YarnYang commited on
Commit
8a54446
·
verified ·
1 Parent(s): 9f5a727

Upload model weights

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. best_weights/best_weights/MdaCD_LEVIRCD/20250403_131147/20250403_131147.log +0 -0
  2. best_weights/best_weights/MdaCD_LEVIRCD/20250403_131147/vis_data/20250403_131147.json +0 -0
  3. best_weights/best_weights/MdaCD_LEVIRCD/20250403_131147/vis_data/config.py +362 -0
  4. best_weights/best_weights/MdaCD_LEVIRCD/20250403_131147/vis_data/scalars.json +0 -0
  5. best_weights/best_weights/MdaCD_LEVIRCD/AugCD_LEVIRCD.py +362 -0
  6. best_weights/best_weights/MdaCD_LEVIRCD/best_mIoU_iter_20000.pth +3 -0
  7. best_weights/best_weights/MdaCD_LEVIRCD/last_checkpoint +1 -0
  8. best_weights/best_weights/MdaCD_LEVIRCD/overall_result.csv +2 -0
  9. best_weights/best_weights/MdaCD_SYSUCD/20250823_154932/20250823_154932.log +0 -0
  10. best_weights/best_weights/MdaCD_SYSUCD/20250823_154932/vis_data/20250823_154932.json +0 -0
  11. best_weights/best_weights/MdaCD_SYSUCD/20250823_154932/vis_data/config.py +362 -0
  12. best_weights/best_weights/MdaCD_SYSUCD/20250823_154932/vis_data/scalars.json +0 -0
  13. best_weights/best_weights/MdaCD_SYSUCD/MdaCD_SYSUCD.py +362 -0
  14. best_weights/best_weights/MdaCD_SYSUCD/best_mIoU_iter_7000.pth +3 -0
  15. best_weights/best_weights/MdaCD_SYSUCD/iter_10000.pth +3 -0
  16. best_weights/best_weights/MdaCD_SYSUCD/last_checkpoint +1 -0
  17. best_weights/best_weights/MdaCD_SYSUCD/overall_result.csv +2 -0
  18. best_weights/best_weights/MdaCD_SYSUCD/test_result/00000.png +0 -0
  19. best_weights/best_weights/MdaCD_SYSUCD/test_result/00001.png +0 -0
  20. best_weights/best_weights/MdaCD_SYSUCD/test_result/00002.png +0 -0
  21. best_weights/best_weights/MdaCD_SYSUCD/test_result/00003.png +0 -0
  22. best_weights/best_weights/MdaCD_SYSUCD/test_result/00004.png +0 -0
  23. best_weights/best_weights/MdaCD_SYSUCD/test_result/00005.png +0 -0
  24. best_weights/best_weights/MdaCD_SYSUCD/test_result/00006.png +0 -0
  25. best_weights/best_weights/MdaCD_SYSUCD/test_result/00007.png +0 -0
  26. best_weights/best_weights/MdaCD_SYSUCD/test_result/00008.png +0 -0
  27. best_weights/best_weights/MdaCD_SYSUCD/test_result/00009.png +0 -0
  28. best_weights/best_weights/MdaCD_SYSUCD/test_result/00010.png +0 -0
  29. best_weights/best_weights/MdaCD_SYSUCD/test_result/00011.png +0 -0
  30. best_weights/best_weights/MdaCD_SYSUCD/test_result/00012.png +0 -0
  31. best_weights/best_weights/MdaCD_SYSUCD/test_result/00013.png +0 -0
  32. best_weights/best_weights/MdaCD_SYSUCD/test_result/00014.png +0 -0
  33. best_weights/best_weights/MdaCD_SYSUCD/test_result/00015.png +0 -0
  34. best_weights/best_weights/MdaCD_SYSUCD/test_result/00016.png +0 -0
  35. best_weights/best_weights/MdaCD_SYSUCD/test_result/00017.png +0 -0
  36. best_weights/best_weights/MdaCD_SYSUCD/test_result/00018.png +0 -0
  37. best_weights/best_weights/MdaCD_SYSUCD/test_result/00019.png +0 -0
  38. best_weights/best_weights/MdaCD_SYSUCD/test_result/00020.png +0 -0
  39. best_weights/best_weights/MdaCD_SYSUCD/test_result/00021.png +0 -0
  40. best_weights/best_weights/MdaCD_SYSUCD/test_result/00022.png +0 -0
  41. best_weights/best_weights/MdaCD_SYSUCD/test_result/00023.png +0 -0
  42. best_weights/best_weights/MdaCD_SYSUCD/test_result/00024.png +0 -0
  43. best_weights/best_weights/MdaCD_SYSUCD/test_result/00025.png +0 -0
  44. best_weights/best_weights/MdaCD_SYSUCD/test_result/00026.png +0 -0
  45. best_weights/best_weights/MdaCD_SYSUCD/test_result/00027.png +0 -0
  46. best_weights/best_weights/MdaCD_SYSUCD/test_result/00028.png +0 -0
  47. best_weights/best_weights/MdaCD_SYSUCD/test_result/00029.png +0 -0
  48. best_weights/best_weights/MdaCD_SYSUCD/test_result/00030.png +0 -0
  49. best_weights/best_weights/MdaCD_SYSUCD/test_result/00031.png +0 -0
  50. best_weights/best_weights/MdaCD_SYSUCD/test_result/00032.png +0 -0
best_weights/best_weights/MdaCD_LEVIRCD/20250403_131147/20250403_131147.log ADDED
The diff for this file is too large to render. See raw diff
 
best_weights/best_weights/MdaCD_LEVIRCD/20250403_131147/vis_data/20250403_131147.json ADDED
The diff for this file is too large to render. See raw diff
 
best_weights/best_weights/MdaCD_LEVIRCD/20250403_131147/vis_data/config.py ADDED
@@ -0,0 +1,362 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ crop_size = (
2
+ 256,
3
+ 256,
4
+ )
5
+ data_preprocessor = dict(
6
+ mean=[
7
+ 123.675,
8
+ 116.28,
9
+ 103.53,
10
+ 123.675,
11
+ 116.28,
12
+ 103.53,
13
+ ],
14
+ size=(
15
+ 256,
16
+ 256,
17
+ ),
18
+ std=[
19
+ 58.395,
20
+ 57.12,
21
+ 57.375,
22
+ 58.395,
23
+ 57.12,
24
+ 57.375,
25
+ ],
26
+ type='SegDataPreProcessor')
27
+ data_root = '/data/Dataset/LEVIRCD'
28
+ dataset_type = 'TXTCDDatasetJSON'
29
+ default_hooks = dict(
30
+ checkpoint=dict(
31
+ by_epoch=False,
32
+ interval=10000,
33
+ save_best='mIoU',
34
+ type='CheckpointHook'),
35
+ logger=dict(interval=10, log_metric_by_epoch=False, type='LoggerHook'),
36
+ param_scheduler=dict(type='ParamSchedulerHook'),
37
+ sampler_seed=dict(type='DistSamplerSeedHook'),
38
+ timer=dict(type='IterTimerHook'),
39
+ visualization=dict(type='SegVisualizationHook'))
40
+ default_scope = 'mmseg'
41
+ env_cfg = dict(
42
