yslan commited on
Commit
78a08b4
·
verified ·
1 Parent(s): 5b1b77f

Add immutable checkpoint backup wan22_5b_sparse_omni_global_8k resolved_run_config

Browse files
stage1_4d_ablations/wan22_5b_sparse_omni_from_15k/step_global_008000/config.yaml ADDED
@@ -0,0 +1,647 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ output_dir: /scratch/shared/engs2907/logs/vgg_forecast_4d/train/flexpi_four_d_pretraining_wan22_5b_separate_streams_full_parameters_video_depth_camera_dino_and_sparse_tracking_video_from_the_15000_update_four_d_run_20261101/attempt03_continue_from_5000/run
2
+ gpu_model: B200
3
+ batch_size: 8
4
+ num_workers: 8
5
+ learning_rate: 5.0e-05
6
+ num_epochs: 20
7
+ max_steps: 100000
8
+ log_every: 10
9
+ save_every: 1000
10
+ eval_every: 0
11
+ eval_num_inference_steps: 25
12
+ keep_last_n_states: 1
13
+ keep_last_n_weights: 2
14
+ eval_video: false
15
+ share_pca_basis: true
16
+ gradient_accumulation_steps: 4
17
+ mixed_precision: bf16
18
+ seed: 42
19
+ max_grad_norm: 1.0
20
+ weight_decay: 0.01
21
+ resume: null
22
+ resume_reanchor_lr_schedule: false
23
+ resume_config_source: current
24
+ pretrained_ckpt: null
25
+ pretrained_ckpt_strict_shape: false
26
+ wandb:
27
+ enabled: true
28
+ workspace: null
29
+ project: omni4d-robotwin
30
+ name: flexpi_four_d_pretraining_wan22_5b_separate_streams_full_parameters_video_depth_camera_dino_and_sparse_tracking_video_from_the_15000_update_four_d_run
31
+ group: flexpi_four_d_pretraining_wan22_5b_separate_streams_lora_modality_ladder
32
+ mode: offline
33
+ data:
34
+ train:
35
+ _target_: four_d_windows.FourDPretrainingWindows
36
+ sources:
37
+ - name: four_d_pretraining_relaunch_syn4d_sources
38
+ kind: track4d_source_config
39
+ config: train/mot/wan_video_depth_vae_from_video_four_d_pretraining_relaunch_rendered_depth_sources_21_frames_512
40
+ weight: 15.2
41
+ left_out_sources:
42
+ - omniworld_game
43
+ - tartanair
44
+ - mvs_synth
45
+ - name: four_d_pretraining_relaunch_omniworld_game_source
46
+ kind: track4d_source_config
47
+ config: train/mot/wan_video_depth_vae_from_video_four_d_pretraining_relaunch_rendered_depth_sources_21_frames_512
48
+ weight: 8.75
49
+ left_out_sources:
50
+ - syn4d_v1
51
+ - syn4d_v2_mixed
52
+ - syn4d_v2_mixed_no_sim
53
+ - tartanair
54
+ - mvs_synth
55
+ - name: four_d_pretraining_relaunch_synthverse_source
56
+ kind: track4d_source_config
57
+ config: train/mot/wan_video_depth_vae_from_video_four_d_pretraining_relaunch_synthverse_dynamic_replica_vkitti_midair_spring_sources_without_midair_quality_masks_21_frames_512
58
+ weight: 11.05
59
+ left_out_sources:
60
+ - dynamic_replica
61
+ - vkitti
62
+ - midair
63
+ - spring
64
+ - name: four_d_pretraining_relaunch_vkitti_source
65
+ kind: track4d_source_config
66
+ config: train/mot/wan_video_depth_vae_from_video_four_d_pretraining_relaunch_synthverse_dynamic_replica_vkitti_midair_spring_sources_without_midair_quality_masks_21_frames_512
67
+ weight: 6.0
68
+ left_out_sources:
69
+ - synthverse_depth
70
+ - dynamic_replica
71
+ - midair
72
+ - spring
73
+ - name: four_d_pretraining_relaunch_tartanair_sky_then_far_percentile_source
74
+ kind: track4d_source_config
75
+ config: train/mot/wan_video_depth_vae_from_video_four_d_pretraining_relaunch_sky_and_far_percentile_depth_validity_rules_21_frames_512
76
+ weight: 6.0
77
+ left_out_sources:
78
+ - dynamic_replica
79
+ - spring
80
+ - midair
81
+ - mvs_synth
82
+ - tartanground
83
+ - paralleldomain4d
