Hugging Face's logo

pubmed-ophtha
/
detection-models

Image Classification
PyTorch
ophthalmology
object-detection
medical-imaging
fundus
oct
detectron2
figure-parsing
Model card Files
xet
 Community
pubmed-ophtha commited on
Commit
632fad0
·
verified ·
1 Parent(s): e223016

Upload folder using huggingface_hub

Browse files
Files changed (5) hide show
  1. models/imaging_type_detection_1515892632/config.yaml +332 -0
  2. models/imaging_type_detection_1515892632/model_0003909.pth +3 -0
  3. models/mark_status_classifier_482239176/model_epoch_7.pth +3 -0
  4. models/panel_detection_1020880423/config.yaml +332 -0
  5. models/panel_detection_1020880423/model_0026865.pth +3 -0
models/imaging_type_detection_1515892632/config.yaml ADDED
@@ -0,0 +1,332 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ CUDNN_BENCHMARK: false
2
+ DATALOADER:
3
+ ASPECT_RATIO_GROUPING: true
4
+ FILTER_EMPTY_ANNOTATIONS: false
5
+ NUM_WORKERS: 2
6
+ REPEAT_SQRT: true
7
+ REPEAT_THRESHOLD: 0.0
8
+ SAMPLER_TRAIN: TrainingSampler
9
+ DATASETS:
10
+ PRECOMPUTED_PROPOSAL_TOPK_TEST: 1000
11
+ PRECOMPUTED_PROPOSAL_TOPK_TRAIN: 2000
12
+ PROPOSAL_FILES_TEST: []
13
+ PROPOSAL_FILES_TRAIN: []
14
+ TEST:
15
+ - imaging_type_detection_test
16
+ TRAIN:
17
+ - imaging_type_detection_train
18
+ VALIDATION: ''
19
+ FLOAT32_PRECISION: ''
20
+ GLOBAL:
21
+ HACK: 1.0
22
+ INPUT:
23
+ CROP:
24
+ ENABLED: false
25
+ SIZE:
26
+ - 0.9
27
+ - 0.9
28
+ TYPE: relative_range
29
+ FORMAT: BGR
30
+ MASK_FORMAT: polygon
31
+ MAX_SIZE_TEST: 1333
32
+ MAX_SIZE_TRAIN: 1333
33
+ MIN_SIZE_TEST: 800
34
+ MIN_SIZE_TRAIN:
35
+ - 640
36
+ - 672
37
+ - 704
38
+ - 736
39
+ - 768
40
+ - 800
41
+ MIN_SIZE_TRAIN_SAMPLING: choice
42
+ RANDOM_FLIP: horizontal
43
+ USE_TRANSFORMS: true
44
+ MODEL:
45
+ ANCHOR_GENERATOR:
46
+ ANGLES:
47
+ - - -90
48
+ - 0
49
+ - 90
50
+ ASPECT_RATIOS:
51
+ - - 0.5
52
+ - 1.0
53
+ - 2.0
54
+ NAME: DefaultAnchorGenerator
55
+ OFFSET: 0.0
56
+ SIZES:
57
+ - - 32
58
+ - 40.31747359663594
59
+ - 50.79683366298238
60
+ - - 64
61
+ - 80.63494719327188
62
+ - 101.59366732596476
63
+ - - 128
64
+ - 161.26989438654377
65
+ - 203.18733465192952
66
+ - - 256
67
+ - 322.53978877308754
68