+ cudnn_benchmark=True,
43
+ dist_cfg=dict(backend='nccl'),
44
+ mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0))
45
+ find_unused_parameters = True
46
+ launcher = 'pytorch'
47
+ load_from = None
48
+ log_level = 'INFO'
49
+ log_processor = dict(by_epoch=False)
50
+ metainfo = dict(
51
+ classes=(
52
+ 'background',
53
+ 'building',
54
+ ),
55
+ palette=[
56
+ [
57
+ 0,
58
+ 0,
59
+ 0,
60
+ ],
61
+ [
62
+ 255,
63
+ 255,
64
+ 255,
65
+ ],
66
+ ])
67
+ model = dict(
68
+ backbone=dict(
69
+ input_resolution=512,
70
+ layers=[
71
+ 3,
72
+ 4,
73
+ 6,
74
+ 3,
75
+ ],
76
+ output_dim=1024,
77
+ style='pytorch',
78
+ type='CLIPResNetWithAttention'),
79
+ context_decoder=dict(
80
+ context_length=16,
81
+ dropout=0.1,
82
+ outdim=1024,
83
+ style='pytorch',
84
+ transformer_heads=4,
85
+ transformer_layers=3,
86
+ transformer_width=256,
87
+ type='ContextDecoder',
88
+ visual_dim=1024),
89
+ context_length=64,
90
+ decode_head=dict(
91
+ align_corners=False,
92
+ channels=256,
93
+ dropout_ratio=0.1,
94
+ feature_strides=[
95
+ 4,
96
+ 8,
97
+ 16,
98
+ 32,
99
+ ],
100
+ in_channels=[
101
+ 256,
102
+ 256,
103
+ 256,
104
+ 256,
105
+ ],
106
+ in_index=[
107
+ 0,
108
+ 1,
109
+ 2,
110
+ 3,
111
+ ],
112
+ loss_decode=dict(
113
+ loss_weight=1.0, type='CrossEntropyLoss', use_sigmoid=False),
114
+ norm_cfg=dict(requires_grad=True, type='SyncBN'),
115
+ num_classes=2,
116
+ type='SwinTextDecode'),
117
+ neck=dict(
118
+ in_channels=[
119
+ 512,
120
+ 1024,
121
+ 2048,
122
+ 4100,
123
+ ],
124
+ num_outs=4,
125
+ out_channels=256,
126
+ type='FPN'),
127
+ pretrained='/data/RN50.pt',
128
+ test_cfg=dict(mode='whole'),
129
+ text_encoder=dict(
130
+ context_length=77,
131
+ embed_dim=1024,
132
+ style='pytorch',
133
+ transformer_heads=8,
134
+ transformer_layers=12,
135
+ transformer_width=512,
136
+ type='CLIPTextContextEncoder'),
137
+ text_head=False,
138
+ train_cfg=dict(),
139
+ type='AugCD')
140
+ norm_cfg = dict(requires_grad=True, type='SyncBN')
141
+ optim_wrapper = dict(
142
+ optimizer=dict(
143
+ betas=(
144
+ 0.9,
145
+ 0.999,
146
+ ), lr=0.0001, type='AdamW', weight_decay=0.01),
147
+ paramwise_cfg=dict(
148
+ custom_keys=dict(
149
+ absolute_pos_embed=dict(decay_mult=0.0),
150
+ norm=dict(decay_mult=0.0),
151
+ relative_position_bias_table=dict(decay_mult=0.0))),
152
+ type='OptimWrapper')
153
+ optimizer = dict(lr=0.01, momentum=0.9, type='SGD', weight_decay=0.0005)
154
+ param_scheduler = [
155
+ dict(
156
+ begin=0, by_epoch=False, end=1500, start_factor=1e-06,
157
+ type='LinearLR'),
158
+ dict(
159
+ begin=1500,
160
+ by_epoch=False,
161
+ end=30000,
162
+ eta_min=0.0,
163
+ power=1.0,
164
+ type='PolyLR'),
165
+ ]
166
+ resume = False
167
+ test_cfg = dict(type='TestLoop')
168
+ test_dataloader = dict(
169
+ batch_size=1,
170
+ dataset=dict(
171
+ ann_file='test.txt',
172
+ data_prefix=dict(
173
+ img_path='test/A', img_path2='test/B', seg_map_path='test/label'),
174
+ data_root='/data/Dataset/LEVIRCD',
175
+ metainfo=dict(
176
+ classes=(
177
+ 'background',
178
+ 'building',
179
+ ),
180
+ palette=[
181
+ [
182
+ 0,
183
+ 0,
184
+ 0,
185
+ ],
186
+ [
187
+ 255,
188
+ 255,
189
+ 255,
190
+ ],
191
+ ]),
192
+ pipeline=[
193
+ dict(type='LoadMultipleRSImageFromFile'),
194
+ dict(type='LoadAnnotations'),
195
+ dict(type='ConcatCDInput'),
196
+ dict(type='PackCDInputs'),
197
+ ],
198
+ type='TXTCDDatasetJSON'),
199
+ num_workers=4,
200
+ persistent_workers=True,
201
+ sampler=dict(shuffle=False, type='DefaultSampler'))
202
+ test_evaluator = dict(
203
+ iou_metrics=[
204
+ 'mIoU',
205
+ ], type='IoUMetric')
206
+ test_pipeline = [
207
+ dict(type='LoadMultipleRSImageFromFile'),
208
+ dict(type='LoadAnnotations'),
209
+ dict(type='ConcatCDInput'),
210
+ dict(type='PackCDInputs'),
211
+ ]
212
+ train_cfg = dict(max_iters=30000, type='IterBasedTrainLoop', val_interval=1000)
213
+ train_dataloader = dict(
214
+ batch_size=20,
215
+ dataset=dict(
216
+ ann_file='train.txt',
217
+ data_prefix=dict(
218
+ img_path='train/A',
219
+ img_path2='train/B',
220
+ seg_map_path='train/label'),
221
+ data_root='/data/Dataset/LEVIRCD',
222
+ metainfo=dict(
223
+ classes=(
224
+ 'background',
225
+ 'building',
226
+ ),
227
+ palette=[
228
+ [
229
+ 0,
230
+ 0,
231
+ 0,
232
+ ],
233
+ [
234
+ 255,
235
+ 255,
236
+ 255,
237
+ ],
238
+ ]),
239
+ pipeline=[
240
+ dict(type='LoadMultipleRSImageFromFile'),
241
+ dict(type='LoadAnnotations'),
242
+ dict(degree=180, prob=0.5, type='MultiImgRandomRotate'),
243
+ dict(
244
+ cat_max_ratio=0.75,
245
+ crop_size=(
246
+ 256,
247
+ 256,
248
+ ),
249
+ type='MultiImgRandomCrop'),
250
+ dict(direction='horizontal', prob=0.5, type='MultiImgRandomFlip'),
251
+ dict(direction='vertical', prob=0.5, type='MultiImgRandomFlip'),
252
+ dict(
253
+ brightness_delta=10,
254
+ contrast_range=(
255
+ 0.8,
256
+ 1.2,
257
+ ),
258
+ hue_delta=10,
259
+ saturation_range=(
260
+ 0.8,
261
+ 1.2,
262
+ ),
263
+ type='MultiImgPhotoMetricDistortion'),
264
+ dict(type='ConcatCDInput'),
265
+ dict(type='PackCDInputs'),
266
+ ],
267
+ type='TXTCDDatasetJSON'),
268
+ num_workers=8,
269
+ persistent_workers=True,
270
+ sampler=dict(shuffle=True, type='InfiniteSampler'))
271
+ train_pipeline = [
272
+ dict(type='LoadMultipleRSImageFromFile'),
273