84
+ - name: four_d_pretraining_relaunch_dynamic_replica_stored_depth_source
85
+ kind: track4d_source_config
86
+ config: train/mot/wan_video_depth_vae_from_video_four_d_pretraining_relaunch_sky_and_far_percentile_depth_validity_rules_21_frames_512
87
+ weight: 5.0
88
+ left_out_sources:
89
+ - tartanair
90
+ - spring
91
+ - midair
92
+ - mvs_synth
93
+ - tartanground
94
+ - paralleldomain4d
95
+ - name: four_d_pretraining_relaunch_spring_stored_depth_source
96
+ kind: track4d_source_config
97
+ config: train/mot/wan_video_depth_vae_from_video_four_d_pretraining_relaunch_sky_and_far_percentile_depth_validity_rules_21_frames_512
98
+ weight: 8.0
99
+ left_out_sources:
100
+ - tartanair
101
+ - dynamic_replica
102
+ - midair
103
+ - mvs_synth
104
+ - tartanground
105
+ - paralleldomain4d
106
+ - name: four_d_pretraining_relaunch_midair_within_200_meters_source
107
+ kind: track4d_source_config
108
+ config: train/mot/wan_video_depth_vae_from_video_four_d_pretraining_relaunch_sky_and_far_percentile_depth_validity_rules_21_frames_512
109
+ weight: 3.0
110
+ left_out_sources:
111
+ - tartanair
112
+ - dynamic_replica
113
+ - spring
114
+ - mvs_synth
115
+ - tartanground
116
+ - paralleldomain4d
117
+ - name: four_d_pretraining_relaunch_mvs_synth_1000_then_far_percentile_source
118
+ kind: track4d_source_config
119
+ config: train/mot/wan_video_depth_vae_from_video_four_d_pretraining_relaunch_sky_and_far_percentile_depth_validity_rules_21_frames_512
120
+ weight: 3.0
121
+ left_out_sources:
122
+ - tartanair
123
+ - dynamic_replica
124
+ - spring
125
+ - midair
126
+ - tartanground
127
+ - paralleldomain4d
128
+ - name: four_d_pretraining_relaunch_tartanground_sky_then_far_percentile_source
129
+ kind: track4d_source_config
130
+ config: train/mot/wan_video_depth_vae_from_video_four_d_pretraining_relaunch_sky_and_far_percentile_depth_validity_rules_21_frames_512
131
+ weight: 3.0
132
+ left_out_sources:
133
+ - tartanair
134
+ - dynamic_replica
135
+ - spring
136
+ - midair
137
+ - mvs_synth
138
+ - paralleldomain4d
139
+ - name: four_d_pretraining_relaunch_paralleldomain4d_sky_then_far_percentile_source
140
+ kind: track4d_source_config
141
+ config: train/mot/wan_video_depth_vae_from_video_four_d_pretraining_relaunch_sky_and_far_percentile_depth_validity_rules_21_frames_512
142
+ weight: 10.0
143
+ left_out_sources:
144
+ - tartanair
145
+ - dynamic_replica
146
+ - spring
147
+ - midair
148
+ - mvs_synth
149
+ - tartanground
150
+ frames: 9
151
+ raster_hw:
152
+ - 288
153
+ - 512
154
+ raymap_camera:
155
+ origin_fraction: 0.3333333333333333
156
+ origin_range: 1.0
157
+ decode_margin: 16
158
+ origin_box_grid: 1
159
+ pointmap_log_factor: 0.4
160
+ pointmap_hole_fill:
161
+ synthverse_depth: far_clip
162
+ syn4d: far_clip
163
+ omniworld_game: far_clip
164
+ dynamic_replica: nearest_labelled_depth_of_3x3_neighbours
165
+ tartanair: far_clip
166
+ dl3dv: far_clip
167
+ scannet: telea_log_depth
168
+ virtual_kitti2: far_clip
169
+ midair: far_clip
170
+ pointodyssey: far_clip
171
+ mvs_synth: far_clip
172
+ wildrgbd: telea_log_depth
173
+ hypersim: far_clip
174
+ spring: far_clip
175
+ kubric49: far_clip
176
+ kubric81: far_clip
177
+ droid_stereo_depth: telea_log_depth
178
+ tartanground: far_clip
179
+ paralleldomain4d: far_clip
180
+ syn4d_v2: far_clip
181
+ pointmap_gauge: window_labelled_median_depth
182
+ geometry_image: genception_log_depth
183
+ context_len: 512
184
+ batch_size: 8
