-
Notifications
You must be signed in to change notification settings - Fork 4
/
Hulk_vit-B.yaml
4466 lines (4122 loc) · 158 KB
/
Hulk_vit-B.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
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
# task 0: attr, task 1: pose, task 2:caption task3: parsing task4: smpl, task 5: det
# fixed parameter with diverse shape among different tasks should also be set in the task_spec_list,
# e.g., text_vectors, pos_embed, etc.
# attr: 5: multi_rap2_PA_100k_parse27k_market_HARDHC 6: luperson
# caption: 7: image_caption_joint
# skeleton action: 0: ntu60+ntu120+gym 1: k400+diving48+gym
# smpl: 2: 3dpw , human3.6m , coco , muco , up3d , mpii , gta_human
# det: 3: Crowdhuman 4: 5set
# pose: 8: coco 9:aic 10: human3.6m 11: posetrack 12: jrdb 13: MHP 14: mpii 15: 3dpw 16: aist++
# parsing: 17:LIP 18:CIHP 19: human3.6m 20: modanet 21: VIP 22: deepfashion 23: PaperDoll
common: # prefix
share_backbone_group: [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0]
share_decoder_group: [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0]
# use modality groups to control the communication of neck, adapter, and output proj
share_rgb_group: [-1, -1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0] # rgb
share_dense_labeling_group: [-1, -1, -1, -1, -1, -1, -1, -1, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0] # dense_labeling
share_text_group: [0, 0, -1, -1, -1, 0, 0, 0, -1, -1, -1, -1, -1, -1, -1, -1,
-1, -1, -1, -1, -1, -1, -1, -1] # text
share_sparse_labeling_group: [ 0, 0, 0, 0, 0, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
-1, -1, -1, -1, -1, -1, -1, -1]
share_video_group: [-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
-1, -1, -1, -1, -1, -1, -1, -1]
# share modality is truly the share task group, e.g., parsing datasets share a group
share_modality_group: [ 2, 2, 3, 4, 4, 0, 0, 1, 5, 5, 5, 5, 5, 5, 5, 5,
5, 6, 6, 6, 6, 6, 6, 6 ]
solver:
type: SolverMAEDev
model_entry_type: aio_entry_v2mae_shareneck
lr_scheduler:
type: 'Cosine'
kwargs:
eta_min: 0.
base_lr: 1.e-5
warmup_lr: 1.e-3
warmup_steps: 1500
backbone_multiplier: 1.
pos_embed_multiplier: 1.
layer_decay:
num_layers: 12
layer_decay_rate: 0.75
lpe_lr: True
optimizer:
type: Adafactor_dev
kwargs:
beta1: 0.9
clip_beta2: 0.999
clip_threshold: 1.
decay_rate: -0.8
scale_parameter: False
relative_step: False
weight_decay: 0.05
auto_denan: False
workers: 2
max_iter: 60000 #61446 # 0.1628001628001628 * |61446 for 149813 // 512 * 210
deterministic: True # seed control
cudnn_deterministic: False
worker_rank: True
random_seed: 233
print_freq: 10
verbose_loss: False
vis_batch: False
save_interval: 10000
use_ceph: True
sync: True
collate: det
# task_specific_param = ['backbone', 'neck', 'decoder', 'dataset', 'sampler', 'lr_scheduler', 'optimizer']
tasks : # prefix
0:
name: NUTRGBD_skeleton #SPECIFIC
loss_weight: 4.4
gres_ratio: 2 # int, > 0, = Task_GPUs / (world_Size/sum(all_gres_ratios))
backbone:
type: vit_base_patch16_mask
kwargs:
task_sp_list: [ 'rel_pos_h', 'rel_pos_w' ] # wrong list would somehow cause .cuda() stuck without error
pretrained: True
lms_checkpoint_train: fairscale
window: False
test_pos_mode: learnable_interpolate
learnable_pos: False
drop_path_rate: 0.1
vis_patch_token_ratio: 1
vis_label_token_ratio: 0.