+ dict(type='LoadAnnotations'),
274
+ dict(degree=180, prob=0.5, type='MultiImgRandomRotate'),
275
+ dict(
276
+ cat_max_ratio=0.75, crop_size=(
277
+ 256,
278
+ 256,
279
+ ), type='MultiImgRandomCrop'),
280
+ dict(direction='horizontal', prob=0.5, type='MultiImgRandomFlip'),
281
+ dict(direction='vertical', prob=0.5, type='MultiImgRandomFlip'),
282
+ dict(
283
+ brightness_delta=10,
284
+ contrast_range=(
285
+ 0.8,
286
+ 1.2,
287
+ ),
288
+ hue_delta=10,
289
+ saturation_range=(
290
+ 0.8,
291
+ 1.2,
292
+ ),
293
+ type='MultiImgPhotoMetricDistortion'),
294
+ dict(type='ConcatCDInput'),
295
+ dict(type='PackCDInputs'),
296
+ ]
297
+ tta_model = dict(type='SegTTAModel')
298
+ tta_pipeline = [
299
+ dict(type='LoadMultipleRSImageFromFile'),
300
+ dict(
301
+ transforms=[
302
+ [
303
+ dict(type='LoadAnnotations'),
304
+ ],
305
+ [
306
+ dict(type='ConcatCDInput'),
307
+ ],
308
+ [
309
+ dict(type='PackCDInputs'),
310
+ ],
311
+ ],
312
+ type='TestTimeAug'),
313
+ ]
314
+ val_cfg = dict(type='ValLoop')
315
+ val_dataloader = dict(
316
+ batch_size=1,
317
+ dataset=dict(
318
+ ann_file='val.txt',
319
+ data_prefix=dict(
320
+ img_path='test/A', img_path2='test/B', seg_map_path='test/label'),
321
+ data_root='/data/Dataset/LEVIRCD',
322
+ metainfo=dict(
323
+ classes=(
324
+ 'background',
325
+ 'building',
326
+ ),
327
+ palette=[
328
+ [
329
+ 0,
330
+ 0,
331
+ 0,
332
+ ],
333
+ [
334
+ 255,
335
+ 255,
336
+ 255,
337
+ ],
338
+ ]),
339
+ pipeline=[
340
+ dict(type='LoadMultipleRSImageFromFile'),
341
+ dict(type='LoadAnnotations'),
342
+ dict(type='ConcatCDInput'),
343
+ dict(type='PackCDInputs'),
344
+ ],
345
+ type='TXTCDDatasetJSON'),
346
+ num_workers=4,
347
+ persistent_workers=True,
348
+ sampler=dict(shuffle=False, type='DefaultSampler'))
349
+ val_evaluator = dict(
350
+ iou_metrics=[
351
+ 'mIoU',
352
+ ], type='IoUMetric')
353
+ vis_backends = [
354
+ dict(type='LocalVisBackend'),
355
+ ]
356
+ visualizer = dict(
357
+ name='visualizer',
358
+ type='SegLocalVisualizer',
359
+ vis_backends=[
360
+ dict(type='LocalVisBackend'),
361
+ ])
362
+ work_dir = '/data/work_dirs/AugCD_LEVIRCD_3'
best_weights/best_weights/MdaCD_LEVIRCD/20250403_131147/vis_data/scalars.json ADDED
The diff for this file is too large to render. See raw diff
 
best_weights/best_weights/MdaCD_LEVIRCD/AugCD_LEVIRCD.py ADDED
@@ -0,0 +1,362 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ crop_size = (
2
+ 256,
3
+ 256,
4
+ )
5
+ data_preprocessor = dict(
6
+ mean=[
7
+ 123.675,
8
+ 116.28,
9
+ 103.53,
10
+ 123.675,
11
+ 116.28,
12
+ 103.53,
13
+ ],
14
+ size=(
15
+ 256,
16
+ 256,
17
+ ),
18
+ std=[
19
+ 58.395,
20
+ 57.12,
21
+ 57.375,
22
+ 58.395,
23
+ 57.12,
24
+ 57.375,
25
+ ],
26
+ type='SegDataPreProcessor')
27
+ data_root = '/data/Dataset/LEVIRCD'
28
+ dataset_type = 'TXTCDDatasetJSON'
29
+ default_hooks = dict(
30
+ checkpoint=dict(
31
+ by_epoch=False,
32
+ interval=10000,
33
+ save_best='mIoU',
34
+ type='CheckpointHook'),
35
+ logger=dict(interval=10, log_metric_by_epoch=False, type='LoggerHook'),
36
+ param_scheduler=dict(type='ParamSchedulerHook'),
37
+ sampler_seed=dict(type='DistSamplerSeedHook'),
38
+ timer=dict(type='IterTimerHook'),
39
+ visualization=dict(type='SegVisualizationHook'))
40
+ default_scope = 'mmseg'
41
+ env_cfg = dict(
42
+ cudnn_benchmark=True,
43
+ dist_cfg=dict(backend='nccl'),
44
+ mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0))
45
+ find_unused_parameters = True
46
+ launcher = 'pytorch'
47
+ load_from = None
48
+ log_level = 'INFO'
49
+ log_processor = dict(by_epoch=False)
50
+ metainfo = dict(
51
+ classes=(
52
+ 'background',
53
+ 'building',
54
+ ),
55
+ palette=[
56
+ [
57
+ 0,
58
+ 0,
59
+ 0,
60
+ ],
61
+ [
62
+ 255,
63
+ 255,
64
+ 255,
65
+ ],
66
+ ])
67
+ model = dict(
68
+ backbone=dict(
69
+ input_resolution=512,
70
+ layers=[
71
+ 3,
72
+ 4,
73
+ 6,
74
+ 3,
75
+ ],
76
+ output_dim=1024,
77
+ style='pytorch',
78
+ type='CLIPResNetWithAttention'),
79
+ context_decoder=dict(
80
+ context_length=16,
81
+ dropout=0.1,
82
+ outdim=1024,
83
+ style='pytorch',
84
+ transformer_heads=4,
85
+ transformer_layers=3,
86
+ transformer_width=256,
87
+ type='ContextDecoder',
88
+ visual_dim=1024),
89
+ context_length=64,
90
+ decode_head=dict(
91
+ align_corners=False,
92
+ channels=256,
93
+ dropout_ratio=0.1,
94
+ feature_strides=[
95
+ 4,
96
+ 8,
97
+ 16,
98
+ 32,
99
+ ],
100
+ in_channels=[
101
+ 256,
102
+ 256,
103
+ 256,
104
+ 256,
105
+ ],
106
+ in_index=[
107
+ 0,
108
+ 1,
109
+ 2,
110
+ 3,
111
+ ],
112
+ loss_decode=dict(
113
+ loss_weight=1.0, type='CrossEntropyLoss', use_sigmoid=False),
114
+ norm_cfg=dict(requires_grad=True, type='SyncBN'),
115
+ num_classes=2,
116
+ type='SwinTextDecode'),
117
+ neck=dict(
118
+ in_channels=[
119
+ 512,
120
+ 1024,
121
+ 2048,
122
+ 4100,
123
+ ],
124
+ num_outs=4,
125
+ out_channels=256,
126
+ type='FPN'),
127
+ pretrained='/data/RN50.pt',
128
+ test_cfg=dict(mode='whole'),
129
+ text_encoder=dict(
130
+ context_length=77,
131
+ embed_dim=1024,
132
+ style='pytorch',
133
+ transformer_heads=8,
134
+ transformer_layers=12,
135
+ transformer_width=512,
136
+ type='CLIPTextContextEncoder'),
137
+ text_head=False,
138
+ train_cfg=dict(),
139
+ type='AugCD')
140
