGImages/2010_003850.jpg
1160 | /JPEGImages/2010_003867.jpg
1161 | /JPEGImages/2010_003882.jpg
1162 | /JPEGImages/2010_003909.jpg
1163 | /JPEGImages/2010_003922.jpg
1164 | /JPEGImages/2010_003923.jpg
1165 | /JPEGImages/2010_003978.jpg
1166 | /JPEGImages/2010_003989.jpg
1167 | /JPEGImages/2010_003990.jpg
1168 | /JPEGImages/2010_004000.jpg
1169 | /JPEGImages/2010_004003.jpg
1170 | /JPEGImages/2010_004068.jpg
1171 | /JPEGImages/2010_004076.jpg
1172 | /JPEGImages/2010_004117.jpg
1173 | /JPEGImages/2010_004136.jpg
1174 | /JPEGImages/2010_004142.jpg
1175 | /JPEGImages/2010_004195.jpg
1176 | /JPEGImages/2010_004200.jpg
1177 | /JPEGImages/2010_004202.jpg
1178 | /JPEGImages/2010_004232.jpg
1179 | /JPEGImages/2010_004261.jpg
1180 | /JPEGImages/2010_004266.jpg
1181 | /JPEGImages/2010_004273.jpg
1182 | /JPEGImages/2010_004305.jpg
1183 | /JPEGImages/2010_004403.jpg
1184 | /JPEGImages/2010_004433.jpg
1185 | /JPEGImages/2010_004434.jpg
1186 | /JPEGImages/2010_004435.jpg
1187 | /JPEGImages/2010_004438.jpg
1188 | /JPEGImages/2010_004442.jpg
1189 | /JPEGImages/2010_004473.jpg
1190 | /JPEGImages/2010_004482.jpg
1191 | /JPEGImages/2010_004487.jpg
1192 | /JPEGImages/2010_004489.jpg
1193 | /JPEGImages/2010_004512.jpg
1194 | /JPEGImages/2010_004525.jpg
1195 | /JPEGImages/2010_004527.jpg
1196 | /JPEGImages/2010_004532.jpg
1197 | /JPEGImages/2010_004566.jpg
1198 | /JPEGImages/2010_004568.jpg
1199 | /JPEGImages/2010_004579.jpg
1200 | /JPEGImages/2010_004611.jpg
1201 | /JPEGImages/2010_004641.jpg
1202 | /JPEGImages/2010_004688.jpg
1203 | /JPEGImages/2010_004699.jpg
1204 | /JPEGImages/2010_004702.jpg
1205 | /JPEGImages/2010_004716.jpg
1206 | /JPEGImages/2010_004754.jpg
1207 | /JPEGImages/2010_004767.jpg
1208 | /JPEGImages/2010_004776.jpg
1209 | /JPEGImages/2010_004811.jpg
1210 | /JPEGImages/2010_004837.jpg
1211 | /JPEGImages/2010_004839.jpg
1212 | /JPEGImages/2010_004845.jpg
1213 | /JPEGImages/2010_004860.jpg
1214 | /JPEGImages/2010_004867.jpg
1215 | /JPEGImages/2010_004881.jpg
1216 | /JPEGImages/2010_004939.jpg
1217 | /JPEGImages/2010_005001.jpg
1218 | /JPEGImages/2010_005047.jpg
1219 | /JPEGImages/2010_005051.jpg
1220 | /JPEGImages/2010_005091.jpg
1221 | /JPEGImages/2010_005095.jpg
1222 | /JPEGImages/2010_005125.jpg
1223 | /JPEGImages/2010_005140.jpg
1224 | /JPEGImages/2010_005177.jpg
1225 | /JPEGImages/2010_005178.jpg
1226 | /JPEGImages/2010_005194.jpg
1227 | /JPEGImages/2010_005197.jpg
1228 | /JPEGImages/2010_005200.jpg
1229 | /JPEGImages/2010_005205.jpg
1230 | /JPEGImages/2010_005212.jpg
1231 | /JPEGImages/2010_005248.jpg
1232 | /JPEGImages/2010_005294.jpg
1233 | /JPEGImages/2010_005298.jpg
1234 | /JPEGImages/2010_005313.jpg
1235 | /JPEGImages/2010_005324.jpg
1236 | /JPEGImages/2010_005328.jpg
1237 | /JPEGImages/2010_005329.jpg
1238 | /JPEGImages/2010_005380.jpg
1239 | /JPEGImages/2010_005404.jpg
1240 | /JPEGImages/2010_005407.jpg
1241 | /JPEGImages/2010_005411.jpg
1242 | /JPEGImages/2010_005423.jpg
1243 | /JPEGImages/2010_005499.jpg
1244 | /JPEGImages/2010_005509.jpg
1245 | /JPEGImages/2010_005510.jpg
1246 | /JPEGImages/2010_005544.jpg
1247 | /JPEGImages/2010_005549.jpg
1248 | /JPEGImages/2010_005590.jpg
1249 | /JPEGImages/2010_005639.jpg
1250 | /JPEGImages/2010_005699.jpg
1251 | /JPEGImages/2010_005704.jpg
1252 | /JPEGImages/2010_005707.jpg
1253 | /JPEGImages/2010_005711.jpg
1254 | /JPEGImages/2010_005726.jpg
1255 | /JPEGImages/2010_005741.jpg
1256 | /JPEGImages/2010_005765.jpg
1257 | /JPEGImages/2010_005790.jpg
1258 | /JPEGImages/2010_005792.jpg
1259 | /JPEGImages/2010_005797.jpg
1260 | /JPEGImages/2010_005812.jpg
1261 | /JPEGImages/2010_005850.jpg
1262 | /JPEGImages/2010_005861.jpg
1263 | /JPEGImages/2010_005869.jpg
1264 | /JPEGImages/2010_005908.jpg
1265 | /JPEGImages/2010_005915.jpg
1266 | /JPEGImages/2010_005946.jpg
1267 | /JPEGImages/2010_005965.jpg
1268 | /JPEGImages/2010_006044.jpg
1269 | /JPEGImages/2010_006047.jpg
1270 | /JPEGImages/2010_006052.jpg
1271 | /JPEGImages/2010_006081.jpg
1272 | /JPEGImages/2011_000001.jpg
1273 | /JPEGImages/2011_000013.jpg
1274 | /JPEGImages/2011_000014.jpg
1275 | /JPEGImages/2011_000020.jpg
1276 | /JPEGImages/2011_000032.jpg
1277 | /JPEGImages/2011_000042.jpg
1278 | /JPEGImages/2011_000063.jpg
1279 | /JPEGImages/2011_000115.jpg
1280 | /JPEGImages/2011_000120.jpg
1281 | /JPEGImages/2011_000240.jpg
1282 | /JPEGImages/2011_000244.jpg
1283 | /JPEGImages/2011_000254.jpg
1284 | /JPEGImages/2011_000261.jpg
1285 | /JPEGImages/2011_000262.jpg
1286 | /JPEGImages/2011_000271.jpg
1287 | /JPEGImages/2011_000274.jpg
1288 | /JPEGImages/2011_000306.jpg
1289 | /JPEGImages/2011_000311.jpg
1290 | /JPEGImages/2011_000316.jpg
1291 | /JPEGImages/2011_000328.jpg
1292 | /JPEGImages/2011_000351.jpg
1293 | /JPEGImages/2011_000352.jpg
1294 | /JPEGImages/2011_000406.jpg
1295 | /JPEGImages/2011_000414.jpg