+ - 406.37466930385904
69
+ - - 512
70
+ - 645.0795775461751
71
+ - 812.7493386077181
72
+ BACKBONE:
73
+ FREEZE_AT: 2
74
+ NAME: build_retinanet_resnet_fpn_backbone
75
+ DEVICE: cuda
76
+ FPN:
77
+ FUSE_TYPE: sum
78
+ IN_FEATURES:
79
+ - res3
80
+ - res4
81
+ - res5
82
+ NORM: ''
83
+ OUT_CHANNELS: 256
84
+ KEYPOINT_ON: false
85
+ LOAD_PROPOSALS: false
86
+ MASK_ON: false
87
+ META_ARCHITECTURE: RetinaNet
88
+ PANOPTIC_FPN:
89
+ COMBINE:
90
+ ENABLED: true
91
+ INSTANCES_CONFIDENCE_THRESH: 0.5
92
+ OVERLAP_THRESH: 0.5
93
+ STUFF_AREA_LIMIT: 4096
94
+ INSTANCE_LOSS_WEIGHT: 1.0
95
+ PIXEL_MEAN:
96
+ - 103.53
97
+ - 116.28
98
+ - 123.675
99
+ PIXEL_STD:
100
+ - 1.0
101
+ - 1.0
102
+ - 1.0
103
+ PROPOSAL_GENERATOR:
104
+ MIN_SIZE: 0
105
+ NAME: RPN
106
+ RESNETS:
107
+ DEFORM_MODULATED: false
108
+ DEFORM_NUM_GROUPS: 1
109
+ DEFORM_ON_PER_STAGE:
110
+ - false
111
+ - false
112
+ - false
113
+ - false
114
+ DEPTH: 50
115
+ NORM: FrozenBN
116
+ NUM_GROUPS: 1
117
+ OUT_FEATURES:
118
+ - res3
119
+ - res4
120
+ - res5
121
+ RES2_OUT_CHANNELS: 256
122
+ RES5_DILATION: 1
123
+ STEM_OUT_CHANNELS: 64
124
+ STRIDE_IN_1X1: true
125
+ WIDTH_PER_GROUP: 64
126
+ RETINANET:
127
+ BBOX_REG_LOSS_TYPE: smooth_l1
128
+ BBOX_REG_WEIGHTS: &id002
129
+ - 1.0
130
+ - 1.0
131
+ - 1.0
132
+ - 1.0
133
+ FOCAL_LOSS_ALPHA: 0.25
134
+ FOCAL_LOSS_GAMMA: 2.0
135
+ IN_FEATURES:
136
+ - p3
137
+ - p4
138
+ - p5
139
+ - p6
140
+ - p7
141
+ IOU_LABELS:
142
+ - 0
143
+ - -1
144
+ - 1
145
+ IOU_THRESHOLDS:
146
+ - 0.4
147
+ - 0.5
148
+ NMS_THRESH_TEST: 0.1
149
+ NORM: ''
150
+ NUM_CLASSES: 4
151
+ NUM_CONVS: 4
152
+ PRIOR_PROB: 0.01
153
+ SCORE_THRESH_TEST: 0.05
154
+ SMOOTH_L1_LOSS_BETA: 0.0
155
+ TOPK_CANDIDATES_TEST: 1000
156
+ ROI_BOX_CASCADE_HEAD:
157
+ BBOX_REG_WEIGHTS:
158
+ - &id001
159
+ - 10.0
160
+ - 10.0
161
+ - 5.0
162
+ - 5.0
163
+ - - 20.0
164
+ - 20.0
165
+ - 10.0
166
+ - 10.0
167
+ - - 30.0
168
+ - 30.0
169
+ - 15.0
170
+ - 15.0
171
+ IOUS:
172
+ - 0.5
173
+ - 0.6
174
+ - 0.7
175
+ ROI_BOX_HEAD:
176
+ BBOX_REG_LOSS_TYPE: smooth_l1
177
+ BBOX_REG_LOSS_WEIGHT: 1.0
178
+ BBOX_REG_WEIGHTS: *id001
179
+ CLS_AGNOSTIC_BBOX_REG: false
180
+ CONV_DIM: 256
181
+ FC_DIM: 1024
182