+ norm_cfg = dict(requires_grad=True, type='SyncBN')
141
+ optim_wrapper = dict(
142
+ optimizer=dict(
143
+ betas=(
144
+ 0.9,
145
+ 0.999,
146
+ ), lr=0.0001, type='AdamW', weight_decay=0.01),
147
+ paramwise_cfg=dict(
148
+ custom_keys=dict(
149
+ absolute_pos_embed=dict(decay_mult=0.0),
150
+ norm=dict(decay_mult=0.0),
151
+ relative_position_bias_table=dict(decay_mult=0.0))),
152
+ type='OptimWrapper')
153
+ optimizer = dict(lr=0.01, momentum=0.9, type='SGD', weight_decay=0.0005)
154
+ param_scheduler = [
155
+ dict(
156
+ begin=0, by_epoch=False, end=1500, start_factor=1e-06,
157
+ type='LinearLR'),
158
+ dict(
159
+ begin=1500,
160
+ by_epoch=False,
161
+ end=30000,
162
+ eta_min=0.0,
163
+ power=1.0,
164
+ type='PolyLR'),
165
+ ]
166
+ resume = False
167
+ test_cfg = dict(type='TestLoop')
168
+ test_dataloader = dict(
169
+ batch_size=1,
170
+ dataset=dict(
171
+ ann_file='test.txt',
172
+ data_prefix=dict(
173
+ img_path='test/A', img_path2='test/B', seg_map_path='test/label'),
174
+ data_root='/data/Dataset/LEVIRCD',
175
+ metainfo=dict(
176
+ classes=(
177
+ 'background',
178
+ 'building',
179
+ ),
180
+ palette=[
181
+ [
182
+ 0,
183
+ 0,
184
+ 0,
185
+ ],
186
+ [
187
+ 255,
188
+ 255,
189
+ 255,
190
+ ],
191
+ ]),
192
+ pipeline=[
193
+ dict(type='LoadMultipleRSImageFromFile'),
194
+ dict(type='LoadAnnotations'),
195
+ dict(type='ConcatCDInput'),
196
+ dict(type='PackCDInputs'),
197
+ ],
198
+ type='TXTCDDatasetJSON'),
199
+ num_workers=4,
200
+ persistent_workers=True,
201
+ sampler=dict(shuffle=False, type='DefaultSampler'))
202
+ test_evaluator = dict(
203
+ iou_metrics=[
204
+ 'mIoU',
205
+ ], type='IoUMetric')
206
+ test_pipeline = [
207
+ dict(type='LoadMultipleRSImageFromFile'),
208
+ dict(type='LoadAnnotations'),
209
+ dict(type='ConcatCDInput'),
210
+ dict(type='PackCDInputs'),
211
+ ]
212
+ train_cfg = dict(max_iters=30000, type='IterBasedTrainLoop', val_interval=1000)
213
+ train_dataloader = dict(
214
+ batch_size=20,
215
+ dataset=dict(
216
+ ann_file='train.txt',
217
+ data_prefix=dict(
218
+ img_path='train/A',
219
+ img_path2='train/B',
220
+ seg_map_path='train/label'),
221
+ data_root='/data/Dataset/LEVIRCD',
222
+ metainfo=dict(
223
+ classes=(
224
+ 'background',
225
+ 'building',
226
+ ),
227
+ palette=[
228
+ [
229
+ 0,
230
+ 0,
231
+ 0,
232
+ ],
233
+ [
234
+ 255,
235
+ 255,
236
+ 255,
237
+ ],
238
+ ]),
239
+ pipeline=[
240
+ dict(type='LoadMultipleRSImageFromFile'),
241
+ dict(type='LoadAnnotations'),
242
+ dict(degree=180, prob=0.5, type='MultiImgRandomRotate'),
243
+ dict(
244
+ cat_max_ratio=0.75,
245
+ crop_size=(
246
+ 256,
247
+ 256,
248
+ ),
249
+ type='MultiImgRandomCrop'),
250
+ dict(direction='horizontal', prob=0.5, type='MultiImgRandomFlip'),
251
+ dict(direction='vertical', prob=0.5, type='MultiImgRandomFlip'),
252
+ dict(
253
+ brightness_delta=10,
254
+ contrast_range=(
255
+ 0.8,
256
+ 1.2,
257
+ ),
258
+ hue_delta=10,
259
+ saturation_range=(
260
+ 0.8,
261
+ 1.2,
262
+ ),
263
+ type='MultiImgPhotoMetricDistortion'),
264
+ dict(type='ConcatCDInput'),
265
+ dict(type='PackCDInputs'),
266
+ ],
267
+ type='TXTCDDatasetJSON'),
268
+ num_workers=8,
269
+ persistent_workers=True,
270
+ sampler=dict(shuffle=True, type='InfiniteSampler'))
271
+ train_pipeline = [
272
+ dict(type='LoadMultipleRSImageFromFile'),
273
+ dict(type='LoadAnnotations'),
274
+ dict(degree=180, prob=0.5, type='MultiImgRandomRotate'),
275
+ dict(
276
+ cat_max_ratio=0.75, crop_size=(
277
+ 256,
278
+ 256,
279
+ ), type='MultiImgRandomCrop'),
280
+ dict(direction='horizontal', prob=0.5, type='MultiImgRandomFlip'),
281
+ dict(direction='vertical', prob=0.5, type='MultiImgRandomFlip'),
282
+ dict(
283
+ brightness_delta=10,
284
+ contrast_range=(
285
+ 0.8,
286
+ 1.2,
287
+ ),
288
+ hue_delta=10,
289
+ saturation_range=(
290
+ 0.8,
291
+ 1.2,
292
+ ),
293
+ type='MultiImgPhotoMetricDistortion'),
294
+ dict(type='ConcatCDInput'),
295
+ dict(type='PackCDInputs'),
296
+ ]
297
+ tta_model = dict(type='SegTTAModel')
298
+ tta_pipeline = [
299
+ dict(type='LoadMultipleRSImageFromFile'),
300
+ dict(
301
+ transforms=[
302
+ [
303
+ dict(type='LoadAnnotations'),
304
+ ],
305
+ [
306
+ dict(type='ConcatCDInput'),
307
+ ],
308
+ [
309
+ dict(type='PackCDInputs'),
310
+ ],
311
+ ],
312
+ type='TestTimeAug'),
313
+ ]
314
+ val_cfg = dict(type='ValLoop')
315
+ val_dataloader = dict(
316
+ batch_size=1,
317
+ dataset=dict(
318
+ ann_file='val.txt',
319
+ data_prefix=dict(
320
+ img_path='test/A', img_path2='test/B', seg_map_path='test/label'),
321
+ data_root='/data/Dataset/LEVIRCD',
322
+ metainfo=dict(
323
+ classes=(
324
+ 'background',
325
+ 'building',
326
+ ),
327
+ palette=[
328
+ [
329
+ 0,
330
+ 0,
331
+ 0,
332
+ ],
333
+ [
334
+ 255,
335
+ 255,
336
+ 255,
337
+ ],
338
+ ]),
339
+ pipeline=[
340
+ dict(type='LoadMultipleRSImageFromFile'),
341
+ dict(type='LoadAnnotations'),
342
+ dict(type='ConcatCDInput'),
343
+ dict(type='PackCDInputs'),
344
+ ],
345
+ type='TXTCDDatasetJSON'),
346
+ num_workers=4,
347
+ persistent_workers=True,
348
+ sampler=dict(shuffle=False, type='DefaultSampler'))
349
+ val_evaluator = dict(
350
+ iou_metrics=[
351
+ 'mIoU',
352
+ ], type='IoUMetric')
353
+ vis_backends = [
354
+ dict(type='LocalVisBackend'),
355
+ ]
356
+ visualizer = dict(