1296 | /JPEGImages/2011_000448.jpg
1297 | /JPEGImages/2011_000451.jpg
1298 | /JPEGImages/2011_000470.jpg
1299 | /JPEGImages/2011_000473.jpg
1300 | /JPEGImages/2011_000515.jpg
1301 | /JPEGImages/2011_000537.jpg
1302 | /JPEGImages/2011_000576.jpg
1303 | /JPEGImages/2011_000603.jpg
1304 | /JPEGImages/2011_000616.jpg
1305 | /JPEGImages/2011_000636.jpg
1306 | /JPEGImages/2011_000639.jpg
1307 | /JPEGImages/2011_000654.jpg
1308 | /JPEGImages/2011_000660.jpg
1309 | /JPEGImages/2011_000664.jpg
1310 | /JPEGImages/2011_000667.jpg
1311 | /JPEGImages/2011_000670.jpg
1312 | /JPEGImages/2011_000676.jpg
1313 | /JPEGImages/2011_000721.jpg
1314 | /JPEGImages/2011_000723.jpg
1315 | /JPEGImages/2011_000762.jpg
1316 | /JPEGImages/2011_000766.jpg
1317 | /JPEGImages/2011_000786.jpg
1318 | /JPEGImages/2011_000802.jpg
1319 | /JPEGImages/2011_000810.jpg
1320 | /JPEGImages/2011_000821.jpg
1321 | /JPEGImages/2011_000841.jpg
1322 | /JPEGImages/2011_000844.jpg
1323 | /JPEGImages/2011_000846.jpg
1324 | /JPEGImages/2011_000869.jpg
1325 | /JPEGImages/2011_000890.jpg
1326 | /JPEGImages/2011_000915.jpg
1327 | /JPEGImages/2011_000924.jpg
1328 | /JPEGImages/2011_000937.jpg
1329 | /JPEGImages/2011_000939.jpg
1330 | /JPEGImages/2011_000952.jpg
1331 | /JPEGImages/2011_000968.jpg
1332 | /JPEGImages/2011_000974.jpg
1333 | /JPEGImages/2011_001037.jpg
1334 | /JPEGImages/2011_001072.jpg
1335 | /JPEGImages/2011_001085.jpg
1336 | /JPEGImages/2011_001089.jpg
1337 | /JPEGImages/2011_001090.jpg
1338 | /JPEGImages/2011_001099.jpg
1339 | /JPEGImages/2011_001104.jpg
1340 | /JPEGImages/2011_001112.jpg
1341 | /JPEGImages/2011_001120.jpg
1342 | /JPEGImages/2011_001132.jpg
1343 | /JPEGImages/2011_001151.jpg
1344 | /JPEGImages/2011_001194.jpg
1345 | /JPEGImages/2011_001258.jpg
1346 | /JPEGImages/2011_001274.jpg
1347 | /JPEGImages/2011_001314.jpg
1348 | /JPEGImages/2011_001317.jpg
1349 | /JPEGImages/2011_001321.jpg
1350 | /JPEGImages/2011_001379.jpg
1351 | /JPEGImages/2011_001425.jpg
1352 | /JPEGImages/2011_001431.jpg
1353 | /JPEGImages/2011_001443.jpg
1354 | /JPEGImages/2011_001446.jpg
1355 | /JPEGImages/2011_001452.jpg
1356 | /JPEGImages/2011_001454.jpg
1357 | /JPEGImages/2011_001477.jpg
1358 | /JPEGImages/2011_001509.jpg
1359 | /JPEGImages/2011_001512.jpg
1360 | /JPEGImages/2011_001515.jpg
1361 | /JPEGImages/2011_001528.jpg
1362 | /JPEGImages/2011_001554.jpg
1363 | /JPEGImages/2011_001561.jpg
1364 | /JPEGImages/2011_001580.jpg
1365 | /JPEGImages/2011_001587.jpg
1366 | /JPEGImages/2011_001623.jpg
1367 | /JPEGImages/2011_001648.jpg
1368 | /JPEGImages/2011_001651.jpg
1369 | /JPEGImages/2011_001654.jpg
1370 | /JPEGImages/2011_001684.jpg
1371 | /JPEGImages/2011_001696.jpg
1372 | /JPEGImages/2011_001697.jpg
1373 | /JPEGImages/2011_001760.jpg
1374 | /JPEGImages/2011_001761.jpg
1375 | /JPEGImages/2011_001798.jpg
1376 | /JPEGImages/2011_001807.jpg
1377 | /JPEGImages/2011_001851.jpg
1378 | /JPEGImages/2011_001852.jpg
1379 | /JPEGImages/2011_001853.jpg
1380 | /JPEGImages/2011_001888.jpg
1381 | /JPEGImages/2011_001940.jpg
1382 | /JPEGImages/2011_002014.jpg
1383 | /JPEGImages/2011_002028.jpg
1384 | /JPEGImages/2011_002056.jpg
1385 | /JPEGImages/2011_002061.jpg
1386 | /JPEGImages/2011_002068.jpg
1387 | /JPEGImages/2011_002076.jpg
1388 | /JPEGImages/2011_002090.jpg
1389 | /JPEGImages/2011_002095.jpg
1390 | /JPEGImages/2011_002104.jpg
1391 | /JPEGImages/2011_002136.jpg
1392 | /JPEGImages/2011_002138.jpg
1393 | /JPEGImages/2011_002151.jpg
1394 | /JPEGImages/2011_002153.jpg
1395 | /JPEGImages/2011_002155.jpg
1396 | /JPEGImages/2011_002197.jpg
1397 | /JPEGImages/2011_002198.jpg
1398 | /JPEGImages/2011_002243.jpg
1399 | /JPEGImages/2011_002250.jpg
1400 | /JPEGImages/2011_002257.jpg
1401 | /JPEGImages/2011_002262.jpg
1402 | /JPEGImages/2011_002264.jpg
1403 | /JPEGImages/2011_002296.jpg
1404 | /JPEGImages/2011_002314.jpg
1405 | /JPEGImages/2011_002331.jpg
1406 | /JPEGImages/2011_002333.jpg
1407 | /JPEGImages/2011_002411.jpg
1408 | /JPEGImages/2011_002417.jpg
1409 | /JPEGImages/2011_002425.jpg
1410 | /JPEGImages/2011_002437.jpg
1411 | /JPEGImages/2011_002444.jpg
1412 | /JPEGImages/2011_002445.jpg
1413 | /JPEGImages/2011_002449.jpg
1414 | /JPEGImages/2011_002468.jpg
1415 | /JPEGImages/2011_002469.jpg
1416 | /JPEGImages/2011_002473.jpg
1417 | /JPEGImages/2011_002508.jpg
1418 | /JPEGImages/2011_002523.jpg
1419 | /JPEGImages/2011_002534.jpg
1420 | /JPEGImages/2011_002557.jpg
1421 | /JPEGImages/2011_002564.jpg
1422 | /JPEGImages/2011_002572.jpg
1423 | /JPEGImages/2011_002597.jpg
1424 | /JPEGImages/2011_002622.jpg
1425 | /JPEGImages/2011_002632.jpg
1426 | /JPEGImages/2011_002635.jpg
1427 | /JPEGImages/2011_002643.jpg
1428 | /JPEGImages/2011_002653.jpg
1429 | /JPEGImages/2011_002667.jpg
1430 | /JPEGImages/2011_002681.jpg
1431 | /JPEGImages/2011_002707.jpg