185
+ stream_seed: 42
186
+ epoch_windows: 32000000
187
+ val_set_proportion: 0.0
188
+ resume_state: null
189
+ gradient_accumulation_steps: 4
190
+ loader_workers: 8
191
+ loader_opencv_threads: 1
192
+ window_captions:
193
+ caption_dropout: 0.1
194
+ missing_caption: empty_prompt
195
+ caches:
196
+ syn4d_v1_gemini_holo4d:
197
+ embeddings: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/syn4d_v1_gemini_holo4d_local_and_global_with_repaired_globals_20261031/captions_t5_emb.safetensors
198
+ meta: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/syn4d_v1_gemini_holo4d_local_and_global_with_repaired_globals_20261031/captions_t5_emb_meta.json
199
+ syn4d_v2_gemini_holo4d:
200
+ embeddings: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/syn4d_v2_gemini_holo4d_20261030/captions_t5_emb.safetensors
201
+ meta: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/syn4d_v2_gemini_holo4d_20261030/captions_t5_emb_meta.json
202
+ tartanair_gemini_holo4d:
203
+ embeddings: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/tartanair_gemini_holo4d/captions_t5_emb.safetensors
204
+ meta: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/tartanair_gemini_holo4d/captions_t5_emb_meta.json
205
+ mvs_synth_gemini_holo4d:
206
+ embeddings: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/mvs_synth_gemini_holo4d/captions_t5_emb.safetensors
207
+ meta: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/mvs_synth_gemini_holo4d/captions_t5_emb_meta.json
208
+ omniworld_game_gemini_local_official_text_chunks_and_gemini_global:
209
+ embeddings: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/omniworld_game_gemini_local_official_text_chunks_and_gemini_global_20261030b/captions_t5_emb.safetensors
210
+ meta: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/omniworld_game_gemini_local_official_text_chunks_and_gemini_global_20261030b/captions_t5_emb_meta.json
211
+ synthverse_gemini_holo4d:
212
+ embeddings: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/synthverse_gemini_holo4d_local96_and_global_20261031/captions_t5_emb.safetensors
213
+ meta: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/synthverse_gemini_holo4d_local96_and_global_20261031/captions_t5_emb_meta.json
214
+ dynamic_replica_gemini_holo4d:
215
+ embeddings: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/dynamic_replica_gemini_holo4d/captions_t5_emb.safetensors
216
+ meta: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/dynamic_replica_gemini_holo4d/captions_t5_emb_meta.json
217
+ vkitti_gemini_holo4d:
218
+ embeddings: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/vkitti_gemini_holo4d/captions_t5_emb.safetensors
219
+ meta: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/vkitti_gemini_holo4d/captions_t5_emb_meta.json
220
+ midair_gemini_holo4d:
221
+ embeddings: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/midair_gemini_holo4d_20261030/captions_t5_emb.safetensors
222
+ meta: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/midair_gemini_holo4d_20261030/captions_t5_emb_meta.json
223
+ spring_gemini_holo4d:
224
+ embeddings: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/spring_gemini_holo4d/captions_t5_emb.safetensors
225
+ meta: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/spring_gemini_holo4d/captions_t5_emb_meta.json
226
+ tartanground_gemini_holo4d:
227
+ embeddings: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/tartanground_gemini_holo4d_20261030/captions_t5_emb.safetensors