dataset:
type: mmSkeletonDataset # train for 150 epochs
kwargs:
ann_file:
- /mnt/path...to...//skaction_public/ntu60_hrnet.pkl
- /mnt/path...to...//skaction_public/ntu120_hrnet.pkl
- /mnt/path...to...//skaction_public/gym_hrnet.pkl
dataset_name:
- 2dntu60
- 2dntu120
- gym
kp_dim: 2d #SPECIFIC
one_hot: True
num_classes:
- 60
- 120
- 99
centernorm: False
scale_range: [ 0.75,1.25 ]
data_pipeline:
- type: PreNormalize2D
kwargs: { }
- type: GenSkeFeat
kwargs:
dataset: coco
feats: [ 'j' ]
- type: UniformSampleGivenFrames
kwargs:
clip_len: 25
given_len: 7
- type: PoseDecode
kwargs: { }
- type: FormatGCNInput2D
kwargs:
num_person: 2
window: False
rotate: True
mode: zero
- type: Collect
kwargs:
keys: [ 'keypoint', 'label' ]
meta_keys: [ ]
- type: ToTensor
kwargs:
keys: [ 'keypoint' ]
flip: True
sampler:
batch_size: 120 # per card
shuffle_strategy: 1
patch_neck:
type: MAEdecoder_proj_neck
kwargs:
mask_dim: 256
modality: sparse_labeling
label_neck:
type: MAEdecoder_proj_neck
kwargs:
mask_dim: 256
modality: text
patch_adapter:
type: sparse_labeling_adapter_skaction
kwargs:
pretrained: True # should be changed to True later
in_chans: 3
num_joints: 17 #SPECIFIC
num_frames: 175
embed_dim: 768
patch_size: [ 7, 2 ]
stride_level: [ 1, 1 ]
use_abs_pos_emb: True
learnable_pos: False
test_pos_mode: learnable_interpolate
type_embed: False
joint_with_text_embedding: True
joint_names: coco_body_17joints #SPECIFIC
proj_norm: 'LN'
stride_text_embedding: True
is_2d_dataset: True #SPECIFIC
modality_share_list: [
'merge_kernel',
'proj_kernel',
'proj', ]
task_sp_list: [ 'text_embedding', 'pos_embed', ]
patch_proj:
type: sparse_labeling_projector
kwargs:
task: skeleton
loss_cfg:
type: MaskDetFocalDiceLoss
kwargs:
cfg:
deep_supervision: True
focal_alpha: 0.25
class_weight: 2.0
bbox_weight: 5.0
giou_weight: 2.
ign_thr: 0.7
dec_layers: 6
num_classes: 1
predict3d: True
xyxy: True
in_chans: 3 # predefined in patch adapter, set in solver
num_joints: 17 #SPECIFIC
num_frames: 175
modality_share_list: [
'output_proj',
'translate_weight',
'translate_bias',
'post_mul_norm',
'patch_proj',
'class_proj'
]
task_sp_list: [
'text_vectors', # useless
'text_features',
]
label_adapter:
type: text_adapter
kwargs:
pretrained: True
#close_set: True
description_dict_name:
- ntu60_name
- ntu120_name
- gym_cls_name
one_way_semantics: False
skeleton_action: True # use skeleton action to Double the text embedding (when M=2)
skeleton_action_one_hot_label: True
task_sp_list: [ 'text_vectors', ]
label_proj:
type: text_projector
kwargs:
one_way_semantics: False
description_dict_name:
- ntu60_name
- ntu120_name
- gym_cls_name
skeleton_action: True
skeleton_action_one_hot_label: True
pre_proj_type: 'pool'
replace_post_mul_norm: False
post_mul_norm: True
task_sp_list: [ 'text_vectors',
'translate_weight',
'translate_bias',
'post_mul_norm', ]
loss_cfg:
type: CELoss
kwargs:
loss_weight: 1.0
decoder:
type: UniHCPv2_Head
kwargs:
predictor: 'hulk'
task: recons
modality_share_list: ['predictor.mask_token', ]
task_sp_list: [
'predictor.query_embed_patch',
'predictor.query_embed_label',
'predictor.class_embed','predictor.fc_bias', # useless in Hulk
] # wrong list would somehow cause .cuda() stuck without error
loss_weight: 1.0
transformer_predictor_cfg:
hidden_dim: 256
num_queries: 20 # useless in Hulk
nheads: 8
dim_feedforward: 2048
dec_layers: 9
pre_norm: False
arch: fan_in
enforce_input_project: False
mask_on: False
num_feature_levels: 1
cross_pos_embed: anchor
self_attn_mask_type: patch_diag_label_row
detach_from_peddet: True
loss_cfg:
type: CEL_Sigmoid
1:
name: k400_skeleton #SPECIFIC
loss_weight: 1
gres_ratio: 1 # int, > 0, = Task_GPUs / (world_Size/sum(all_gres_ratios))
backbone:
type: vit_base_patch16_mask
kwargs:
task_sp_list: [ 'rel_pos_h', 'rel_pos_w' ] # wrong list would somehow cause .cuda() stuck without error
pretrained: True
lms_checkpoint_train: fairscale
window: False
test_pos_mode: learnable_interpolate
learnable_pos: False
drop_path_rate: 0.1
vis_patch_token_ratio: 1
vis_label_token_ratio: 0.
