-
Notifications
You must be signed in to change notification settings - Fork 17
/
dinov_sam_coco_swinl_train.yaml
404 lines (392 loc) · 9.46 KB
/
dinov_sam_coco_swinl_train.yaml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
# --------------------------------------------------------
# X-Decoder -- Generalized Decoding for Pixel, Image, and Language
# Copyright (c) 2022 Microsoft
# Licensed under The MIT License [see LICENSE for details]
# Written by Xueyan Zou ([email protected])
# --------------------------------------------------------
# Copyright (c) 2024 Microsoft
# Licensed under The MIT License [see LICENSE for details]
# Written by Feng Li ([email protected])
# --------------------------------------------------------
##################
# Task settings
##################
WEIGHT: ''
PORT: 53711
VERBOSE: true
track_thresh_hold: 1
OUTPUT_DIR: '../data/output/test'
# misc
LOADER:
JOINT: True
KEY_DATASET: 'coco'
# Spatial sampler
STROKE_SAMPLER:
MAX_CANDIDATE: 50
CANDIDATE_PROBS: [1.0] # for training only
CANDIDATE_NAMES: ["Scribble"]
DILATION: 3
SCRIBBLE:
NUM_STROKES: 5
STROKE_PRESET: ['rand_curve', 'rand_curve_small']
STROKE_PROB: [0.5, 0.5]
EVAL:
MODE: 'best' # best/random
NEGATIVE: False
MAX_ITER: 20
# model
MODEL:
NAME: dinov
HEAD: general_head
MASK_ON: false
KEYPOINT_ON: false
LOAD_PROPOSALS: false
DIM_PROJ: 512
BACKBONE_DIM: 768
BACKGROUND: False
WEIGHTS: ''
TEXT:
ARCH: noencoder
NAME: transformer
TOKENIZER: clip
CONTEXT_LENGTH: 18 # 77
WIDTH: 512
HEADS: 8
LAYERS: 12 # 6
AUTOGRESSIVE: True
BACKBONE:
NAME: swin
PRETRAINED: 'https://github.com/SwinTransformer/storage/releases/download/v1.0.0/swin_large_patch4_window12_384_22k.pth'
LOAD_PRETRAINED: true
SWIN:
PRETRAIN_IMG_SIZE: 384
PATCH_SIZE: 4
EMBED_DIM: 192
DEPTHS: [ 2, 2, 18, 2 ]
NUM_HEADS: [ 6, 12, 24, 48 ]
WINDOW_SIZE: 12
MLP_RATIO: 4.0
QKV_BIAS: true
QK_SCALE: ~
DROP_RATE: 0.0
ATTN_DROP_RATE: 0.0
DROP_PATH_RATE: 0.3
APE: false
PATCH_NORM: true
USE_CHECKPOINT: true
OUT_FEATURES: [ 'res2', 'res3', 'res4', 'res5' ]
ENCODER:
NAME: encoder_deform
IGNORE_VALUE: 255
NUM_CLASSES: 133
LOSS_WEIGHT: 1.0
CONVS_DIM: 256
MASK_DIM: 256
NORM: "GN"
IN_FEATURES: [ "res2", "res3", "res4", "res5" ]
DEFORMABLE_TRANSFORMER_ENCODER_IN_FEATURES: [ "res2", "res3", "res4", "res5" ]
COMMON_STRIDE: 4
TRANSFORMER_ENC_LAYERS: 6
TOTAL_NUM_FEATURE_LEVELS: 5
NUM_FEATURE_LEVELS: 4
FEATURE_ORDER: "low2high"
DECODER:
NAME: dinov_openset_decoder
TRANSFORMER_IN_FEATURE: "multi_scale_pixel_decoder"
MASK: True
BOX: True
PART: True
GROUNDING:
ENABLED: False
MAX_LEN: 5
TEXT_WEIGHT: 2.0
CLASS_WEIGHT: 0.5
CAPTION:
ENABLED: False
PHRASE_PROB: 0.0
SIM_THRES: 0.95
CAPTIONING:
ENABLED: False
STEP: 50
RETRIEVAL:
ENABLED: False
DIM_IMG: 768
ENSEMBLE: True
OPENIMAGE:
ENABLED: False
NEGATIVE_SAMPLES: 5
GROUNDING:
ENABLED: False
MAX_LEN: 5
DEEP_SUPERVISION: True
NO_OBJECT_WEIGHT: 0.1