1432 | /JPEGImages/2011_002736.jpg
1433 | /JPEGImages/2011_002759.jpg
1434 | /JPEGImages/2011_002783.jpg
1435 | /JPEGImages/2011_002792.jpg
1436 | /JPEGImages/2011_002799.jpg
1437 | /JPEGImages/2011_002824.jpg
1438 | /JPEGImages/2011_002835.jpg
1439 | /JPEGImages/2011_002866.jpg
1440 | /JPEGImages/2011_002876.jpg
1441 | /JPEGImages/2011_002888.jpg
1442 | /JPEGImages/2011_002894.jpg
1443 | /JPEGImages/2011_002903.jpg
1444 | /JPEGImages/2011_002905.jpg
1445 | /JPEGImages/2011_002986.jpg
1446 | /JPEGImages/2011_003045.jpg
1447 | /JPEGImages/2011_003064.jpg
1448 | /JPEGImages/2011_003070.jpg
1449 | /JPEGImages/2011_003083.jpg
1450 | /JPEGImages/2011_003093.jpg
1451 | /JPEGImages/2011_003096.jpg
1452 | /JPEGImages/2011_003102.jpg
1453 | /JPEGImages/2011_003156.jpg
1454 | /JPEGImages/2011_003170.jpg
1455 | /JPEGImages/2011_003178.jpg
1456 | /JPEGImages/2011_003231.jpg
1457 | 


--------------------------------------------------------------------------------
/voc12/val_id.txt:
--------------------------------------------------------------------------------
   1 | 2007_000033
   2 | 2007_000042
   3 | 2007_000061
   4 | 2007_000123
   5 | 2007_000129
   6 | 2007_000175
   7 | 2007_000187
   8 | 2007_000323
   9 | 2007_000332
  10 | 2007_000346
  11 | 2007_000452
  12 | 2007_000464
  13 | 2007_000491
  14 | 2007_000529
  15 | 2007_000559
  16 | 2007_000572
  17 | 2007_000629
  18 | 2007_000636
  19 | 2007_000661
  20 | 2007_000663
  21 | 2007_000676
  22 | 2007_000727
  23 | 2007_000762
  24 | 2007_000783
  25 | 2007_000799
  26 | 2007_000804
  27 | 2007_000830
  28 | 2007_000837
  29 | 2007_000847
  30 | 2007_000862
  31 | 2007_000925
  32 | 2007_000999
  33 | 2007_001154
  34 | 2007_001175
  35 | 2007_001239
  36 | 2007_001284
  37 | 2007_001288
  38 | 2007_001289
  39 | 2007_001299
  40 | 2007_001311
  41 | 2007_001321
  42 | 2007_001377
  43 | 2007_001408
  44 | 2007_001423
  45 | 2007_001430
  46 | 2007_001457
  47 | 2007_001458
  48 | 2007_001526
  49 | 2007_001568
  50 | 2007_001585
  51 | 2007_001586
  52 | 2007_001587
  53 | 2007_001594
  54 | 2007_001630
  55 | 2007_001677
  56 | 2007_001678
  57 | 2007_001717
  58 | 2007_001733
  59 | 2007_001761
  60 | 2007_001763
  61 | 2007_001774
  62 | 2007_001884
  63 | 2007_001955
  64 | 2007_002046
  65 | 2007_002094
  66 | 2007_002119
  67 | 2007_002132
  68 | 2007_002260
  69 | 2007_002266
  70 | 2007_002268
  71 | 2007_002284
  72 | 2007_002376
  73 | 2007_002378
  74 | 2007_002387
  75 | 2007_002400
  76 | 2007_002412
  77 | 2007_002426
  78 | 2007_002427
  79 | 2007_002445
  80 | 2007_002470
  81 | 2007_002539
  82 | 2007_002565
  83 | 2007_002597
  84 | 2007_002618
  85 | 2007_002619
  86 | 2007_002624
  87 | 2007_002643
  88 | 2007_002648
  89 | 2007_002719
  90 | 2007_002728
  91 | 2007_002823
  92 | 2007_002824
  93 | 2007_002852
  94 | 2007_002903
  95 | 2007_003011
  96 | 2007_003020
  97 | 2007_003022
  98 | 2007_003051
  99 | 2007_003088
 100 | 2007_003101
 101 | 2007_003106
 102 | 2007_003110
 103 | 2007_003131
 104 | 2007_003134
 105 | 2007_003137
 106 | 2007_003143
 107 | 2007_003169
 108 | 2007_003188
 109 | 2007_003194
 110 | 2007_003195
 111 | 2007_003201
 112 | 2007_003349
 113 | 2007_003367
 114 | 2007_003373
 115 | 2007_003499
 116 | 2007_003503
 117 | 2007_003506
 118 | 2007_003530
 119 | 2007_003571
 120 | 2007_003587
 121 | 2007_003611
 122 | 2007_003621
 123 | 2007_003682
 124 | 2007_003711
 125 | 2007_003714
 126 | 2007_003742
 127 | 2007_003786
 128 | 2007_003841
 129 | 2007_003848
 130 | 2007_003861
 131 | 2007_003872
 132 | 2007_003917
 133 | 2007_003957
 134 | 2007_003991
 135 | 2007_004033
 136 | 2007_004052
 137 | 2007_004112
 138 | 2007_004121
 139 | 2007_004143
 140 | 2007_004189
 141 | 2007_004190
 142 | 2007_004193
 143 | 2007_004241
 144 | 2007_004275
 145 | 2007_004281
 146 | 2007_004380
 147 | 2007_004392
 148 | 2007_004405
 149 | 2007_004468
 150 | 2007_004483
 151 | 2007_004510
 152 | 2007_004538
 153 | 2007_004558
 154 | 2007_004644
 155 | 2007_004649
 156 | 2007_004712
 157 | 2007_004722
 158 | 2007_004856
 159 | 2007_004866
 160 | 2007_004902
 161 | 2007_004969
 162 | 2007_005058
 163 | 2007_005074
 164 | 2007_005107
 165 | 2007_005114
 166 | 2007_005149
 167 | 2007_005173
 168 | 2007_005281
 169 | 2007_005294
 170 | 2007_005296
 171 | 2007_005304
 172 | 2007_005331
 173 | 2007_005354
 174 | 2007_005358
 175 | 2007_005428
 176 | 2007_005460
 177 | 2007_005469
 178 | 2007_005509
 179 | 2007_005547
 180 | 2007_005600
 181 | 2007_005608
 182 | 2007_005626
 183 | 2007_005689
 184 | 2007_005696
 185 | 2007_005705
 186 | 2007_005759
 187 | 2007_005803
 188 | 2007_005813
 189 | 2007_005828
 190 | 2007_005844
 191 | 2007_005845
 192 | 2007_005857
 193 | 2007_005911
 194 | 2007_005915
 195 | 2007_005978