dataset:
type: mmSkeletonDataset # train for 150 epochs
kwargs:
ann_file:
- /mnt/path...to.../skaction_public/diving48_hrnet.pkl
- /mnt/path...to.../skaction_public/ucf101_hrnet.pkl
- /mnt/path...to.../skaction_public/k400_hrnet.pkl
dataset_name:
- diving
- ucf
- k400
kp_dim: 2d #SPECIFIC
one_hot: True
num_classes:
- 48
- 101
- 400
centernorm: False
scale_range: [ 0.75,1.25 ]
data_pipeline:
- type: PreNormalize2D
kwargs: { }
- type: GenSkeFeat
kwargs:
dataset: coco
feats: [ 'j' ]
- type: UniformSampleGivenFrames
kwargs:
clip_len: 25
given_len: 7
- type: PoseDecode
kwargs: { }
- type: FormatGCNInput2D
kwargs:
num_person: 2
window: False
rotate: True
mode: zero
- type: Collect
kwargs:
keys: [ 'keypoint', 'label' ]
meta_keys: [ ]
- type: ToTensor
kwargs:
keys: [ 'keypoint' ]
flip: True
sampler:
batch_size: 90 # per card
shuffle_strategy: 1
patch_neck:
type: MAEdecoder_proj_neck
kwargs:
mask_dim: 256
modality: sparse_labeling
label_neck:
type: MAEdecoder_proj_neck
kwargs:
mask_dim: 256
modality: text
patch_adapter:
type: sparse_labeling_adapter_skaction
kwargs:
pretrained: True # should be changed to True later
in_chans: 3
num_joints: 17 #SPECIFIC
num_frames: 175
embed_dim: 768
patch_size: [ 7, 2 ]
stride_level: [ 1, 1 ]
use_abs_pos_emb: True
learnable_pos: False
test_pos_mode: learnable_interpolate
type_embed: False
joint_with_text_embedding: True
joint_names: coco_body_17joints #SPECIFIC
proj_norm: 'LN'
stride_text_embedding: True
is_2d_dataset: True #SPECIFIC
modality_share_list: [
'merge_kernel',
'proj_kernel',
'proj', ]
task_sp_list: [ 'text_embedding', 'pos_embed', ]
patch_proj:
type: sparse_labeling_projector
kwargs:
task: skeleton
loss_cfg:
type: MaskDetFocalDiceLoss
kwargs:
cfg:
deep_supervision: True
focal_alpha: 0.25
class_weight: 2.0
bbox_weight: 5.0
giou_weight: 2.