228
+ meta: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/tartanground_gemini_holo4d_20261030/captions_t5_emb_meta.json
229
+ paralleldomain4d_gemini_holo4d:
230
+ embeddings: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/paralleldomain4d_gemini_holo4d_20261030/captions_t5_emb.safetensors
231
+ meta: /scratch/shared/engs2907/logs/vgg_forecast_4d/cache/wan_umt5_captions/four_d_window_captions_512/paralleldomain4d_gemini_holo4d_20261030/captions_t5_emb_meta.json
232
+ stanzas:
233
+ syn4d_v1:
234
+ product: syn4d_v1_gemini_holo4d
235
+ join: local_covering_then_global
236
+ syn4d_v2_mixed:
237
+ product: syn4d_v2_gemini_holo4d
238
+ join: local_covering_then_global
239
+ syn4d_v2_mixed_no_sim:
240
+ product: syn4d_v2_gemini_holo4d
241
+ join: local_covering_then_global
242
+ tartanair:
243
+ product: tartanair_gemini_holo4d
244
+ join: local_covering
245
+ mvs_synth:
246
+ product: mvs_synth_gemini_holo4d
247
+ join: local_covering
248
+ omniworld_game:
249
+ product: omniworld_game_gemini_local_official_text_chunks_and_gemini_global
250
+ join: local_covering_then_second_local_covering_then_global
251
+ pointodyssey_test: none
252
+ kubric49_validation: none
253
+ wildrgbd_test: none
254
+ vkitti_test:
255
+ product: vkitti_gemini_holo4d
256
+ join: local_covering_then_global
257
+ droid_held_out: none
258
+ synthverse_depth:
259
+ product: synthverse_gemini_holo4d
260
+ join: local_covering_then_global
261
+ dynamic_replica:
262
+ product: dynamic_replica_gemini_holo4d
263
+ join: global
264
+ vkitti:
265
+ product: vkitti_gemini_holo4d
266
+ join: local_covering_then_global
267
+ midair:
268
+ product: midair_gemini_holo4d
269
+ join: local_covering_then_global
270
+ spring:
271
+ product: spring_gemini_holo4d
272
+ join: global
273
+ tartanground:
274
+ product: tartanground_gemini_holo4d
275
+ join: local_covering_then_global
276
+ paralleldomain4d:
277
+ product: paralleldomain4d_gemini_holo4d
278
+ join: global
279
+ paralleldomain4d_validation:
280
+ product: paralleldomain4d_gemini_holo4d
281
+ join: global
282
+ training_scene_exclusions:
283
+ reason: rendered depth passes through glass (raw EXR depth through glass panes
284
+ at 190-260 m), owner 10-29; evidence /scratch/shared/engs2907/logs/vgg_forecast_4d/tools/syn4d_glass_depth_check_20261029;
285
+ scifiroom_bald added 10-30 from /scratch/shared/engs2907/logs/vgg_forecast_4d/tools/depth_clipping_audit_20261030
286
+ stanzas:
287
+ syn4d_v1:
288
+ - bigoffice_v1
289
+ - lab_bald
290
+ - office_bald
291
+ - scifiroom_bald
292
+ syn4d_v2_mixed:
293
+ - citynight
294
+ - mall
295
+ - metro
296
+ syn4d_v2_mixed_no_sim:
297
+ - citynight
298
+ - mall
299
+ - metro
300
+ tracking_video:
301
+ query_grid_px: 10
302
+ disc_radius_px: 2
303
+ occluded_style: ring
304
+ ring_radius_px: 3
305
+ ring_width_px: 1
306
+ xyz_normalization: quantile_0.02_0.98
307
+ colour_floor: 0.2
308
+ min_frame_zero_points: 16
309
+ surface_test:
310
+ relative_tolerance: 0.1
311
+ absolute_tolerance: 0.05
312
+ stanza_rules:
313
+ tartanair: static_scene_points
314
+ midair: static_scene_points
315
+ mvs_synth: static_scene_points
316
+ tartanground: static_scene_points
317
+ syn4d_v1: tracked_points
318
+ syn4d_v2_mixed: tracked_points
319
+ syn4d_v2_mixed_no_sim: tracked_points
320
+ omniworld_game: no_tracking_video
321
+ synthverse_depth: no_tracking_video
322
+ dynamic_replica: tracked_points
323