357
+ name='visualizer',
358
+ type='SegLocalVisualizer',
359
+ vis_backends=[
360
+ dict(type='LocalVisBackend'),
361
+ ])
362
+ work_dir = '/data/work_dirs/AugCD_LEVIRCD_3'
best_weights/best_weights/MdaCD_LEVIRCD/best_mIoU_iter_20000.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5135ac21f8261974d1c658717844d1fad0b333b25f36eaf947d4912c28f6d227
3
+ size 474216637
best_weights/best_weights/MdaCD_LEVIRCD/last_checkpoint ADDED
@@ -0,0 +1 @@
 
 
1
+ /data/work_dirs/AugCD_LEVIRCD_3/iter_30000.pth
best_weights/best_weights/MdaCD_LEVIRCD/overall_result.csv ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ OverallAccuracy,MeanF1,MeanIoU,ClassIoU,ClassF1,ClassRecall,ClassPrecision
2
+ 99.18,95.7,92.01,84.88,91.82,90.09,93.62
best_weights/best_weights/MdaCD_SYSUCD/20250823_154932/20250823_154932.log ADDED
The diff for this file is too large to render. See raw diff
 
best_weights/best_weights/MdaCD_SYSUCD/20250823_154932/vis_data/20250823_154932.json ADDED
The diff for this file is too large to render. See raw diff
 
best_weights/best_weights/MdaCD_SYSUCD/20250823_154932/vis_data/config.py ADDED
@@ -0,0 +1,362 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ crop_size = (
2
+ 256,
3
+ 256,
4
+ )
5
+ data_preprocessor = dict(
6
+ mean=[
7
+ 123.675,
8
+ 116.28,
9
+ 103.53,
10
+ 123.675,
11
+ 116.28,
12
+ 103.53,
13
+ ],
14
+ size=(
15
+ 256,
16
+ 256,
17
+ ),
18
+ std=[
19
+ 58.395,
20
+ 57.12,
21
+ 57.375,
22
+ 58.395,
23
+ 57.12,
24
+ 57.375,
25
+ ],
26
+ type='SegDataPreProcessor')
27
+ data_root = '/home/haoyang/data/data/MdaCD/Dataset/SYSUCD'
28
+ dataset_type = 'TXTCDDatasetJSON'
29
+ default_hooks = dict(
30
+ checkpoint=dict(
31
+ by_epoch=False,
32
+ interval=10000,
33
+ save_best='mIoU',
34
+ type='CheckpointHook'),
35
+ logger=dict(interval=10, log_metric_by_epoch=False, type='LoggerHook'),
36
+ param_scheduler=dict(type='ParamSchedulerHook'),
37
+ sampler_seed=dict(type='DistSamplerSeedHook'),
38
+ timer=dict(type='IterTimerHook'),
39
+ visualization=dict(type='SegVisualizationHook'))
40
+ default_scope = 'mmseg'
41
+ env_cfg = dict(
42
+ cudnn_benchmark=True,
43
+ dist_cfg=dict(backend='nccl'),
44
+ mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0))
45
+ find_unused_parameters = True
46
+ launcher = 'pytorch'
47
+ load_from = None
48
+ log_level = 'INFO'
49
+ log_processor = dict(by_epoch=False)
50
+ metainfo = dict(
51
+ classes=(
52
+ 'background',
53
+ 'building',
54
+ ),
55
+ palette=[
56
+ [
57
+ 0,
58
+ 0,
59
+ 0,
60
+ ],
61
+ [
62
+ 255,
63
+ 255,
64
+ 255,
65
+ ],
66
+ ])
67
+ model = dict(
68
+ backbone=dict(
69
+ input_resolution=512,
70
+ layers=[
71
+ 3,
72
+ 4,
73
+ 6,
74
+ 3,
75
+ ],
76
+ output_dim=1024,
77
+ style='pytorch',
78
+ type='CLIPResNetWithAttention'),
79
+ context_decoder=dict(
80
+ context_length=16,
81
+ dropout=0.1,
82
+ outdim=1024,
83
+ style='pytorch',
84
+ transformer_heads=4,
85
+ transformer_layers=3,
86
+ transformer_width=256,
87
+ type='ContextDecoder',
88
+ visual_dim=1024),
89
+ context_length=64,
90
+ decode_head=dict(
91
+ align_corners=False,
92
+ channels=256,
93
+ dropout_ratio=0.1,
94
+ feature_strides=[
95
+ 4,
96
+ 8,
97
+ 16,
98
+ 32,
99
+ ],
100
+ in_channels=[
101
+ 256,
102
+ 256,
103
+ 256,
104
+ 256,
105
+ ],
106
+ in_index=[
107
+ 0,
108
+ 1,
109
+ 2,
110
+ 3,
111
+ ],
112
+ loss_decode=dict(
113
+ loss_weight=1.0, type='CrossEntropyLoss', use_sigmoid=False),
114
+ norm_cfg=dict(requires_grad=True, type='BN'),
115
+ num_classes=2,
116
+ type='SwinTextDecode'),
117
+ neck=dict(
118
+ in_channels=[
119
+ 512,
120
+ 1024,
121
+ 2048,
122
+ 4100,
123
+ ],
124
+ num_outs=4,
125
+ out_channels=256,
126
+ type='FPN'),
127
+ pretrained='/home/haoyang/data/data/MdaCD/RN50.pt',
128
+ test_cfg=dict(mode='whole'),
129
+ text_encoder=dict(
130
+ context_length=77,
131
+ embed_dim=1024,
132
+ style='pytorch',
133
+ transformer_heads=8,
134
+ transformer_layers=12,
135
+ transformer_width=512,
136
+ type='CLIPTextContextEncoder'),
137
+ text_head=False,
138
+ train_cfg=dict(),
139
+ type='AugCD')
140
+ norm_cfg = dict(requires_grad=True, type='BN')
141
+ optim_wrapper = dict(
142
+ optimizer=dict(
143
+ betas=(
144
+ 0.9,
145
+ 0.999,
146
+ ), lr=3e-05, type='AdamW', weight_decay=0.01),
147
+ paramwise_cfg=dict(
148
+ custom_keys=dict(
149
+ absolute_pos_embed=dict(decay_mult=0.0),
150
+ norm=dict(decay_mult=0.0),
151
+ relative_position_bias_table=dict(decay_mult=0.0))),
152
+ type='OptimWrapper')
153
+ optimizer = dict(lr=0.01, momentum=0.9, type='SGD', weight_decay=0.0005)
154
+ param_scheduler = [
155
+ dict(
156
+ begin=0, by_epoch=False, end=1500, start_factor=1e-06,
157
+ type='LinearLR'),
158
+ dict(
159
+ begin=1500,
160
+ by_epoch=False,
161
+ end=10000,
162
+ eta_min=0.0,
163
+ power=1.0,
164
+ type='PolyLR'),
165
+ ]
166
+ resume = False
167
+ test_cfg = dict(type='TestLoop')
168
+ test_dataloader = dict(
169
+ batch_size=1,
170
+ dataset=dict(
171
+ ann_file='test.txt',
172
+ data_prefix=dict(
173
+ img_path='test/A', img_path2='test/B', seg_map_path='test/label'),
174
+ data_root='/home/haoyang/data/data/MdaCD/Dataset/SYSUCD',
175
+ metainfo=dict(
176
+ classes=(
177
+ 'background',
178
+ 'building',
179
+ ),
180
+ palette=[
181
+ [
182
+ 0,
183
+ 0,
184
+ 0,
185