ign_thr: 0.7
dec_layers: 6
num_classes: 1
predict3d: True
xyxy: True
in_chans: 3 # predefined in patch adapter, set in solver
num_joints: 17 #SPECIFIC
num_frames: 175
modality_share_list: [
'output_proj',
'translate_weight',
'translate_bias',
'post_mul_norm',
'patch_proj',
'class_proj'
]
task_sp_list: [
'text_vectors', # useless
'text_features',
]
label_adapter:
type: text_adapter
kwargs:
pretrained: True
description_dict_name:
- diving48_cls_name
- ucf101_cls_name
- k400_cls_name
one_way_semantics: False
skeleton_action: True # use skeleton action to Double the text embedding (when M=2)
skeleton_action_one_hot_label: True
task_sp_list: [ 'text_vectors', ]
label_proj:
type: text_projector
kwargs:
one_way_semantics: False
description_dict_name:
- diving48_cls_name
- ucf101_cls_name
- k400_cls_name
skeleton_action: True
skeleton_action_one_hot_label: True
pre_proj_type: 'pool'
replace_post_mul_norm: False
post_mul_norm: True
task_sp_list: [ 'text_vectors',
'translate_weight',
'translate_bias',
'post_mul_norm', ]
loss_cfg:
type: CELoss
kwargs:
loss_weight: 1.0
decoder:
type: UniHCPv2_Head
kwargs:
predictor: 'hulk'
task: recons
modality_share_list: [ 'predictor.mask_token', ]
task_sp_list: [
'predictor.query_embed_patch',
'predictor.query_embed_label',
'predictor.class_embed','predictor.fc_bias', # useless in Hulk
] # wrong list would somehow cause .cuda() stuck without error
loss_weight: 1.0
transformer_predictor_cfg:
hidden_dim: 256
num_queries: 20 # useless in Hulk
nheads: 8
dim_feedforward: 2048
dec_layers: 9
pre_norm: False
arch: fan_in
enforce_input_project: False
mask_on: False
num_feature_levels: 1
cross_pos_embed: anchor
self_attn_mask_type: patch_diag_label_row
detach_from_peddet: True
loss_cfg:
type: CEL_Sigmoid
2:
name: smpl
loss_weight: 0.5
gres_ratio: 3
dataset:
type: MeshTSVYamlDataset # train for 150 epochs
kwargs:
is_composite: True
is_train: True
cv2_output: False
augmentation:
scale_factor: 0.25
noise_factor: 0.4
rot_factor: 30
img_res: 224
cfg:
data_path:
- /mnt/path...to.../Processed_SMPL/3dpw/dataset.pkl # problem
- /mnt/path...to.../Processed_SMPL/human3.6m/dataset.pkl #running
- /mnt/path...to.../Processed_SMPL/coco_smpl/dataset.pkl # problem
- /mnt/path...to.../Processed_SMPL/muco/dataset.pkl #running
- /mnt/path...to.../Processed_SMPL/up3d/dataset.pkl # done
- /mnt/path...to.../Processed_SMPL/mpii/dataset.pkl #done
- /mnt/path...to.../Processed_SMPL/gta_human/dataset_pkl/v2_dataset_1396913.pkl
- /mnt/path...to.../Processed_SMPL/gta_human/dataset_pkl/v2_dataset_200000.pkl
- /mnt/path...to.../Processed_SMPL/gta_human/dataset_pkl/v2_dataset_400000.pkl
- /mnt/path...to.../Processed_SMPL/gta_human/dataset_pkl/v2_dataset_600000.pkl
- /mnt/path...to.../Processed_SMPL/gta_human/dataset_pkl/v2_dataset_800000.pkl
- /mnt/path...to.../Processed_SMPL/gta_human/dataset_pkl/v2_dataset_1000000.pkl
- /mnt/path...to.../Processed_SMPL/gta_human/dataset_pkl/v2_dataset_1200000.pkl
root_path:
- /mnt/path...to.../Processed_SMPL/3dpw/images
- /mnt/path...to.../Processed_SMPL/human3.6m/images
- /mnt/path...to.../Processed_SMPL/coco_smpl/images
- /mnt/path...to.../Processed_SMPL/muco/images
- /mnt/path...to.../Processed_SMPL/up3d/images
- /mnt/path...to.../Processed_SMPL/mpii/images
- /mnt/path...to.../Processed_SMPL/gta_human_openxlab/gta_human
- /mnt/path...to.../Processed_SMPL/gta_human_openxlab/gta_human
- /mnt/path...to.../Processed_SMPL/gta_human_openxlab/gta_human
- /mnt/path...to.../Processed_SMPL/gta_human_openxlab/gta_human
- /mnt/path...to.../Processed_SMPL/gta_human_openxlab/gta_human
- /mnt/path...to.../Processed_SMPL/gta_human_openxlab/gta_human
- /mnt/path...to.../Processed_SMPL/gta_human_openxlab/gta_human
sampler:
batch_size: 165 # per card
shuffle_strategy: 1
backbone:
type: vit_base_patch16_mask
kwargs:
task_sp_list: [ 'rel_pos_h', 'rel_pos_w' ] # wrong list would somehow cause .cuda() stuck without error
pretrained: True
lms_checkpoint_train: fairscale
window: False
test_pos_mode: learnable_interpolate
learnable_pos: True
drop_path_rate: 0.2
vis_patch_token_ratio: 1
vis_label_token_ratio: 0.