 196 | 2007_006028
 197 | 2007_006035
 198 | 2007_006046
 199 | 2007_006076
 200 | 2007_006086
 201 | 2007_006117
 202 | 2007_006171
 203 | 2007_006241
 204 | 2007_006260
 205 | 2007_006277
 206 | 2007_006348
 207 | 2007_006364
 208 | 2007_006373
 209 | 2007_006444
 210 | 2007_006449
 211 | 2007_006549
 212 | 2007_006553
 213 | 2007_006560
 214 | 2007_006647
 215 | 2007_006678
 216 | 2007_006680
 217 | 2007_006698
 218 | 2007_006761
 219 | 2007_006802
 220 | 2007_006837
 221 | 2007_006841
 222 | 2007_006864
 223 | 2007_006866
 224 | 2007_006946
 225 | 2007_007007
 226 | 2007_007084
 227 | 2007_007109
 228 | 2007_007130
 229 | 2007_007165
 230 | 2007_007168
 231 | 2007_007195
 232 | 2007_007196
 233 | 2007_007203
 234 | 2007_007211
 235 | 2007_007235
 236 | 2007_007341
 237 | 2007_007414
 238 | 2007_007417
 239 | 2007_007470
 240 | 2007_007477
 241 | 2007_007493
 242 | 2007_007498
 243 | 2007_007524
 244 | 2007_007534
 245 | 2007_007624
 246 | 2007_007651
 247 | 2007_007688
 248 | 2007_007748
 249 | 2007_007795
 250 | 2007_007810
 251 | 2007_007815
 252 | 2007_007818
 253 | 2007_007836
 254 | 2007_007849
 255 | 2007_007881
 256 | 2007_007996
 257 | 2007_008051
 258 | 2007_008084
 259 | 2007_008106
 260 | 2007_008110
 261 | 2007_008204
 262 | 2007_008222
 263 | 2007_008256
 264 | 2007_008260
 265 | 2007_008339
 266 | 2007_008374
 267 | 2007_008415
 268 | 2007_008430
 269 | 2007_008543
 270 | 2007_008547
 271 | 2007_008596
 272 | 2007_008645
 273 | 2007_008670
 274 | 2007_008708
 275 | 2007_008722
 276 | 2007_008747
 277 | 2007_008802
 278 | 2007_008815
 279 | 2007_008897
 280 | 2007_008944
 281 | 2007_008964
 282 | 2007_008973
 283 | 2007_008980
 284 | 2007_009015
 285 | 2007_009068
 286 | 2007_009084
 287 | 2007_009088
 288 | 2007_009096
 289 | 2007_009221
 290 | 2007_009245
 291 | 2007_009251
 292 | 2007_009252
 293 | 2007_009258
 294 | 2007_009320
 295 | 2007_009323
 296 | 2007_009331
 297 | 2007_009346
 298 | 2007_009392
 299 | 2007_009413
 300 | 2007_009419
 301 | 2007_009446
 302 | 2007_009458
 303 | 2007_009521
 304 | 2007_009562
 305 | 2007_009592
 306 | 2007_009654
 307 | 2007_009655
 308 | 2007_009684
 309 | 2007_009687
 310 | 2007_009691
 311 | 2007_009706
 312 | 2007_009750
 313 | 2007_009756
 314 | 2007_009764
 315 | 2007_009794
 316 | 2007_009817
 317 | 2007_009841
 318 | 2007_009897
 319 | 2007_009911
 320 | 2007_009923
 321 | 2007_009938
 322 | 2008_000009
 323 | 2008_000016
 324 | 2008_000073
 325 | 2008_000075
 326 | 2008_000080
 327 | 2008_000107
 328 | 2008_000120
 329 | 2008_000123
 330 | 2008_000149
 331 | 2008_000182
 332 | 2008_000213
 333 | 2008_000215
 334 | 2008_000223
 335 | 2008_000233
 336 | 2008_000234
 337 | 2008_000239
 338 | 2008_000254
 339 | 2008_000270
 340 | 2008_000271
 341 | 2008_000345
 342 | 2008_000359
 343 | 2008_000391
 344 | 2008_000401
 345 | 2008_000464
 346 | 2008_000469
 347 | 2008_000474
 348 | 2008_000501
 349 | 2008_000510
 350 | 2008_000533
 351 | 2008_000573
 352 | 2008_000589
 353 | 2008_000602
 354 | 2008_000630
 355 | 2008_000657
 356 | 2008_000661
 357 | 2008_000662
 358 | 2008_000666
 359 | 2008_000673
 360 | 2008_000700
 361 | 2008_000725
 362 | 2008_000731
 363 | 2008_000763
 364 | 2008_000765
 365 | 2008_000782
 366 | 2008_000795
 367 | 2008_000811
 368 | 2008_000848
 369 | 2008_000853
 370 | 2008_000863
 371 | 2008_000911
 372 | 2008_000919
 373 | 2008_000943
 374 | 2008_000992
 375 | 2008_001013
 376 | 2008_001028
 377 | 2008_001040
 378 | 2008_001070
 379 | 2008_001074
 380 | 2008_001076
 381 | 2008_001078
 382 | 2008_001135
 383 | 2008_001150
 384 | 2008_001170
 385 | 2008_001231
 386 | 2008_001249
 387 | 2008_001260
 388 | 2008_001283
 389 | 2008_001308
 390 | 2008_001379
 391 | 2008_001404
 392 | 2008_001433
 393 | 2008_001439
 394 | 2008_001478
 395 | 2008_001491
 396 | 2008_001504
 397 | 2008_001513
 398 | 2008_001514
 399 | 2008_001531
 400 | 2008_001546
 401 | 2008_001547
 402 | 2008_001580
 403 | 2008_001629
 404 | 2008_001640
 405 | 2008_001682
 406 | 2008_001688
 407 | 2008_001715
 408 | 2008_001821
 409 | 2008_001874
 410 | 2008_001885
 411 | 2008_001895
 412 | 2008_001966
 413 | 2008_001971
 414 | 2008_001992
 415 | 2008_002043
 416 | 2008_002152
 417 | 2008_002205
 418 | 2008_002212
 419 | 2008_002239
 420 | 2008_002240
 421 | 2008_002241
 422 | 2008_002269
 423 | 2008_002273
 424 | 2008_002358
 425 | 2008_002379
 426 | 2008_002383
 427 | 2008_002429
 428 | 2008_002464
 429 | 2008_002467
 430 | 2008_002492
 431 | 2008_002495
 432 | 2008_002504
 433 | 2008_002521
 434 | 2008_002536
 435 | 2008_002588
 436 | 2008_002623
 437 | 2008_002680
 438 | 2008_002681
 439 | 2008_002775
 440 | 2008_002778
 441 | 2008_002835
 442 | 2008_002859
 443 | 2008_002864
 444 | 2008_002900
 445 | 2008_002904
 446 | 2008_002929
 447 | 2008_002936