CLASS_WEIGHT: 4.0
MASK_WEIGHT: 5.0
DICE_WEIGHT: 5.0
BOX_WEIGHT: 5.0
GIOU_WEIGHT: 2.0
IOU_WEIGHT: 1.0
COST_CLASS_WEIGHT: 4.0
COST_DICE_WEIGHT: 5.0
COST_MASK_WEIGHT: 5.0
COST_BOX_WEIGHT: 5.0
COST_GIOU_WEIGHT: 2.0
MATCH_CLASS_WEIGHT: 2.0
HIDDEN_DIM: 256
NUM_OBJECT_QUERIES: 300
NHEADS: 8
DROPOUT: 0.0
DIM_FEEDFORWARD: 2048
ENC_LAYERS: 0
PRE_NORM: False
ENFORCE_INPUT_PROJ: False
SIZE_DIVISIBILITY: 32
DEC_LAYERS: 9 # 9 decoder layers, add one for the loss on learnable query
TRAIN_NUM_POINTS: 12544
OVERSAMPLE_RATIO: 3.0
IMPORTANCE_SAMPLE_RATIO: 0.75
TWO_STAGE: False
INITIALIZE_BOX_TYPE: 'no'
DN: seg
DN_NOISE_SCALE: 0.4
DN_NUM: 100
INITIAL_PRED: False
LEARN_TGT: False
TOTAL_NUM_FEATURE_LEVELS: 5
SEMANTIC_CE_LOSS: False
PANO_BOX_LOSS: False
COCO: True
O365: False
SAM: True
COCO_TRACK: False
PASCAL: False
RE_POINT: True
NUM_INTERACTIVE_TOKENS: 6
NUM_CONTENT_TOKENS: 1
MAX_NUM_INSTANCE: 50
MAX_NUM_INSTANCE_CONTENT: 8
USE_SHAPE_SAMPLER: True
OPENSET_USE_SHAPE_SAMPLER: False
INFERENCE_EXAMPLE: 4
MAX_TRAIN_EXAMPLE: 4
CROSS_GPU_BATCH: True
CONTENT_INDEPENDENT: True
NMS_THRESHOLD: 0.7
REFER_IMAGE: False
point_per_side: 20
MAX_MEMORY_SIZE: 9
softmax: False
MANY2MANY: True
VIS: False
freeze_all: False
freeze_backbone_enc: False
freeze_backbone_enc_decoder: False
TEST:
SEMANTIC_ON: True
INSTANCE_ON: True
PANOPTIC_ON: True
BOX_INTERACTIVE: False
CLASSIFICATION_ON: False
OVERLAP_THRESHOLD: 0.8
OBJECT_MASK_THRESHOLD: 0.25
SEM_SEG_POSTPROCESSING_BEFORE_INFERENCE: false
TEST_FOUCUS_ON_BOX: False
PANO_TRANSFORM_EVAL: True
PANO_TEMPERATURE: 0.06
TEST:
EVAL_PERIOD: 500000
PRECISE_BN:
NUM_ITER: 1
ENABLED: False
AUG:
ENABLED: False
DATASETS:
TRAIN: ["sam_train", "coco_2017_train_panoptic"]
TEST: ["coco_2017_val_panoptic_with_sem_seg"]
# TEST: ["ade20k_panoptic_val", "ade20k_full_sem_seg_val", "sunrgbd_37_val_seg", "scannet_21_val_seg", "scannet_21_panoptic_val", "scannet_41_val_seg", "cityscapes_fine_panoptic_val", "cityscapes_fine_instance_seg_val", "cityscapes_fine_sem_seg_val", "bdd10k_val_sem_seg", "bdd10k_40_panoptic_val"]
CLASS_CONCAT: false
SIZE_DIVISIBILITY: 32
PROPOSAL_FILES_TRAIN: []
SAM:
INPUT:
MIN_SIZE_TEST: 800
MAX_SIZE_TEST: 1333
IMAGE_SIZE: 1024
MIN_SCALE: 0.99
MAX_SCALE: 1.01
DATASET_MAPPER_NAME: "sam_content"
IGNORE_VALUE: 255
COLOR_AUG_SSD: False
SIZE_DIVISIBILITY: 32
RANDOM_FLIP: "horizontal"
MASK_FORMAT: "polygon"
FORMAT: "RGB"
CROP:
ENABLED: True
DATASET:
DATASET: 'sam'
TEST:
DETECTIONS_PER_IMAGE: 100
NAME: coco_eval
IOU_TYPE: ['bbox', 'segm']
USE_MULTISCALE: false
BATCH_SIZE_TOTAL: 8
MODEL_FILE: ''
AUG:
ENABLED: False
TRAIN:
BATCH_SIZE_TOTAL: 1
BATCH_SIZE_PER_GPU: 1
SHUFFLE: true
DATALOADER:
FILTER_EMPTY_ANNOTATIONS: False
NUM_WORKERS: 4
LOAD_PROPOSALS: False
SAMPLER_TRAIN: "TrainingSampler"
ASPECT_RATIO_GROUPING: True
COCO:
INPUT:
MIN_SIZE_TEST: 800
MAX_SIZE_TEST: 1333
IMAGE_SIZE: 1024
MIN_SCALE: 0.1