patch_neck:
type: MAEdecoder_proj_neck
kwargs:
mask_dim: 256
modality: rgb
label_neck:
type: MAEdecoder_proj_neck
kwargs:
mask_dim: 256
modality: sparse_labeling
patch_adapter:
type: rgb_adapter # change to adapter_rgb
kwargs:
pretrained: True
stride_level: 1
in_chans: 3
use_abs_pos_emb: True
learnable_pos: False # useless
test_pos_mode: False
img_size: [ 224, 224 ]
task_sp_list: [ 'pos_embed' ]
label_adapter:
type: sparse_labeling_adapter
kwargs:
pretrained: True
in_chans: 3
num_joints: 446 # 1 + 14 + 431
num_frames: 1
embed_dim: 768
patch_size: [ 1,1 ]
stride_level: [ 1, 1 ]
use_abs_pos_emb: True
learnable_pos: False
test_pos_mode: learnable_interpolate
type_embed: False
proj_norm: 'LN'
task_sp_list: [ 'pos_embed',
'text_embedding',
'proj_kernel',
'proj',]
patch_proj:
type: rgb_projector
kwargs:
loss_cfg:
type: MaskedMSELoss
kwargs:
stride: 1
norm_pix_loss: True
pix_loss: True
pix_loss_weight: 1.
norm_pix_loss_weight: 1.
label_proj:
type: sparse_labeling_projector
kwargs:
task_sp_list: [ 'output_proj',
'text_features',
'loss_fn',
'translate',
'post_mul_norm',
'patch_proj',
'class_proj',
'proj'
]
pre_proj_type: 'fix_text_tokens'
num_classes: 14
# pred_joints_class: True
reference_type: 'smpl'
in_chans: 3 # XYZ
num_joints: 446
num_frames: 1
hidden_dim: 256
patch_size: [ 1, 1 ]
stride_level: [ 1, 1 ]
replace_post_mul_norm: False
task: smpl
# for smpl task, do not predict joints classes, so text_prototype and learn_text is not useful
text_prototype: True
learn_text: True
loss_cfg:
type: SMPL_LOSS_FASTMETRO
kwargs:
# use_pred_joints_class_loss: True
cfg:
use_smpl_param_regressor: True
joints_2d_loss_weight: 100.0
vertices_3d_loss_weight: 100.0
edge_normal_loss_weight: 100.0
joints_3d_loss_weight: 1000.0
vertices_fine_loss_weight: 0.25
vertices_intermediate_loss_weight: 0.50
vertices_coarse_loss_weight: 0.25
edge_gt_loss_weight: 5.0
edge_self_loss_weight: 1.e-4
normal_loss_weight: 0.1
smpl_param_loss_weight: 1000.0
except_smpl_param_loss_weight: 1.e-8
decoder:
type: UniHCPv2_Head
kwargs:
predictor: 'hulk'
task: recons
modality_share_list: ['predictor.mask_token']
task_sp_list: [ 'predictor.query_embed_patch',
'predictor.query_embed_label',
'predictor.class_embed', 'predictor.fc_bias', # useless in Hulk
] # wrong list would somehow cause .cuda() stuck without error
loss_weight: 1.0
transformer_predictor_cfg:
hidden_dim: 256
num_queries: 20
nheads: 8
dim_feedforward: 2048
dec_layers: 9
pre_norm: False
arch: fan_in
enforce_input_project: False
mask_on: False
num_feature_levels: 1
cross_pos_embed: anchor
cls_out_dim: 1
smpl_attention_mask_flag: True
smpl_mae_pe: True
use_adapt_pos2d: True
use_adapt_pos1d: True
self_attn_mask_type: full
adding_per_layer_pe: True
detach_from_peddet: True
use_adapt_position: 'before'
use_smpl_label_attention_mask: True
label_pos_mode: 'smpl_xyz'
loss_cfg:
type: CEL_Sigmoid # useless
3:
name: Peddet
loss_weight: 15
gres_ratio: 8 # int, > 0, = Task_GPUs / (world_Size/sum(all_gres_ratios))
dataset:
type: PedestrainDetectionDataset_v2 # train for 150 epochs
kwargs:
task_spec:
img_folder:
- /mnt/path...to.../PedDet2d/CrowdHuman/Images
ann_file:
- /mnt/path...to.../PedDet2d/CrowdHuman/annotations/train.json
return_masks: False
augmentation:
max_size: 1120
vit: True
num_append_fake_boxes: 867
return_box_xyxy: True
append_z: True
sampler:
batch_size: 4 # per card
shuffle_strategy: 1
batch_accumulation: 1
backbone:
type: vit_base_patch16_mask
kwargs:
task_sp_list: [ 'rel_pos_h', 'rel_pos_w' ] # wrong list would somehow cause .cuda() stuck without error
pretrained: True
lms_checkpoint_train: fairscale
window: False
test_pos_mode: learnable_interpolate
learnable_pos: True
drop_path_rate: 0.2
attn_calcul_method: 'math'
vis_patch_token_ratio: 1
vis_label_token_ratio: 0.