 448 | 2008_002942
 449 | 2008_002958
 450 | 2008_003003
 451 | 2008_003026
 452 | 2008_003034
 453 | 2008_003076
 454 | 2008_003105
 455 | 2008_003108
 456 | 2008_003110
 457 | 2008_003135
 458 | 2008_003141
 459 | 2008_003155
 460 | 2008_003210
 461 | 2008_003238
 462 | 2008_003270
 463 | 2008_003330
 464 | 2008_003333
 465 | 2008_003369
 466 | 2008_003379
 467 | 2008_003451
 468 | 2008_003461
 469 | 2008_003477
 470 | 2008_003492
 471 | 2008_003499
 472 | 2008_003511
 473 | 2008_003546
 474 | 2008_003576
 475 | 2008_003577
 476 | 2008_003676
 477 | 2008_003709
 478 | 2008_003733
 479 | 2008_003777
 480 | 2008_003782
 481 | 2008_003821
 482 | 2008_003846
 483 | 2008_003856
 484 | 2008_003858
 485 | 2008_003874
 486 | 2008_003876
 487 | 2008_003885
 488 | 2008_003886
 489 | 2008_003926
 490 | 2008_003976
 491 | 2008_004069
 492 | 2008_004101
 493 | 2008_004140
 494 | 2008_004172
 495 | 2008_004175
 496 | 2008_004212
 497 | 2008_004279
 498 | 2008_004339
 499 | 2008_004345
 500 | 2008_004363
 501 | 2008_004367
 502 | 2008_004396
 503 | 2008_004399
 504 | 2008_004453
 505 | 2008_004477
 506 | 2008_004552
 507 | 2008_004562
 508 | 2008_004575
 509 | 2008_004610
 510 | 2008_004612
 511 | 2008_004621
 512 | 2008_004624
 513 | 2008_004654
 514 | 2008_004659
 515 | 2008_004687
 516 | 2008_004701
 517 | 2008_004704
 518 | 2008_004705
 519 | 2008_004754
 520 | 2008_004758
 521 | 2008_004854
 522 | 2008_004910
 523 | 2008_004995
 524 | 2008_005049
 525 | 2008_005089
 526 | 2008_005097
 527 | 2008_005105
 528 | 2008_005145
 529 | 2008_005197
 530 | 2008_005217
 531 | 2008_005242
 532 | 2008_005245
 533 | 2008_005254
 534 | 2008_005262
 535 | 2008_005338
 536 | 2008_005398
 537 | 2008_005399
 538 | 2008_005422
 539 | 2008_005439
 540 | 2008_005445
 541 | 2008_005525
 542 | 2008_005544
 543 | 2008_005628
 544 | 2008_005633
 545 | 2008_005637
 546 | 2008_005642
 547 | 2008_005676
 548 | 2008_005680
 549 | 2008_005691
 550 | 2008_005727
 551 | 2008_005738
 552 | 2008_005812
 553 | 2008_005904
 554 | 2008_005915
 555 | 2008_006008
 556 | 2008_006036
 557 | 2008_006055
 558 | 2008_006063
 559 | 2008_006108
 560 | 2008_006130
 561 | 2008_006143
 562 | 2008_006159
 563 | 2008_006216
 564 | 2008_006219
 565 | 2008_006229
 566 | 2008_006254
 567 | 2008_006275
 568 | 2008_006325
 569 | 2008_006327
 570 | 2008_006341
 571 | 2008_006408
 572 | 2008_006480
 573 | 2008_006523
 574 | 2008_006526
 575 | 2008_006528
 576 | 2008_006553
 577 | 2008_006554
 578 | 2008_006703
 579 | 2008_006722
 580 | 2008_006752
 581 | 2008_006784
 582 | 2008_006835
 583 | 2008_006874
 584 | 2008_006981
 585 | 2008_006986
 586 | 2008_007025
 587 | 2008_007031
 588 | 2008_007048
 589 | 2008_007120
 590 | 2008_007123
 591 | 2008_007143
 592 | 2008_007194
 593 | 2008_007219
 594 | 2008_007273
 595 | 2008_007350
 596 | 2008_007378
 597 | 2008_007392
 598 | 2008_007402
 599 | 2008_007497
 600 | 2008_007498
 601 | 2008_007507
 602 | 2008_007513
 603 | 2008_007527
 604 | 2008_007548
 605 | 2008_007596
 606 | 2008_007677
 607 | 2008_007737
 608 | 2008_007797
 609 | 2008_007804
 610 | 2008_007811
 611 | 2008_007814
 612 | 2008_007828
 613 | 2008_007836
 614 | 2008_007945
 615 | 2008_007994
 616 | 2008_008051
 617 | 2008_008103
 618 | 2008_008127
 619 | 2008_008221
 620 | 2008_008252
 621 | 2008_008268
 622 | 2008_008296
 623 | 2008_008301
 624 | 2008_008335
 625 | 2008_008362
 626 | 2008_008392
 627 | 2008_008393
 628 | 2008_008421
 629 | 2008_008434
 630 | 2008_008469
 631 | 2008_008629
 632 | 2008_008682
 633 | 2008_008711
 634 | 2008_008746
 635 | 2009_000012
 636 | 2009_000013
 637 | 2009_000022
 638 | 2009_000032
 639 | 2009_000037
 640 | 2009_000039
 641 | 2009_000074
 642 | 2009_000080
 643 | 2009_000087
 644 | 2009_000096
 645 | 2009_000121
 646 | 2009_000136
 647 | 2009_000149
 648 | 2009_000156
 649 | 2009_000201
 650 | 2009_000205
 651 | 2009_000219
 652 | 2009_000242
 653 | 2009_000309
 654 | 2009_000318
 655 | 2009_000335
 656 | 2009_000351
 657 | 2009_000354
 658 | 2009_000387
 659 | 2009_000391
 660 | 2009_000412
 661 | 2009_000418
 662 | 2009_000421
 663 | 2009_000426
 664 | 2009_000440
 665 | 2009_000446
 666 | 2009_000455
 667 | 2009_000457
 668 | 2009_000469
 669 | 2009_000487
 670 | 2009_000488
 671 | 2009_000523
 672 | 2009_000573
 673 | 2009_000619
 674 | 2009_000628
 675 | 2009_000641
 676 | 2009_000664
 677 | 2009_000675
 678 | 2009_000704
 679 | 2009_000705
 680 | 2009_000712
 681 | 2009_000716
 682 | 2009_000723
 683 | 2009_000727
 684 | 2009_000730
 685 | 2009_000731
 686 | 2009_000732
 687 | 2009_000771
 688 | 2009_000825
 689 | 2009_000828
 690 | 2009_000839
 691 | 2009_000840
 692 | 2009_000845
 693 | 2009_000879
 694 | 2009_000892
 695 | 2009_000919
 696 | 2009_000924
 697 | 2009_000931
 698 | 2009_000935