+ FED_LOSS_FREQ_WEIGHT_POWER: 0.5
183
+ FED_LOSS_NUM_CLASSES: 50
184
+ NAME: ''
185
+ NORM: ''
186
+ NUM_CONV: 0
187
+ NUM_FC: 0
188
+ POOLER_RESOLUTION: 14
189
+ POOLER_SAMPLING_RATIO: 0
190
+ POOLER_TYPE: ROIAlignV2
191
+ SMOOTH_L1_BETA: 0.0
192
+ TRAIN_ON_PRED_BOXES: false
193
+ USE_FED_LOSS: false
194
+ USE_SIGMOID_CE: false
195
+ ROI_HEADS:
196
+ BATCH_SIZE_PER_IMAGE: 512
197
+ IN_FEATURES:
198
+ - res4
199
+ IOU_LABELS:
200
+ - 0
201
+ - 1
202
+ IOU_THRESHOLDS:
203
+ - 0.5
204
+ NAME: Res5ROIHeads
205
+ NMS_THRESH_TEST: 0.5
206
+ NUM_CLASSES: 80
207
+ POSITIVE_FRACTION: 0.25
208
+ PROPOSAL_APPEND_GT: true
209
+ SCORE_THRESH_TEST: 0.05
210
+ ROI_KEYPOINT_HEAD:
211
+ CONV_DIMS:
212
+ - 512
213
+ - 512
214
+ - 512
215
+ - 512
216
+ - 512
217
+ - 512
218
+ - 512
219
+ - 512
220
+ LOSS_WEIGHT: 1.0
221
+ MIN_KEYPOINTS_PER_IMAGE: 1
222
+ NAME: KRCNNConvDeconvUpsampleHead
223
+ NORMALIZE_LOSS_BY_VISIBLE_KEYPOINTS: true
224
+ NUM_KEYPOINTS: 17
225
+ POOLER_RESOLUTION: 14
226
+ POOLER_SAMPLING_RATIO: 0
227
+ POOLER_TYPE: ROIAlignV2
228
+ ROI_MASK_HEAD:
229
+ CLS_AGNOSTIC_MASK: false
230
+ CONV_DIM: 256
231
+ NAME: MaskRCNNConvUpsampleHead
232
+ NORM: ''
233
+ NUM_CONV: 0
234
+ POOLER_RESOLUTION: 14
235
+ POOLER_SAMPLING_RATIO: 0
236
+ POOLER_TYPE: ROIAlignV2
237
+ RPN:
238
+ BATCH_SIZE_PER_IMAGE: 256
239
+ BBOX_REG_LOSS_TYPE: smooth_l1
240
+ BBOX_REG_LOSS_WEIGHT: 1.0
241
+ BBOX_REG_WEIGHTS: *id002
242
+ BOUNDARY_THRESH: -1
243
+ CONV_DIMS:
244
+ - -1
245
+ HEAD_NAME: StandardRPNHead
246
+ IN_FEATURES:
247
+ - res4
248
+ IOU_LABELS:
249
+ - 0
250
+ - -1
251
+ - 1
252
+ IOU_THRESHOLDS:
253
+ - 0.3
254
+ - 0.7
255
+ LOSS_WEIGHT: 1.0
256
+ NMS_THRESH: 0.7
257
+ POSITIVE_FRACTION: 0.5
258
+ POST_NMS_TOPK_TEST: 1000
259
+ POST_NMS_TOPK_TRAIN: 2000
260
+ PRE_NMS_TOPK_TEST: 6000
261
+ PRE_NMS_TOPK_TRAIN: 12000
262
+ SMOOTH_L1_BETA: 0.0
263
+ SEM_SEG_HEAD:
264
+ COMMON_STRIDE: 4
265
+ CONVS_DIM: 128
266
+ IGNORE_VALUE: 255
267
+ IN_FEATURES:
268
+ - p2
269
+ - p3
270
+ - p4
271
+ - p5
272
+ LOSS_WEIGHT: 1.0
273
+ NAME: SemSegFPNHead
274
+ NORM: GN
275
+ NUM_CLASSES: 54
276
+ WEIGHTS: models/pretrained_checkpoint_3104848164/model_0002099.pth
277