patch_neck:
type: MAEdecoder_proj_neck
kwargs:
mask_dim: 256
modality: rgb
label_neck:
type: MAEdecoder_proj_neck
kwargs:
mask_dim: 256
modality: sparse_labeling
patch_adapter:
type: rgb_adapter
kwargs:
pretrained: True
stride_level: 1
in_chans: 3
learnable_pos: False
use_abs_pos_emb: True
test_pos_mode: interpolate_with_nomask
img_size: 1344 # dynamic input size: TODO: nested
round_padding: True # should fix in rgb
pad_attn_mask: True
task_sp_list: [ 'pos_embed' ]
label_adapter:
type: sparse_labeling_adapter
kwargs:
pretrained: True
in_chans: 3 # xyz
num_joints: 867 # boxes with random gts
num_frames: 2 # 2 for x1y1 and x2y2
embed_dim: 768
patch_size: [ 2, 1 ]
stride_level: [ 1, 1 ]
use_abs_pos_emb: True
learnable_pos: False
test_pos_mode: learnable_interpolate
type_embed: False
proj_norm: 'LN'
task_sp_list: [ 'pos_embed',
'text_embedding',
'proj_kernel',
'proj',
'merge_kernel',
]
patch_proj:
type: rgb_projector
kwargs:
loss_cfg:
type: MaskedMSELoss
kwargs:
stride: 1
norm_pix_loss: True
pix_loss: True
pix_loss_weight: 1.
norm_pix_loss_weight: 1.
label_proj:
type: sparse_labeling_projector
kwargs:
task_sp_list: [ 'text_vectors', # useless
'text_features',
]
modality_share_list: [
'text_vectors', # useless
'output_proj',
'translate_weight',
'translate_bias',
'post_mul_norm',
'patch_proj',
'class_proj'
]
in_chans: 3
num_joints: 867 # boxes with random gts
num_frames: 2 # 2 for x1y1 and x2y2
pre_proj_type: fix_text_tokens
num_classes: 1
reference_type: four_points
box_mlp: True
replace_post_mul_norm: True
translate_weight_scale: 4
text_prototype: True
loss_cfg:
type: MaskDetFocalDiceLoss
kwargs:
cfg:
deep_supervision: True
focal_alpha: 0.25
class_weight: 2.0
bbox_weight: 5.0
giou_weight: 2.