MAX_SCALE: 2.0
DATASET_MAPPER_NAME: "coco_interactive_panoptic_lsj"
IGNORE_VALUE: 255
COLOR_AUG_SSD: False
SIZE_DIVISIBILITY: 32
RANDOM_FLIP: "horizontal"
MASK_FORMAT: "polygon"
FORMAT: "RGB"
CROP:
ENABLED: True
DATASET:
DATASET: 'coco'
TEST:
DETECTIONS_PER_IMAGE: 100
NAME: coco_eval
IOU_TYPE: ['bbox', 'segm']
USE_MULTISCALE: false
BATCH_SIZE_TOTAL: 1
MODEL_FILE: ''
AUG:
ENABLED: False
TRAIN:
BATCH_SIZE_TOTAL: 1
BATCH_SIZE_PER_GPU: 1
SHUFFLE: true
DATALOADER:
FILTER_EMPTY_ANNOTATIONS: False
NUM_WORKERS: 2
LOAD_PROPOSALS: False
SAMPLER_TRAIN: "TrainingSampler"
ASPECT_RATIO_GROUPING: True
INPUT:
PIXEL_MEAN: [123.675, 116.280, 103.530]
PIXEL_STD: [58.395, 57.120, 57.375]
DATALOADER:
FILTER_EMPTY_ANNOTATIONS: False
NUM_WORKERS: 16
LOAD_PROPOSALS: False
SAMPLER_TRAIN: "TrainingSampler"
ASPECT_RATIO_GROUPING: True
# Detectron2 training config for optimizer and lr scheduler
SOLVER:
BASE_LR_END: 0.0
MOMENTUM: 0.9
NESTEROV: False
CHECKPOINT_PERIOD: 5000
IMS_PER_BATCH: 1
REFERENCE_WORLD_SIZE: 0
BIAS_LR_FACTOR: 1.0
WEIGHT_DECAY_BIAS: None
# original
BASE_LR: 0.0001
# STEPS: [327778, 355092]
STEPS: [67778, 105092]
# MAX_ITER: 368750
MAX_ITER: 168750
GAMMA: 0.1
WARMUP_FACTOR: 1.0
WARMUP_ITERS: 10
WARMUP_METHOD: "linear"
WEIGHT_DECAY: 0.05
OPTIMIZER: "ADAMW"
LR_SCHEDULER_NAME: "WarmupMultiStepLR"
LR_MULTIPLIER:
backbone: 0.1
lang_encoder: 0.1
WEIGHT_DECAY_NORM: 0.0
WEIGHT_DECAY_EMBED: 0.0
CLIP_GRADIENTS:
ENABLED: True
CLIP_TYPE: "full_model"
CLIP_VALUE: 0.01
NORM_TYPE: 2.0
AMP:
ENABLED: True
# Evaluation Dataset
ADE20K:
INPUT:
MIN_SIZE_TRAIN: [320, 384, 448, 512, 576, 640, 704, 768, 832, 896, 960, 1024, 1088, 1152, 1216, 1280]
MIN_SIZE_TRAIN_SAMPLING: "choice"
MIN_SIZE_TEST: 640
MAX_SIZE_TRAIN: 2560
MAX_SIZE_TEST: 2560
MASK_FORMAT: "polygon"
CROP:
ENABLED: True
TYPE: "absolute"
SIZE: [640, 640]
SINGLE_CATEGORY_MAX_AREA: 1.0
IGNORE_VALUE: 255
COLOR_AUG_SSD: True
SIZE_DIVISIBILITY: 640 # used in dataset mapper
DATASET_MAPPER_NAME: "mask_former_panoptic"
FORMAT: "RGB"
DATASET:
DATASET: 'ade'
TRAIN:
ASPECT_RATIO_GROUPING: true
BATCH_SIZE_TOTAL: 16
BATCH_SIZE_PER_GPU: 2
SHUFFLE: true
TEST:
DETECTIONS_PER_IMAGE: 100
NAME: coco_eval
IOU_TYPE: ['bbox', 'segm']
USE_MULTISCALE: false
BATCH_SIZE_TOTAL: 1
MODEL_FILE: ''
AUG:
ENABLED: False
DATALOADER:
FILTER_EMPTY_ANNOTATIONS: False
NUM_WORKERS: 8
LOAD_PROPOSALS: False
SAMPLER_TRAIN: "TrainingSampler"
ASPECT_RATIO_GROUPING: True
DAVIS:
INPUT:
MIN_SIZE_TEST: 800
MAX_SIZE_TEST: 1333
DATALOADER:
FILTER_EMPTY_ANNOTATIONS: False
NUM_WORKERS: 0
LOAD_PROPOSALS: False
SAMPLER_TRAIN: "TrainingSampler"
ASPECT_RATIO_GROUPING: False
TEST:
BATCH_SIZE_TOTAL: 1
VOS:
INPUT:
MIN_SIZE_TEST: 800
MAX_SIZE_TEST: 1333
DATALOADER:
FILTER_EMPTY_ANNOTATIONS: False
NUM_WORKERS: 0
LOAD_PROPOSALS: False
SAMPLER_TRAIN: "TrainingSampler"
ASPECT_RATIO_GROUPING: False
TEST:
BATCH_SIZE_TOTAL: 1