 699 | 2009_000964
 700 | 2009_000989
 701 | 2009_000991
 702 | 2009_000998
 703 | 2009_001008
 704 | 2009_001082
 705 | 2009_001108
 706 | 2009_001160
 707 | 2009_001215
 708 | 2009_001240
 709 | 2009_001255
 710 | 2009_001278
 711 | 2009_001299
 712 | 2009_001300
 713 | 2009_001314
 714 | 2009_001332
 715 | 2009_001333
 716 | 2009_001363
 717 | 2009_001391
 718 | 2009_001411
 719 | 2009_001433
 720 | 2009_001505
 721 | 2009_001535
 722 | 2009_001536
 723 | 2009_001565
 724 | 2009_001607
 725 | 2009_001644
 726 | 2009_001663
 727 | 2009_001683
 728 | 2009_001684
 729 | 2009_001687
 730 | 2009_001718
 731 | 2009_001731
 732 | 2009_001765
 733 | 2009_001768
 734 | 2009_001775
 735 | 2009_001804
 736 | 2009_001816
 737 | 2009_001818
 738 | 2009_001850
 739 | 2009_001851
 740 | 2009_001854
 741 | 2009_001941
 742 | 2009_001991
 743 | 2009_002012
 744 | 2009_002035
 745 | 2009_002042
 746 | 2009_002082
 747 | 2009_002094
 748 | 2009_002097
 749 | 2009_002122
 750 | 2009_002150
 751 | 2009_002155
 752 | 2009_002164
 753 | 2009_002165
 754 | 2009_002171
 755 | 2009_002185
 756 | 2009_002202
 757 | 2009_002221
 758 | 2009_002238
 759 | 2009_002239
 760 | 2009_002265
 761 | 2009_002268
 762 | 2009_002291
 763 | 2009_002295
 764 | 2009_002317
 765 | 2009_002320
 766 | 2009_002346
 767 | 2009_002366
 768 | 2009_002372
 769 | 2009_002382
 770 | 2009_002390
 771 | 2009_002415
 772 | 2009_002445
 773 | 2009_002487
 774 | 2009_002521
 775 | 2009_002527
 776 | 2009_002535
 777 | 2009_002539
 778 | 2009_002549
 779 | 2009_002562
 780 | 2009_002568
 781 | 2009_002571
 782 | 2009_002573
 783 | 2009_002584
 784 | 2009_002591
 785 | 2009_002594
 786 | 2009_002604
 787 | 2009_002618
 788 | 2009_002635
 789 | 2009_002638
 790 | 2009_002649
 791 | 2009_002651
 792 | 2009_002727
 793 | 2009_002732
 794 | 2009_002749
 795 | 2009_002753
 796 | 2009_002771
 797 | 2009_002808
 798 | 2009_002856
 799 | 2009_002887
 800 | 2009_002888
 801 | 2009_002928
 802 | 2009_002936
 803 | 2009_002975
 804 | 2009_002982
 805 | 2009_002990
 806 | 2009_003003
 807 | 2009_003005
 808 | 2009_003043
 809 | 2009_003059
 810 | 2009_003063
 811 | 2009_003065
 812 | 2009_003071
 813 | 2009_003080
 814 | 2009_003105
 815 | 2009_003123
 816 | 2009_003193
 817 | 2009_003196
 818 | 2009_003217
 819 | 2009_003224
 820 | 2009_003241
 821 | 2009_003269
 822 | 2009_003273
 823 | 2009_003299
 824 | 2009_003304
 825 | 2009_003311
 826 | 2009_003323
 827 | 2009_003343
 828 | 2009_003378
 829 | 2009_003387
 830 | 2009_003406
 831 | 2009_003433
 832 | 2009_003450
 833 | 2009_003466
 834 | 2009_003481
 835 | 2009_003494
 836 | 2009_003498
 837 | 2009_003504
 838 | 2009_003507
 839 | 2009_003517
 840 | 2009_003523
 841 | 2009_003542
 842 | 2009_003549
 843 | 2009_003551
 844 | 2009_003564
 845 | 2009_003569
 846 | 2009_003576
 847 | 2009_003589
 848 | 2009_003607
 849 | 2009_003640
 850 | 2009_003666
 851 | 2009_003696
 852 | 2009_003703
 853 | 2009_003707
 854 | 2009_003756
 855 | 2009_003771
 856 | 2009_003773
 857 | 2009_003804
 858 | 2009_003806
 859 | 2009_003810
 860 | 2009_003849
 861 | 2009_003857
 862 | 2009_003858
 863 | 2009_003895
 864 | 2009_003903
 865 | 2009_003904
 866 | 2009_003928
 867 | 2009_003938
 868 | 2009_003971
 869 | 2009_003991
 870 | 2009_004021
 871 | 2009_004033
 872 | 2009_004043
 873 | 2009_004070
 874 | 2009_004072
 875 | 2009_004084
 876 | 2009_004099
 877 | 2009_004125
 878 | 2009_004140
 879 | 2009_004217
 880 | 2009_004221
 881 | 2009_004247
 882 | 2009_004248
 883 | 2009_004255
 884 | 2009_004298
 885 | 2009_004324
 886 | 2009_004455
 887 | 2009_004494
 888 | 2009_004497
 889 | 2009_004504
 890 | 2009_004507
 891 | 2009_004509
 892 | 2009_004540
 893 | 2009_004568
 894 | 2009_004579
 895 | 2009_004581
 896 | 2009_004590
 897 | 2009_004592
 898 | 2009_004594
 899 | 2009_004635
 900 | 2009_004653
 901 | 2009_004687
 902 | 2009_004721
 903 | 2009_004730
 904 | 2009_004732
 905 | 2009_004738
 906 | 2009_004748
 907 | 2009_004789
 908 | 2009_004799
 909 | 2009_004801
 910 | 2009_004848
 911 | 2009_004859
 912 | 2009_004867
 913 | 2009_004882
 914 | 2009_004886
 915 | 2009_004895
 916 | 2009_004942
 917 | 2009_004969
 918 | 2009_004987
 919 | 2009_004993
 920 | 2009_004994
 921 | 2009_005038
 922 | 2009_005078
 923 | 2009_005087
 924 | 2009_005089
 925 | 2009_005137
 926 | 2009_005148
 927 | 2009_005156
 928 | 2009_005158
 929 | 2009_005189
 930 | 2009_005190
 931 | 2009_005217
 932 | 2009_005219
 933 | 2009_005220
 934 | 2009_005231
 935 | 2009_005260
 936 | 2009_005262
 937 | 2009_005302
 938 | 2010_000003
 939 | 2010_000038
 940 | 2010_000065
 941 | 2010_000083
 942 | 2010_000084
 943 | 2010_000087
 944 | 2010_000110
 945 | 2010_000159
 946 | 2010_000160
 947 | 2010_000163
 948 | 2010_000174
 949 | 2010_000216