+ OUTPUT_DIR: models/imaging_type_detection_1515892632
278
+ SEED: 1515892632
279
+ SOLVER:
280
+ AMP:
281
+ ENABLED: false
282
+ BASE_LR: 0.001
283
+ BASE_LR_END: 0.0
284
+ BIAS_LR_FACTOR: 1.0
285
+ CHECKPOINT_PERIOD: 85
286
+ CLIP_GRADIENTS:
287
+ CLIP_TYPE: value
288
+ CLIP_VALUE: 1.0
289
+ ENABLED: false
290
+ NORM_TYPE: 2.0
291
+ GAMMA: 0.1
292
+ IMS_PER_BATCH: 16
293
+ LR_SCHEDULER_NAME: WarmupMultiStepLR
294
+ MAX_ITER: 15000
295
+ MOMENTUM: 0.9
296
+ NESTEROV: false
297
+ NUM_DECAYS: 3
298
+ REFERENCE_WORLD_SIZE: 0
299
+ RESCALE_INTERVAL: false
300
+ STEPS:
301
+ - 4000
302
+ - 12000
303
+ WARMUP_FACTOR: 0.001
304
+ WARMUP_ITERS: 1000
305
+ WARMUP_METHOD: linear
306
+ WEIGHT_DECAY: 0.0001
307
+ WEIGHT_DECAY_BIAS: null
308
+ WEIGHT_DECAY_NORM: 0.0
309
+ TEST:
310
+ AUG:
311
+ ENABLED: false
312
+ FLIP: true
313
+ MAX_SIZE: 4000
314
+ MIN_SIZES:
315
+ - 400
316
+ - 500
317
+ - 600
318
+ - 700
319
+ - 800
320
+ - 900
321
+ - 1000
322
+ - 1100
323
+ - 1200
324
+ DETECTIONS_PER_IMAGE: 60
325
+ EVAL_PERIOD: 85
326
+ EXPECTED_RESULTS: []
327
+ KEYPOINT_OKS_SIGMAS: []
328
+ PRECISE_BN:
329
+ ENABLED: false
330
+ NUM_ITER: 200
331
+ VERSION: 2
332
+ VIS_PERIOD: 0
models/imaging_type_detection_1515892632/model_0003909.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f0629ab545756f2f0a4814f939090f6599ea40c11f09e4f67b0a074a5f41bf40
3
+ size 290380169
models/mark_status_classifier_482239176/model_epoch_7.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fca459a8738f15ad9462e6f32a5af1c53cc509ee6f6916e4b75145ff08ec894b
3
+ size 94368483
models/panel_detection_1020880423/config.yaml ADDED
@@ -0,0 +1,332 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ CUDNN_BENCHMARK: false
2
+ DATALOADER:
3
+ ASPECT_RATIO_GROUPING: true
4
+ FILTER_EMPTY_ANNOTATIONS: false
5
+ NUM_WORKERS: 2
6
+ REPEAT_SQRT: true
7
+ REPEAT_THRESHOLD: 0.0
8
+ SAMPLER_TRAIN: TrainingSampler
9
+ DATASETS:
10
+ PRECOMPUTED_PROPOSAL_TOPK_TEST: 1000
11
+ PRECOMPUTED_PROPOSAL_TOPK_TRAIN: 2000
12
+ PROPOSAL_FILES_TEST: []
13
+ PROPOSAL_FILES_TRAIN: []
14
+ TEST:
15
+ - panel_detection_test
16
+ TRAIN:
17
+ - panel_detection_train
18
+ VALIDATION: ''
19
+ FLOAT32_PRECISION: ''
20
+ GLOBAL:
21
+ HACK: 1.0