+ ],
186
+ [
187
+ 255,
188
+ 255,
189
+ 255,
190
+ ],
191
+ ]),
192
+ pipeline=[
193
+ dict(type='LoadMultipleRSImageFromFile'),
194
+ dict(type='LoadAnnotations'),
195
+ dict(type='ConcatCDInput'),
196
+ dict(type='PackCDInputs'),
197
+ ],
198
+ type='TXTCDDatasetJSON'),
199
+ num_workers=4,
200
+ persistent_workers=True,
201
+ sampler=dict(shuffle=False, type='DefaultSampler'))
202
+ test_evaluator = dict(
203
+ iou_metrics=[
204
+ 'mIoU',
205
+ ], type='IoUMetric')
206
+ test_pipeline = [
207
+ dict(type='LoadMultipleRSImageFromFile'),
208
+ dict(type='LoadAnnotations'),
209
+ dict(type='ConcatCDInput'),
210
+ dict(type='PackCDInputs'),
211
+ ]
212
+ train_cfg = dict(max_iters=10000, type='IterBasedTrainLoop', val_interval=500)
213
+ train_dataloader = dict(
214
+ batch_size=20,
215
+ dataset=dict(
216
+ ann_file='train.txt',
217
+ data_prefix=dict(
218
+ img_path='train/A',
219
+ img_path2='train/B',
220
+ seg_map_path='train/label'),
221
+ data_root='/home/haoyang/data/data/MdaCD/Dataset/SYSUCD',
222
+ metainfo=dict(
223
+ classes=(
224
+ 'background',
225
+ 'building',
226
+ ),
227
+ palette=[
228
+ [
229
+ 0,
230
+ 0,
231
+ 0,
232
+ ],
233
+ [
234
+ 255,
235
+ 255,
236
+ 255,
237
+ ],
238
+ ]),
239
+ pipeline=[
240
+ dict(type='LoadMultipleRSImageFromFile'),
241
+ dict(type='LoadAnnotations'),
242
+ dict(degree=180, prob=0.5, type='MultiImgRandomRotate'),
243
+ dict(
244
+ cat_max_ratio=0.75,
245
+ crop_size=(
246
+ 256,
247
+ 256,
248
+ ),
249
+ type='MultiImgRandomCrop'),
250
+ dict(direction='horizontal', prob=0.5, type='MultiImgRandomFlip'),
251
+ dict(direction='vertical', prob=0.5, type='MultiImgRandomFlip'),
252
+ dict(
253
+ brightness_delta=10,
254
+ contrast_range=(
255
+ 0.8,
256
+ 1.2,
257
+ ),
258
+ hue_delta=10,
259
+ saturation_range=(
260
+ 0.8,
261
+ 1.2,
262
+ ),
263
+ type='MultiImgPhotoMetricDistortion'),
264
+ dict(type='ConcatCDInput'),
265
+ dict(type='PackCDInputs'),
266
+ ],
267
+ type='TXTCDDatasetJSON'),
268
+ num_workers=8,
269
+ persistent_workers=True,
270
+ sampler=dict(shuffle=True, type='InfiniteSampler'))
271
+ train_pipeline = [
272
+ dict(type='LoadMultipleRSImageFromFile'),
273
+ dict(type='LoadAnnotations'),
274
+ dict(degree=180, prob=0.5, type='MultiImgRandomRotate'),
275
+ dict(
276
+ cat_max_ratio=0.75, crop_size=(
277
+ 256,
278
+ 256,
279
+ ), type='MultiImgRandomCrop'),
280
+ dict(direction='horizontal', prob=0.5, type='MultiImgRandomFlip'),
281
+ dict(direction='vertical', prob=0.5, type='MultiImgRandomFlip'),
282
+ dict(
283
+ brightness_delta=10,
284
+ contrast_range=(
285
+ 0.8,
286
+ 1.2,
287
+ ),
288
+ hue_delta=10,
289
+ saturation_range=(
290
+ 0.8,
291
+ 1.2,
292
+ ),
293
+ type='MultiImgPhotoMetricDistortion'),
294
+ dict(type='ConcatCDInput'),
295
+ dict(type='PackCDInputs'),
296
+ ]
297
+ tta_model = dict(type='SegTTAModel')
298
+ tta_pipeline = [
299
+ dict(type='LoadMultipleRSImageFromFile'),
300
+ dict(
301
+ transforms=[
302
+ [
303
+ dict(type='LoadAnnotations'),
304
+ ],
305
+ [
306
+ dict(type='ConcatCDInput'),
307
+ ],
308
+ [
309
+ dict(type='PackCDInputs'),
310
+ ],
311
+ ],
312
+ type='TestTimeAug'),
313
+ ]
314
+ val_cfg = dict(type='ValLoop')
315
+ val_dataloader = dict(
316
+ batch_size=1,
317
+ dataset=dict(
318
+ ann_file='val.txt',
319
+ data_prefix=dict(
320
+ img_path='test/A', img_path2='test/B', seg_map_path='test/label'),
321
+ data_root='/home/haoyang/data/data/MdaCD/Dataset/SYSUCD',
322
+ metainfo=dict(
323
+ classes=(
324
+ 'background',
325
+ 'building',
326
+ ),
327
+ palette=[
328
+ [
329
+ 0,
330
+ 0,
331
+ 0,
332
+ ],
333
+ [
334
+ 255,
335
+ 255,
336
+ 255,
337
+ ],
338
+ ]),
339
+ pipeline=[
340
+ dict(type='LoadMultipleRSImageFromFile'),
341
+ dict(type='LoadAnnotations'),
342
+ dict(type='ConcatCDInput'),
343
+ dict(type='PackCDInputs'),
344
+ ],
345
+ type='TXTCDDatasetJSON'),
346
+ num_workers=4,
347
+ persistent_workers=True,
348
+ sampler=dict(shuffle=False, type='DefaultSampler'))
349
+ val_evaluator = dict(
350
+ iou_metrics=[
351
+ 'mIoU',
352
+ ], type='IoUMetric')
353
+ vis_backends = [
354
+ dict(type='LocalVisBackend'),
355
+ ]
356
+ visualizer = dict(
357
+ name='visualizer',
358
+ type='SegLocalVisualizer',
359
+ vis_backends=[
360
+ dict(type='LocalVisBackend'),
361
+ ])
362
+ work_dir = '/home/haoyang/data/data/MdaCD/work_dirs/MdaCD_SYSUCD_4'
best_weights/best_weights/MdaCD_SYSUCD/20250823_154932/vis_data/scalars.json ADDED
The diff for this file is too large to render. See raw diff
 
best_weights/best_weights/MdaCD_SYSUCD/MdaCD_SYSUCD.py ADDED
@@ -0,0 +1,362 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ crop_size = (
2
+ 256,
3
+ 256,
4
+ )
5
+ data_preprocessor = dict(
6
+ mean=[
7
+ 123.675,
8
+ 116.28,
9
+ 103.53,
10
+ 123.675,
11
+ 116.28,
12
+ 103.53,
13
+ ],
14
+ size=(
15
+ 256,
16
+ 256,
17
+ ),
18
+ std=[
19
+ 58.395,
20
+ 57.12,
21
+ 57.375,
22
+ 58.395,
23
+ 57.12,
24
+ 57.375,
25
+ ],
26
+ type='SegDataPreProcessor')
27
+ data_root = '/home/haoyang/data/data/MdaCD/Dataset/SYSUCD'
28
+ dataset_type = 'TXTCDDatasetJSON'
29
+ default_hooks = dict(
30
+ checkpoint=dict(
31
+ by_epoch=False,
32
+ interval=10000,
33
+ save_best='mIoU',
34
+ type='CheckpointHook'),
35
+ logger=dict(interval=10, log_metric_by_epoch=False, type='LoggerHook'),