 950 | 2010_000238
 951 | 2010_000241
 952 | 2010_000256
 953 | 2010_000272
 954 | 2010_000284
 955 | 2010_000309
 956 | 2010_000318
 957 | 2010_000330
 958 | 2010_000335
 959 | 2010_000342
 960 | 2010_000372
 961 | 2010_000422
 962 | 2010_000426
 963 | 2010_000427
 964 | 2010_000502
 965 | 2010_000530
 966 | 2010_000552
 967 | 2010_000559
 968 | 2010_000572
 969 | 2010_000573
 970 | 2010_000622
 971 | 2010_000628
 972 | 2010_000639
 973 | 2010_000666
 974 | 2010_000679
 975 | 2010_000682
 976 | 2010_000683
 977 | 2010_000724
 978 | 2010_000738
 979 | 2010_000764
 980 | 2010_000788
 981 | 2010_000814
 982 | 2010_000836
 983 | 2010_000874
 984 | 2010_000904
 985 | 2010_000906
 986 | 2010_000907
 987 | 2010_000918
 988 | 2010_000929
 989 | 2010_000941
 990 | 2010_000952
 991 | 2010_000961
 992 | 2010_001000
 993 | 2010_001010
 994 | 2010_001011
 995 | 2010_001016
 996 | 2010_001017
 997 | 2010_001024
 998 | 2010_001036
 999 | 2010_001061
1000 | 2010_001069
1001 | 2010_001070
1002 | 2010_001079
1003 | 2010_001104
1004 | 2010_001124
1005 | 2010_001149
1006 | 2010_001151
1007 | 2010_001174
1008 | 2010_001206
1009 | 2010_001246
1010 | 2010_001251
1011 | 2010_001256
1012 | 2010_001264
1013 | 2010_001292
1014 | 2010_001313
1015 | 2010_001327
1016 | 2010_001331
1017 | 2010_001351
1018 | 2010_001367
1019 | 2010_001376
1020 | 2010_001403
1021 | 2010_001448
1022 | 2010_001451
1023 | 2010_001522
1024 | 2010_001534
1025 | 2010_001553
1026 | 2010_001557
1027 | 2010_001563
1028 | 2010_001577
1029 | 2010_001579
1030 | 2010_001646
1031 | 2010_001656
1032 | 2010_001692
1033 | 2010_001699
1034 | 2010_001734
1035 | 2010_001752
1036 | 2010_001767
1037 | 2010_001768
1038 | 2010_001773
1039 | 2010_001820
1040 | 2010_001830
1041 | 2010_001851
1042 | 2010_001908
1043 | 2010_001913
1044 | 2010_001951
1045 | 2010_001956
1046 | 2010_001962
1047 | 2010_001966
1048 | 2010_001995
1049 | 2010_002017
1050 | 2010_002025
1051 | 2010_002030
1052 | 2010_002106
1053 | 2010_002137
1054 | 2010_002142
1055 | 2010_002146
1056 | 2010_002147
1057 | 2010_002150
1058 | 2010_002161
1059 | 2010_002200
1060 | 2010_002228
1061 | 2010_002232
1062 | 2010_002251
1063 | 2010_002271
1064 | 2010_002305
1065 | 2010_002310
1066 | 2010_002336
1067 | 2010_002348
1068 | 2010_002361
1069 | 2010_002390
1070 | 2010_002396
1071 | 2010_002422
1072 | 2010_002450
1073 | 2010_002480
1074 | 2010_002512
1075 | 2010_002531
1076 | 2010_002536
1077 | 2010_002538
1078 | 2010_002546
1079 | 2010_002623
1080 | 2010_002682
1081 | 2010_002691
1082 | 2010_002693
1083 | 2010_002701
1084 | 2010_002763
1085 | 2010_002792
1086 | 2010_002868
1087 | 2010_002900
1088 | 2010_002902
1089 | 2010_002921
1090 | 2010_002929
1091 | 2010_002939
1092 | 2010_002988
1093 | 2010_003014
1094 | 2010_003060
1095 | 2010_003123
1096 | 2010_003127
1097 | 2010_003132
1098 | 2010_003168
1099 | 2010_003183
1100 | 2010_003187
1101 | 2010_003207
1102 | 2010_003231
1103 | 2010_003239
1104 | 2010_003275
1105 | 2010_003276
1106 | 2010_003293
1107 | 2010_003302
1108 | 2010_003325
1109 | 2010_003362
1110 | 2010_003365
1111 | 2010_003381
1112 | 2010_003402
1113 | 2010_003409
1114 | 2010_003418
1115 | 2010_003446
1116 | 2010_003453
1117 | 2010_003468
1118 | 2010_003473
1119 | 2010_003495
1120 | 2010_003506
1121 | 2010_003514
1122 | 2010_003531
1123 | 2010_003532
1124 | 2010_003541
1125 | 2010_003547
1126 | 2010_003597
1127 | 2010_003675
1128 | 2010_003708
1129 | 2010_003716
1130 | 2010_003746
1131 | 2010_003758
1132 | 2010_003764
1133 | 2010_003768
1134 | 2010_003771
1135 | 2010_003772
1136 | 2010_003781
1137 | 2010_003813
1138 | 2010_003820
1139 | 2010_003854
1140 | 2010_003912
1141 | 2010_003915
1142 | 2010_003947
1143 | 2010_003956
1144 | 2010_003971
1145 | 2010_004041
1146 | 2010_004042
1147 | 2010_004056
1148 | 2010_004063
1149 | 2010_004104
1150 | 2010_004120
1151 | 2010_004149
1152 | 2010_004165
1153 | 2010_004208
1154 | 2010_004219
1155 | 2010_004226
1156 | 2010_004314
1157 | 2010_004320
1158 | 2010_004322
1159 | 2010_004337
1160 | 2010_004348
1161 | 2010_004355
1162 | 2010_004369
1163 | 2010_004382
1164 | 2010_004419
1165 | 2010_004432
1166 | 2010_004472
1167 | 2010_004479
1168 | 2010_004519
1169 | 2010_004520
1170 | 2010_004529
1171 | 2010_004543
1172 | 2010_004550
1173 | 2010_004551
1174 | 2010_004556
1175 | 2010_004559
1176 | 2010_004628
1177 | 2010_004635
1178 | 2010_004662
1179 | 2010_004697
1180 | 2010_004757
1181 | 2010_004763
1182 | 2010_004772
1183 | 2010_004783
1184 | 2010_004789
1185 | 2010_004795
1186 | 2010_004815
1187 | 2010_004825
1188 | 2010_004828
1189 | 2010_004856
1190 | 2010_004857
1191 | 2010_004861
1192 | 2010_004941
1193 | 2010_004946
1194 | 2010_004951
1195 | 2010_004980
1196 | 2010_004994
1197 | 2010_005013
1198 | 2010_005021
1199 | 2010_005046
1200 | 2010_005063
1201 | 2010_005108