22
+ INPUT:
23
+ CROP:
24
+ ENABLED: false
25
+ SIZE:
26
+ - 0.9
27
+ - 0.9
28
+ TYPE: relative_range
29
+ FORMAT: BGR
30
+ MASK_FORMAT: polygon
31
+ MAX_SIZE_TEST: 1333
32
+ MAX_SIZE_TRAIN: 1333
33
+ MIN_SIZE_TEST: 800
34
+ MIN_SIZE_TRAIN:
35
+ - 640
36
+ - 672
37
+ - 704
38
+ - 736
39
+ - 768
40
+ - 800
41
+ MIN_SIZE_TRAIN_SAMPLING: choice
42
+ RANDOM_FLIP: horizontal
43
+ USE_TRANSFORMS: true
44
+ MODEL:
45
+ ANCHOR_GENERATOR:
46
+ ANGLES:
47
+ - - -90
48
+ - 0
49
+ - 90
50
+ ASPECT_RATIOS:
51
+ - - 0.5
52
+ - 1.0
53
+ - 2.0
54
+ NAME: DefaultAnchorGenerator
55
+ OFFSET: 0.0
56
+ SIZES:
57
+ - - 32
58
+ - 40.31747359663594
59
+ - 50.79683366298238
60
+ - - 64
61
+ - 80.63494719327188
62
+ - 101.59366732596476
63
+ - - 128
64
+ - 161.26989438654377
65
+ - 203.18733465192952
66
+ - - 256
67
+ - 322.53978877308754
68
+ - 406.37466930385904
69
+ - - 512
70
+ - 645.0795775461751
71
+ - 812.7493386077181
72
+ BACKBONE:
73
+ FREEZE_AT: 2
74
+ NAME: build_retinanet_resnet_fpn_backbone
75
+ DEVICE: cuda
76
+ FPN:
77
+ FUSE_TYPE: sum
78
+ IN_FEATURES:
79
+ - res3
80
+ - res4
81
+ - res5
82
+ NORM: ''
83
+ OUT_CHANNELS: 256
84
+ KEYPOINT_ON: false
85
+ LOAD_PROPOSALS: false
86
+ MASK_ON: false
87
+ META_ARCHITECTURE: RetinaNet
88
+ PANOPTIC_FPN:
89
+ COMBINE:
90
+ ENABLED: true
91
+ INSTANCES_CONFIDENCE_THRESH: 0.5
92
+ OVERLAP_THRESH: 0.5
93
+ STUFF_AREA_LIMIT: 4096
94
+ INSTANCE_LOSS_WEIGHT: 1.0
95
+ PIXEL_MEAN:
96
+ - 103.53
97
+ - 116.28
98
+ - 123.675
99
+ PIXEL_STD:
100
+ - 1.0
101
+ - 1.0
102
+ - 1.0
103
+ PROPOSAL_GENERATOR:
104
+ MIN_SIZE: 0
105
+ NAME: RPN
106
+ RESNETS:
107
+ DEFORM_MODULATED: false
108
+ DEFORM_NUM_GROUPS: 1
109
+ DEFORM_ON_PER_STAGE:
110
+ - false
111
+ - false
112
+ - false
113
+ - false
114
+ DEPTH: 50
115
+ NORM: FrozenBN
116
+ NUM_GROUPS: 1
117
+ OUT_FEATURES:
118
+ - res3
119
+ - res4
120
+ - res5
121
+ RES2_OUT_CHANNELS: 256
122
+ RES5_DILATION: 1
123
+ STEM_OUT_CHANNELS: 64
124
+ STRIDE_IN_1X1: true
125
+ WIDTH_PER_GROUP: 64
126
+ RETINANET:
127
+ BBOX_REG_LOSS_TYPE: smooth_l1
128
+ BBOX_REG_WEIGHTS: &id002
129
+ - 1.0
130
+ - 1.0
131
+ - 1.0
132
+ - 1.0
133
+ FOCAL_LOSS_ALPHA: 0.25
134
+ FOCAL_LOSS_GAMMA: 2.0
135