36
+ param_scheduler=dict(type='ParamSchedulerHook'),
37
+ sampler_seed=dict(type='DistSamplerSeedHook'),
38
+ timer=dict(type='IterTimerHook'),
39
+ visualization=dict(type='SegVisualizationHook'))
40
+ default_scope = 'mmseg'
41
+ env_cfg = dict(
42
+ cudnn_benchmark=True,
43
+ dist_cfg=dict(backend='nccl'),
44
+ mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0))
45
+ find_unused_parameters = True
46
+ launcher = 'pytorch'
47
+ load_from = None
48
+ log_level = 'INFO'
49
+ log_processor = dict(by_epoch=False)
50
+ metainfo = dict(
51
+ classes=(
52
+ 'background',
53
+ 'building',
54
+ ),
55
+ palette=[
56
+ [
57
+ 0,
58
+ 0,
59
+ 0,
60
+ ],
61
+ [
62
+ 255,
63
+ 255,
64
+ 255,
65
+ ],
66
+ ])
67
+ model = dict(
68
+ backbone=dict(
69
+ input_resolution=512,
70
+ layers=[
71
+ 3,
72
+ 4,
73
+ 6,
74
+ 3,
75
+ ],
76
+ output_dim=1024,
77
+ style='pytorch',
78
+ type='CLIPResNetWithAttention'),
79
+ context_decoder=dict(
80
+ context_length=16,
81
+ dropout=0.1,
82
+ outdim=1024,
83
+ style='pytorch',
84
+ transformer_heads=4,
85
+ transformer_layers=3,
86
+ transformer_width=256,
87
+ type='ContextDecoder',
88
+ visual_dim=1024),
89
+ context_length=64,
90
+ decode_head=dict(
91
+ align_corners=False,
92
+ channels=256,
93
+ dropout_ratio=0.1,
94
+ feature_strides=[
95
+ 4,
96
+ 8,
97
+ 16,
98
+ 32,
99
+ ],
100
+ in_channels=[
101
+ 256,
102
+ 256,
103
+ 256,
104
+ 256,
105
+ ],
106
+ in_index=[
107
+ 0,
108
+ 1,
109
+ 2,
110
+ 3,
111
+ ],
112
+ loss_decode=dict(
113
+ loss_weight=1.0, type='CrossEntropyLoss', use_sigmoid=False),
114
+ norm_cfg=dict(requires_grad=True, type='BN'),
115
+ num_classes=2,
116
+ type='SwinTextDecode'),
117
+ neck=dict(
118
+ in_channels=[
119
+ 512,
120
+ 1024,
121
+ 2048,
122
+ 4100,
123
+ ],
124
+ num_outs=4,
125
+ out_channels=256,
126
+ type='FPN'),
127
+ pretrained='/home/haoyang/data/data/MdaCD/RN50.pt',
128
+ test_cfg=dict(mode='whole'),
129
+ text_encoder=dict(
130
+ context_length=77,
131
+ embed_dim=1024,
132
+ style='pytorch',
133
+ transformer_heads=8,
134
+ transformer_layers=12,
135
+ transformer_width=512,
136
+ type='CLIPTextContextEncoder'),
137
+ text_head=False,
138
+ train_cfg=dict(),
139
+ type='AugCD')
140
+ norm_cfg = dict(requires_grad=True, type='BN')
141
+ optim_wrapper = dict(
142
+ optimizer=dict(
143
+ betas=(
144
+ 0.9,
145
+ 0.999,
146
+ ), lr=3e-05, type='AdamW', weight_decay=0.01),
147
+ paramwise_cfg=dict(
148
+ custom_keys=dict(
149
+ absolute_pos_embed=dict(decay_mult=0.0),
150
+ norm=dict(decay_mult=0.0),
151
+ relative_position_bias_table=dict(decay_mult=0.0))),
152
+ type='OptimWrapper')
153
+ optimizer = dict(lr=0.01, momentum=0.9, type='SGD', weight_decay=0.0005)
154
+ param_scheduler = [
155
+ dict(
156
+ begin=0, by_epoch=False, end=1500, start_factor=1e-06,
157
+ type='LinearLR'),
158
+ dict(
159
+ begin=1500,
160
+ by_epoch=False,
161
+ end=10000,
162
+ eta_min=0.0,
163
+ power=1.0,
164
+ type='PolyLR'),
165
+ ]
166
+ resume = False
167
+ test_cfg = dict(type='TestLoop')
168
+ test_dataloader = dict(
169
+ batch_size=1,
170
+ dataset=dict(
171
+ ann_file='test.txt',
172
+ data_prefix=dict(
173
+ img_path='test/A', img_path2='test/B', seg_map_path='test/label'),
174
+ data_root='/home/haoyang/data/data/MdaCD/Dataset/SYSUCD',
175
+ metainfo=dict(
176
+ classes=(
177
+ 'background',
178
+ 'building',
179
+ ),
180
+ palette=[
181
+ [
182
+ 0,
183
+ 0,
184
+ 0,
185
+ ],
186
+ [
187
+ 255,
188
+ 255,
189
+ 255,
190
+ ],
191
+ ]),
192
+ pipeline=[
193
+ dict(type='LoadMultipleRSImageFromFile'),
194
+ dict(type='LoadAnnotations'),
195
+ dict(type='ConcatCDInput'),
196
+ dict(type='PackCDInputs'),
197
+ ],
198
+ type='TXTCDDatasetJSON'),
199
+ num_workers=4,
200
+ persistent_workers=True,
201
+ sampler=dict(shuffle=False, type='DefaultSampler'))
202
+ test_evaluator = dict(
203
+ iou_metrics=[
204
+ 'mIoU',
205
+ ], type='IoUMetric')
206
+ test_pipeline = [
207
+ dict(type='LoadMultipleRSImageFromFile'),
208
+ dict(type='LoadAnnotations'),
209
+ dict(type='ConcatCDInput'),
210
+ dict(type='PackCDInputs'),
211
+ ]
212
+ train_cfg = dict(max_iters=10000, type='IterBasedTrainLoop', val_interval=500)
213
+ train_dataloader = dict(
214
+ batch_size=20,
215
+ dataset=dict(
216
+ ann_file='train.txt',
217
+ data_prefix=dict(
218
+ img_path='train/A',
219
+ img_path2='train/B',
220
+ seg_map_path='train/label'),
221
+ data_root='/home/haoyang/data/data/MdaCD/Dataset/SYSUCD',
222
+ metainfo=dict(
223
+ classes=(
224
+ 'background',
225
+ 'building',
226
+ ),
227
+ palette=[
228
+ [
229
+ 0,
230
+ 0,
231
+ 0,
232
+ ],
233
+ [
234
+ 255,
235
+ 255,
236
+ 255,
237
+ ],
238
+ ]),
239
+ pipeline=[
240
+ dict(type='LoadMultipleRSImageFromFile'),
241
+ dict(type='LoadAnnotations'),
242
+ dict(degree=180, prob=0.5, type='MultiImgRandomRotate'),
243
+ dict(
244
+ cat_max_ratio=0.75,
245
+ crop_size=(
246
+ 256,
247
+ 256,
248
+ ),
249
+ type='MultiImgRandomCrop'),
250
+ dict(direction='horizontal', prob=0.5, type='MultiImgRandomFlip'),
251
+ dict(direction='vertical', prob=0.5, type='MultiImgRandomFlip'),
252
+ dict(
253
+ brightness_delta=10,
254
+ contrast_range=(
255
+ 0.8,
256
+ 1.2,
257
+ ),
258
+ hue_delta=10,
259
+ saturation_range=(
260
+ 0.8,
261
+ 1.2,
262
+ ),