1202 | 2010_005118
1203 | 2010_005159
1204 | 2010_005160
1205 | 2010_005166
1206 | 2010_005174
1207 | 2010_005180
1208 | 2010_005187
1209 | 2010_005206
1210 | 2010_005245
1211 | 2010_005252
1212 | 2010_005284
1213 | 2010_005305
1214 | 2010_005344
1215 | 2010_005353
1216 | 2010_005366
1217 | 2010_005401
1218 | 2010_005421
1219 | 2010_005428
1220 | 2010_005432
1221 | 2010_005433
1222 | 2010_005496
1223 | 2010_005501
1224 | 2010_005508
1225 | 2010_005531
1226 | 2010_005534
1227 | 2010_005575
1228 | 2010_005582
1229 | 2010_005606
1230 | 2010_005626
1231 | 2010_005644
1232 | 2010_005664
1233 | 2010_005705
1234 | 2010_005706
1235 | 2010_005709
1236 | 2010_005718
1237 | 2010_005719
1238 | 2010_005727
1239 | 2010_005762
1240 | 2010_005788
1241 | 2010_005860
1242 | 2010_005871
1243 | 2010_005877
1244 | 2010_005888
1245 | 2010_005899
1246 | 2010_005922
1247 | 2010_005991
1248 | 2010_005992
1249 | 2010_006026
1250 | 2010_006034
1251 | 2010_006054
1252 | 2010_006070
1253 | 2011_000045
1254 | 2011_000051
1255 | 2011_000054
1256 | 2011_000066
1257 | 2011_000070
1258 | 2011_000112
1259 | 2011_000173
1260 | 2011_000178
1261 | 2011_000185
1262 | 2011_000226
1263 | 2011_000234
1264 | 2011_000238
1265 | 2011_000239
1266 | 2011_000248
1267 | 2011_000283
1268 | 2011_000291
1269 | 2011_000310
1270 | 2011_000312
1271 | 2011_000338
1272 | 2011_000396
1273 | 2011_000412
1274 | 2011_000419
1275 | 2011_000435
1276 | 2011_000436
1277 | 2011_000438
1278 | 2011_000455
1279 | 2011_000456
1280 | 2011_000479
1281 | 2011_000481
1282 | 2011_000482
1283 | 2011_000503
1284 | 2011_000512
1285 | 2011_000521
1286 | 2011_000526
1287 | 2011_000536
1288 | 2011_000548
1289 | 2011_000566
1290 | 2011_000585
1291 | 2011_000598
1292 | 2011_000607
1293 | 2011_000618
1294 | 2011_000638
1295 | 2011_000658
1296 | 2011_000661
1297 | 2011_000669
1298 | 2011_000747
1299 | 2011_000780
1300 | 2011_000789
1301 | 2011_000807
1302 | 2011_000809
1303 | 2011_000813
1304 | 2011_000830
1305 | 2011_000843
1306 | 2011_000874
1307 | 2011_000888
1308 | 2011_000900
1309 | 2011_000912
1310 | 2011_000953
1311 | 2011_000969
1312 | 2011_001005
1313 | 2011_001014
1314 | 2011_001020
1315 | 2011_001047
1316 | 2011_001060
1317 | 2011_001064
1318 | 2011_001069
1319 | 2011_001071
1320 | 2011_001082
1321 | 2011_001110
1322 | 2011_001114
1323 | 2011_001159
1324 | 2011_001161
1325 | 2011_001190
1326 | 2011_001232
1327 | 2011_001263
1328 | 2011_001276
1329 | 2011_001281
1330 | 2011_001287
1331 | 2011_001292
1332 | 2011_001313
1333 | 2011_001341
1334 | 2011_001346
1335 | 2011_001350
1336 | 2011_001407
1337 | 2011_001416
1338 | 2011_001421
1339 | 2011_001434
1340 | 2011_001447
1341 | 2011_001489
1342 | 2011_001529
1343 | 2011_001530
1344 | 2011_001534
1345 | 2011_001546
1346 | 2011_001567
1347 | 2011_001589
1348 | 2011_001597
1349 | 2011_001601
1350 | 2011_001607
1351 | 2011_001613
1352 | 2011_001614
1353 | 2011_001619
1354 | 2011_001624
1355 | 2011_001642
1356 | 2011_001665
1357 | 2011_001669
1358 | 2011_001674
1359 | 2011_001708
1360 | 2011_001713
1361 | 2011_001714
1362 | 2011_001722
1363 | 2011_001726
1364 | 2011_001745
1365 | 2011_001748
1366 | 2011_001775
1367 | 2011_001782
1368 | 2011_001793
1369 | 2011_001794
1370 | 2011_001812
1371 | 2011_001862
1372 | 2011_001863
1373 | 2011_001868
1374 | 2011_001880
1375 | 2011_001910
1376 | 2011_001984
1377 | 2011_001988
1378 | 2011_002002
1379 | 2011_002040
1380 | 2011_002041
1381 | 2011_002064
1382 | 2011_002075
1383 | 2011_002098
1384 | 2011_002110
1385 | 2011_002121
1386 | 2011_002124
1387 | 2011_002150
1388 | 2011_002156
1389 | 2011_002178
1390 | 2011_002200
1391 | 2011_002223
1392 | 2011_002244
1393 | 2011_002247
1394 | 2011_002279
1395 | 2011_002295
1396 | 2011_002298
1397 | 2011_002308
1398 | 2011_002317
1399 | 2011_002322
1400 | 2011_002327
1401 | 2011_002343
1402 | 2011_002358
1403 | 2011_002371
1404 | 2011_002379
1405 | 2011_002391
1406 | 2011_002498
1407 | 2011_002509
1408 | 2011_002515
1409 | 2011_002532
1410 | 2011_002535
1411 | 2011_002548
1412 | 2011_002575
1413 | 2011_002578
1414 | 2011_002589
1415 | 2011_002592
1416 | 2011_002623
1417 | 2011_002641
1418 | 2011_002644
1419 | 2011_002662
1420 | 2011_002675
1421 | 2011_002685
1422 | 2011_002713
1423 | 2011_002730
1424 | 2011_002754
1425 | 2011_002812
1426 | 2011_002863
1427 | 2011_002879
1428 | 2011_002885
1429 | 2011_002929
1430 | 2011_002951
1431 | 2011_002975
1432 | 2011_002993
1433 | 2011_002997
1434 | 2011_003003
1435 | 2011_003011
1436 | 2011_003019
1437 | 2011_003030
1438 | 2011_003055
1439 | 2011_003085
1440 | 2011_003103
1441 | 2011_003114
1442 | 2011_003145
1443 | 2011_003146
1444 | 2011_003182
1445 | 2011_003197
1446 | 2011_003205
1447 | 2011_003240
1448 | 2011_003256
1449 | 2011_003271


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