+ IN_FEATURES:
136
+ - p3
137
+ - p4
138
+ - p5
139
+ - p6
140
+ - p7
141
+ IOU_LABELS:
142
+ - 0
143
+ - -1
144
+ - 1
145
+ IOU_THRESHOLDS:
146
+ - 0.4
147
+ - 0.5
148
+ NMS_THRESH_TEST: 0.1
149
+ NORM: ''
150
+ NUM_CLASSES: 2
151
+ NUM_CONVS: 4
152
+ PRIOR_PROB: 0.01
153
+ SCORE_THRESH_TEST: 0.05
154
+ SMOOTH_L1_LOSS_BETA: 0.0
155
+ TOPK_CANDIDATES_TEST: 1000
156
+ ROI_BOX_CASCADE_HEAD:
157
+ BBOX_REG_WEIGHTS:
158
+ - &id001
159
+ - 10.0
160
+ - 10.0
161
+ - 5.0
162
+ - 5.0
163
+ - - 20.0
164
+ - 20.0
165
+ - 10.0
166
+ - 10.0
167
+ - - 30.0
168
+ - 30.0
169
+ - 15.0
170
+ - 15.0
171
+ IOUS:
172
+ - 0.5
173
+ - 0.6
174
+ - 0.7
175
+ ROI_BOX_HEAD:
176
+ BBOX_REG_LOSS_TYPE: smooth_l1
177
+ BBOX_REG_LOSS_WEIGHT: 1.0
178
+ BBOX_REG_WEIGHTS: *id001
179
+ CLS_AGNOSTIC_BBOX_REG: false
180
+ CONV_DIM: 256
181
+ FC_DIM: 1024
182
+ FED_LOSS_FREQ_WEIGHT_POWER: 0.5
183
+ FED_LOSS_NUM_CLASSES: 50
184
+ NAME: ''
185
+ NORM: ''
186
+ NUM_CONV: 0
187
+ NUM_FC: 0
188
+ POOLER_RESOLUTION: 14
189
+ POOLER_SAMPLING_RATIO: 0
190
+ POOLER_TYPE: ROIAlignV2
191
+ SMOOTH_L1_BETA: 0.0
192
+ TRAIN_ON_PRED_BOXES: false
193
+ USE_FED_LOSS: false
194
+ USE_SIGMOID_CE: false
195
+ ROI_HEADS:
196
+ BATCH_SIZE_PER_IMAGE: 512
197
+ IN_FEATURES:
198
+ - res4
199
+ IOU_LABELS:
200
+ - 0
201
+ - 1
202
+ IOU_THRESHOLDS:
203
+ - 0.5
204
+ NAME: Res5ROIHeads
205
+ NMS_THRESH_TEST: 0.5
206
+ NUM_CLASSES: 80
207
+ POSITIVE_FRACTION: 0.25
208
+ PROPOSAL_APPEND_GT: true
209
+ SCORE_THRESH_TEST: 0.05
210
+ ROI_KEYPOINT_HEAD:
211
+ CONV_DIMS:
212
+ - 512
213
+ - 512
214
+ - 512
215
+ - 512
216
+ - 512
217
+ - 512
218
+ - 512
219
+ - 512
220
+ LOSS_WEIGHT: 1.0
221
+ MIN_KEYPOINTS_PER_IMAGE: 1
222
+ NAME: KRCNNConvDeconvUpsampleHead
223
+ NORMALIZE_LOSS_BY_VISIBLE_KEYPOINTS: true
224
+ NUM_KEYPOINTS: 17
225
+ POOLER_RESOLUTION: 14
226
+ POOLER_SAMPLING_RATIO: 0
227
+ POOLER_TYPE: ROIAlignV2
228
+ ROI_MASK_HEAD:
229
+ CLS_AGNOSTIC_MASK: false
230
+ CONV_DIM: 256
231
+ NAME: MaskRCNNConvUpsampleHead
232
+ NORM: ''
233
+ NUM_CONV: 0
234
+ POOLER_RESOLUTION: 14
235
+ POOLER_SAMPLING_RATIO: 0
236
+ POOLER_TYPE: ROIAlignV2
237
+ RPN:
238
+ BATCH_SIZE_PER_IMAGE: 256
239
+ BBOX_REG_LOSS_TYPE: smooth_l1