ign_thr: 0.7
dec_layers: 9
num_classes: 1
predict3d: True
xyxy: True
decoder:
type: UniHCPv2_Head
kwargs:
predictor: 'hulk'
task: recons
modality_share_list: ['predictor.mask_token']
task_sp_list: [
'predictor.query_embed_patch',
'predictor.query_embed_label',
'predictor.anchor',
'predictor.class_embed','predictor.fc_bias', # useless in Hulk
] # wrong list would somehow cause .cuda() stuck without error
loss_weight: 1.0
transformer_predictor_cfg:
hidden_dim: 256
num_queries: 20 # useless in Hulk
nheads: 8
dim_feedforward: 2048
dec_layers: 9
pre_norm: False
arch: fan_in
enforce_input_project: False
mask_on: False
num_feature_levels: 1
cross_pos_embed: anchor
patch_pos_mode: interpolate_with_nomask
label_pos_mode: simple_interpolate
self_attn_mask_type: patch_diag_label_row_nested
adding_per_layer_pe: True
mask_token_normal_init: True
intermediate_output: True
peddet_cfgs:
share_content_query: 3
num_queries: 867
pre_defined_path: '289_points_3d.npy'
query_pe_dim: 3
xattn: False
anchor_requires_grad: False
loss_cfg:
type: CEL_Sigmoid
4:
name: Peddet_5set
loss_weight: 42.4
gres_ratio: 20 # int, > 0, = Task_GPUs / (world_Size/sum(all_gres_ratios))
dataset:
type: PedestrainDetectionDataset_v2 # train for 150 epochs
kwargs:
task_spec:
img_folder:
- /mnt/path...to.../peddet_public/CrowdHuman/Images
- /mnt/path...to.../peddet_public/ECP/
- /mnt/path...to.../peddet_public/CityPersons/
- /mnt/path...to.../peddet_public/WiderPerson/Images
- /mnt/path...to.../peddet_public/coco/train2017/
- /mnt/path...to.../peddet_public/WIDER_Pedestrian/Images/
ann_file:
- /mnt/path...to.../peddet_public/CrowdHuman/annotations/train.json
- /mnt/path...to.../peddet_public/ECP/ECP_remove_no_person_img.json
- /mnt/path...to.../peddet_public/CityPersons/CityPersons_remove_no_person_img.json
- /mnt/path...to.../peddet_public/WiderPerson/WiderPerson_remove_no_person_img.json
- /mnt/path...to.../peddet_public/cocopersons/coco_person_remove_no_person_img.json
- /mnt/path...to.../peddet_public/WIDER_Pedestrian/WIDER_Pedestrian_remove_no_person_img.json
return_masks: False
augmentation:
max_size: 1120
vit: True
num_append_fake_boxes: 867
return_box_xyxy: True
append_z: True
sampler:
batch_size: 4 # per card
shuffle_strategy: 1
batch_accumulation: 1
backbone:
type: vit_base_patch16_mask
kwargs:
task_sp_list: [ 'rel_pos_h', 'rel_pos_w' ] # wrong list would somehow cause .cuda() stuck without error
pretrained: True
lms_checkpoint_train: fairscale
window: False
test_pos_mode: learnable_interpolate
learnable_pos: True
drop_path_rate: 0.2
attn_calcul_method: 'math'
vis_patch_token_ratio: 1
vis_label_token_ratio: 0.
patch_neck:
type: MAEdecoder_proj_neck
kwargs:
mask_dim: 256
modality: rgb
label_neck:
type: MAEdecoder_proj_neck
kwargs:
mask_dim: 256
modality: sparse_labeling
patch_adapter:
type: rgb_adapter
kwargs:
pretrained: True
stride_level: 1
in_chans: 3
learnable_pos: False
use_abs_pos_emb: True
test_pos_mode: interpolate_with_nomask
img_size: 1344
round_padding: True # should fix in rgb
pad_attn_mask: True
task_sp_list: [ 'pos_embed' ]
label_adapter:
type: sparse_labeling_adapter
kwargs:
pretrained: True
in_chans: 3 # xyz
num_joints: 867 # boxes with random gts
num_frames: 2 # 2 for x1y1 and x2y2
embed_dim: 768
patch_size: [ 2, 1 ]
stride_level: [ 1, 1 ]
use_abs_pos_emb: True
learnable_pos: False
test_pos_mode: learnable_interpolate
type_embed: False
proj_norm: 'LN'
task_sp_list: [ 'pos_embed',
'text_embedding',
'proj_kernel',
'proj',
'merge_kernel',
]
patch_proj:
type: rgb_projector
kwargs:
loss_cfg:
type: MaskedMSELoss
kwargs:
stride: 1
norm_pix_loss: True
pix_loss: True
pix_loss_weight: 1.
norm_pix_loss_weight: 1.
label_proj:
type: sparse_labeling_projector
kwargs:
task_sp_list: [ 'text_vectors', # useless