+ vkitti: no_tracking_video
324
+ spring: no_tracking_video
325
+ paralleldomain4d: no_tracking_video
326
+ scannet: static_scene_points
327
+ wildrgbd_test: static_scene_points
328
+ pointodyssey_test: tracked_points
329
+ kubric49_validation: no_tracking_video
330
+ vkitti_test: no_tracking_video
331
+ droid_held_out: no_tracking_video
332
+ paralleldomain4d_validation: no_tracking_video
333
+ dynamic_replica_test: tracked_points
334
+ stanza_track_reading:
335
+ syn4d_v1:
336
+ stanza:
337
+ capability: dynamic_cross_time
338
+ supervision:
339
+ - pointmap
340
+ - source_motion
341
+ annotation_provenance: syn4d_mesh_surface_forward_track_from_frame0
342
+ allow_missing_tracks: false
343
+ kwargs:
344
+ return_tracks: true
345
+ depth_only: false
346
+ metadata_root: /vgg/chenzhao/syn4d_metadata
347
+ wrapped_face_distance_thresh: 0.1
348
+ query_inconsistent_distance_thresh: 0.1
349
+ pointodyssey_test:
350
+ stanza:
351
+ supervision:
352
+ - pointmap
353
+ - source_motion
354
+ kwargs:
355
+ return_tracks: true
356
+ dynamic_replica_test:
357
+ stanza:
358
+ supervision:
359
+ - pointmap
360
+ - source_motion
361
+ kwargs:
362
+ return_tracks: true
363
+ syn4d_v2_mixed:
364
+ loader: syn4d_v2
365
+ stanza:
366
+ capability: dynamic_cross_time
367
+ supervision:
368
+ - pointmap
369
+ - source_motion
370
+ annotation_provenance: syn4d_v2_mesh_surface_forward_track_from_frame0
371
+ allow_missing_tracks: false
372
+ kwargs:
373
+ return_tracks: true
374
+ depth_only: false
375
+ metadata_root: /vgg/chenzhao/syn4d_metadata
376
+ tracking_format: safetensor
377
+ mask_bg: false
378
+ trusted_background_depth_max: 80.0
379
+ wrapped_face_distance_thresh: 0.1
380
+ query_inconsistent_distance_thresh: 0.1
381
+ syn4d_v2_mixed_no_sim:
382
+ loader: syn4d_v2
383
+ stanza:
384
+ capability: dynamic_cross_time
385
+ supervision:
386
+ - pointmap
387
+ - source_motion
388
+ annotation_provenance: syn4d_v2_mesh_surface_forward_track_from_frame0
389
+ allow_missing_tracks: false
390
+ kwargs:
391
+ return_tracks: true
392
+ depth_only: false
393
+ metadata_root: /vgg/chenzhao/syn4d_metadata
394
+ tracking_format: safetensor
395
+ mask_bg: false
396
+ trusted_background_depth_max: 80.0
397
+ wrapped_face_distance_thresh: 0.1
398
+ query_inconsistent_distance_thresh: 0.1
399
+ dynamic_replica:
400
+ stanza:
401
+ supervision:
402
+ - pointmap
403
+ - source_motion
404
+ kwargs:
405
+ return_tracks: true
406
+ model:
407
+ _target_: flexpi_track_video.create_flexpi_track_video
408
+ model_id: Wan-AI/Wan2.2-TI2V-5B
409
+ tokenizer_model_id: Wan-AI/Wan2.1-T2V-1.3B
410
+ tokenizer_max_len: 128
411
+ load_text_encoder: false
412
+ proprio_dim: null
413
+ redirect_common_files: true
414
+ mot_checkpoint_mixed_attn: true
415
+ action_dit_pretrained_path: null
416
+ skip_dit_load_from_pretrain: false
417
+ dino_dim: 768
418
+ dino_model_name: vit_base_patch16_dinov3.lvd1689m
419
+ composite_layout: tshape_robotwin_384x320_uniform_dino_on_video_grid
420
+ dino_cam_patches: null
421
+ dino_cam_regions: null
422
+ dino_temporal_stride: 1
423
+ dino_stride_keep_far: true
424
+ dino_pred_x0: true
425
+ dino_pool_mode: avg
426
+ freeze_dino_encoder: true
427
+ dino_pixel_unshuffle: 2
428
+ enable_pointmap: true
429
+ pointmap_max_depth_m: 2.0
430
+ pointmap_norm_bounds:
431
+ x:
432
+ - -0.5
433
+ - 0.5
434
+ 'y':
435
+ - -0.5
436
+ - 0.5
437
+ z:
438
+ - 0.0
439
+ - 1.5
440
+ pointmap_depth_vis_mode: turbo
441