240
+ BBOX_REG_LOSS_WEIGHT: 1.0
241
+ BBOX_REG_WEIGHTS: *id002
242
+ BOUNDARY_THRESH: -1
243
+ CONV_DIMS:
244
+ - -1
245
+ HEAD_NAME: StandardRPNHead
246
+ IN_FEATURES:
247
+ - res4
248
+ IOU_LABELS:
249
+ - 0
250
+ - -1
251
+ - 1
252
+ IOU_THRESHOLDS:
253
+ - 0.3
254
+ - 0.7
255
+ LOSS_WEIGHT: 1.0
256
+ NMS_THRESH: 0.7
257
+ POSITIVE_FRACTION: 0.5
258
+ POST_NMS_TOPK_TEST: 1000
259
+ POST_NMS_TOPK_TRAIN: 2000
260
+ PRE_NMS_TOPK_TEST: 6000
261
+ PRE_NMS_TOPK_TRAIN: 12000
262
+ SMOOTH_L1_BETA: 0.0
263
+ SEM_SEG_HEAD:
264
+ COMMON_STRIDE: 4
265
+ CONVS_DIM: 128
266
+ IGNORE_VALUE: 255
267
+ IN_FEATURES:
268
+ - p2
269
+ - p3
270
+ - p4
271
+ - p5
272
+ LOSS_WEIGHT: 1.0
273
+ NAME: SemSegFPNHead
274
+ NORM: GN
275
+ NUM_CLASSES: 54
276
+ WEIGHTS: models/pretrained_checkpoint_3104848164/model_0002099.pth
277
+ OUTPUT_DIR: models/panel_detection_1020880423
278
+ SEED: 1020880423
279
+ SOLVER:
280
+ AMP:
281
+ ENABLED: false
282
+ BASE_LR: 0.001
283
+ BASE_LR_END: 0.0
284
+ BIAS_LR_FACTOR: 1.0
285
+ CHECKPOINT_PERIOD: 707
286
+ CLIP_GRADIENTS:
287
+ CLIP_TYPE: value
288
+ CLIP_VALUE: 1.0
289
+ ENABLED: false
290
+ NORM_TYPE: 2.0
291
+ GAMMA: 0.1
292
+ IMS_PER_BATCH: 16
293
+ LR_SCHEDULER_NAME: WarmupMultiStepLR
294
+ MAX_ITER: 100000
295
+ MOMENTUM: 0.9
296
+ NESTEROV: false
297
+ NUM_DECAYS: 3
298
+ REFERENCE_WORLD_SIZE: 0
299
+ RESCALE_INTERVAL: false
300
+ STEPS:
301
+ - 26000
302
+ - 80000
303
+ WARMUP_FACTOR: 0.001
304
+ WARMUP_ITERS: 1000
305
+ WARMUP_METHOD: linear
306
+ WEIGHT_DECAY: 0.0001
307
+ WEIGHT_DECAY_BIAS: null
308
+ WEIGHT_DECAY_NORM: 0.0
309
+ TEST:
310
+ AUG:
311
+ ENABLED: false
312
+ FLIP: true
313
+ MAX_SIZE: 4000
314
+ MIN_SIZES:
315
+ - 400
316
+ - 500
317
+ - 600
318
+ - 700
319
+ - 800
320
+ - 900
321
+ - 1000
322
+ - 1100
323
+ - 1200
324
+ DETECTIONS_PER_IMAGE: 60
325
+ EVAL_PERIOD: 707
326
+ EXPECTED_RESULTS: []
327
+ KEYPOINT_OKS_SIGMAS: []
328
+ PRECISE_BN:
329
+ ENABLED: false
330
+ NUM_ITER: 200
331
+ VERSION: 2
332
+ VIS_PERIOD: 0
models/panel_detection_1020880423/model_0026865.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e1c672c53b4b30090ffb7282fd3290e40869521031f5b7378c20436f3f890766
3
+ size 290048265