263
+ type='MultiImgPhotoMetricDistortion'),
264
+ dict(type='ConcatCDInput'),
265
+ dict(type='PackCDInputs'),
266
+ ],
267
+ type='TXTCDDatasetJSON'),
268
+ num_workers=8,
269
+ persistent_workers=True,
270
+ sampler=dict(shuffle=True, type='InfiniteSampler'))
271
+ train_pipeline = [
272
+ dict(type='LoadMultipleRSImageFromFile'),
273
+ dict(type='LoadAnnotations'),
274
+ dict(degree=180, prob=0.5, type='MultiImgRandomRotate'),
275
+ dict(
276
+ cat_max_ratio=0.75, crop_size=(
277
+ 256,
278
+ 256,
279
+ ), type='MultiImgRandomCrop'),
280
+ dict(direction='horizontal', prob=0.5, type='MultiImgRandomFlip'),
281
+ dict(direction='vertical', prob=0.5, type='MultiImgRandomFlip'),
282
+ dict(
283
+ brightness_delta=10,
284
+ contrast_range=(
285
+ 0.8,
286
+ 1.2,
287
+ ),
288
+ hue_delta=10,
289
+ saturation_range=(
290
+ 0.8,
291
+ 1.2,
292
+ ),
293
+ type='MultiImgPhotoMetricDistortion'),
294
+ dict(type='ConcatCDInput'),
295
+ dict(type='PackCDInputs'),
296
+ ]
297
+ tta_model = dict(type='SegTTAModel')
298
+ tta_pipeline = [
299
+ dict(type='LoadMultipleRSImageFromFile'),
300
+ dict(
301
+ transforms=[
302
+ [
303
+ dict(type='LoadAnnotations'),
304
+ ],
305
+ [
306
+ dict(type='ConcatCDInput'),
307
+ ],
308
+ [
309
+ dict(type='PackCDInputs'),
310
+ ],
311
+ ],
312
+ type='TestTimeAug'),
313
+ ]
314
+ val_cfg = dict(type='ValLoop')
315
+ val_dataloader = dict(
316
+ batch_size=1,
317
+ dataset=dict(
318
+ ann_file='val.txt',
319
+ data_prefix=dict(
320
+ img_path='test/A', img_path2='test/B', seg_map_path='test/label'),
321
+ data_root='/home/haoyang/data/data/MdaCD/Dataset/SYSUCD',
322
+ metainfo=dict(
323
+ classes=(
324
+ 'background',
325
+ 'building',
326
+ ),
327
+ palette=[
328
+ [
329
+ 0,
330
+ 0,
331
+ 0,
332
+ ],
333
+ [
334
+ 255,
335
+ 255,
336
+ 255,
337
+ ],
338
+ ]),
339
+ pipeline=[
340
+ dict(type='LoadMultipleRSImageFromFile'),
341
+ dict(type='LoadAnnotations'),
342
+ dict(type='ConcatCDInput'),
343
+ dict(type='PackCDInputs'),
344
+ ],
345
+ type='TXTCDDatasetJSON'),
346
+ num_workers=4,
347
+ persistent_workers=True,
348
+ sampler=dict(shuffle=False, type='DefaultSampler'))
349
+ val_evaluator = dict(
350
+ iou_metrics=[
351
+ 'mIoU',
352
+ ], type='IoUMetric')
353
+ vis_backends = [
354
+ dict(type='LocalVisBackend'),
355
+ ]
356
+ visualizer = dict(
357
+ name='visualizer',
358
+ type='SegLocalVisualizer',
359
+ vis_backends=[
360
+ dict(type='LocalVisBackend'),
361
+ ])
362
+ work_dir = '/home/haoyang/data/data/MdaCD/work_dirs/MdaCD_SYSUCD_4'
best_weights/best_weights/MdaCD_SYSUCD/best_mIoU_iter_7000.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6dc18bed199b6639c85031d5e733df8eeb55ab1228601502f792acde02428b7c
3
+ size 474475325
best_weights/best_weights/MdaCD_SYSUCD/iter_10000.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8e9f3c2dd72ff0d3cb9c512129ceec2fdc8928b550a0da71c9bae3a4bd163969
3
+ size 1416057321
best_weights/best_weights/MdaCD_SYSUCD/last_checkpoint ADDED
@@ -0,0 +1 @@
 
 
1
+ /home/haoyang/data/data/MdaCD/work_dirs/MdaCD_SYSUCD_4/iter_10000.pth
best_weights/best_weights/MdaCD_SYSUCD/overall_result.csv ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ OverallAccuracy,MeanF1,MeanIoU,ClassIoU,ClassF1,ClassRecall,ClassPrecision
2
+ 92.45,89.39,81.3,71.96,83.7,82.2,85.25
best_weights/best_weights/MdaCD_SYSUCD/test_result/00000.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00001.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00002.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00003.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00004.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00005.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00006.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00007.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00008.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00009.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00010.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00011.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00012.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00013.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00014.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00015.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00016.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00017.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00018.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00019.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00020.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00021.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00022.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00023.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00024.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00025.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00026.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00027.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00028.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00029.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00030.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00031.png ADDED
best_weights/best_weights/MdaCD_SYSUCD/test_result/00032.png ADDED