+ joint_video: true
442
+ joint_dino: true
443
+ joint_pointmap: true
444
+ flex_joint:
445
+ enabled: true
446
+ p_present_video: 0.5
447
+ p_present_dino: 0.5
448
+ p_present_pointmap: 0.5
449
+ p_jv: 0.5
450
+ p_jd: 0.5
451
+ p_jp: 0.5
452
+ cross_modal_predict_video: true
453
+ cross_modal_predict_dino: true
454
+ cross_modal_predict_pointmap: true
455
+ video_dit_config:
456
+ has_image_input: false
457
+ patch_size:
458
+ - 1
459
+ - 2
460
+ - 2
461
+ in_dim: 48
462
+ hidden_dim: 3072
463
+ ffn_dim: 14336
464
+ freq_dim: 256
465
+ text_dim: 4096
466
+ out_dim: 48
467
+ num_heads: 24
468
+ attn_head_dim: 128
469
+ num_layers: 30
470
+ eps: 1.0e-06
471
+ seperated_timestep: true
472
+ require_clip_embedding: false
473
+ require_vae_embedding: false
474
+ fuse_vae_embedding_in_latents: true
475
+ use_gradient_checkpointing: true
476
+ video_attention_mask_mode: first_frame_causal
477
+ action_conditioned: false
478
+ action_dim: 14
479
+ action_group_causal_mask_mode: group_diagonal
480
+ action_dit_config: null
481
+ video_scheduler:
482
+ train_shift: 6.0
483
+ infer_shift: 6.0
484
+ num_train_timesteps: 1000
485
+ action_scheduler:
486
+ train_shift: 1.0
487
+ infer_shift: 1.0
488
+ num_train_timesteps: 1000
489
+ dino_scheduler:
490
+ train_shift: 6.0
491
+ infer_shift: 6.0
492
+ num_train_timesteps: 1000
493
+ pointmap_scheduler:
494
+ train_shift: 6.0
495
+ infer_shift: 6.0
496
+ num_train_timesteps: 1000
497
+ loss:
498
+ lambda_action: 1.0
499
+ lambda_dino: 1.0
500
+ lambda_pointmap: 1.0
501
+ lambda_video: 1.0
502
+ hbridge:
503
+ enabled: true
504
+ bottom_ratio: 0.25
505
+ top_ratio: 0.25
506
+ track_video:
507
+ attention: visual_stream
508
+ sparse_action: true
509
+ train_shift: 6.0
510
+ infer_shift: 6.0
511
+ loss_lambda: 1.0
512
+ action_edge: two_way
513
+ flex:
514
+ p_present: 1.0
515
+ p_joint: 0.5
516
+ cross_modal_predict: true
517
+ noise_level_coupling: shared_uniform
518
+ raymap_camera:
519
+ origin_fraction: 0.3333333333333333
520
+ origin_range: 1.0
521
+ decode_margin: 16
522
+ origin_box_grid: 1
523
+ train_shift: 6.0
524
+ infer_shift: 6.0
525
+ loss_lambda: 1.0
526
+ four_d_pretraining:
527
+ geometry_streams: generated
528
+ dino_stream: generated
529
+ video_loss_lambda: 1.0
530
+ modality_forcing:
531
+ video:
532
+ p_present: 0.5
533
+ p_joint: 0.5
534
+ cross_modal: true
535
+ dino:
536
+ p_present: 0.5
537
+ p_joint: 0.5
538
+ cross_modal: true
539
+ pointmap:
540
+ p_present: 0.5
541
+ p_joint: 0.5
542
+ cross_modal: true
543
+ camera:
544
+ p_present: 0.5
545
+ p_joint: 0.5
546
+ cross_modal: true
547
+ training_mode_shares:
548
+ flexible_joint: 1.0
549
+ v2d: 0.0
550
+ v2c: 0.0
551
+ c2v_from_first_frame: 0.0
552
+ c2vd_from_first_frame: 0.0
553
+ c2v: 0.0
554
+ c2vd: 0.0
555
+ vd2c: 0.0
556
+ vc2d: 0.0
557
+ v2dc: 0.0
558
+ dc2v_from_first_frame: 0.0
559
+ d2vc_from_first_frame: 0.0
560
+ d2v_from_first_frame: 0.0
561
+ dc2v: 0.0
562
+ d2vc: 0.0
563
+ d2v: 0.0
564
+ d2c: 0.0
565
+ c2d: 0.0
566
+ v2d_dino_generated: 0.0
567
+ v2c_dino_generated: 0.0
568
+ v2dc_dino_generated: 0.0
569
+ draw_rules:
570
+ first_frame_anchor: false
571
+ absent:
572
+ video: 0.0
573
+ dino: 0.0
574
+ pointmap: 0.0
575
+ camera: 0.0
576
+ whole_given:
577
+ video: 0.3333333333333333
578
+ dino: 0.3333333333333333
579
+ pointmap: 0.3333333333333333
580
+ camera: 0.3333333333333333
581
+ pointmap_log_factor: 0.4
582
+ tracking_video:
583
+ p_present: 0.5
584
+ p_joint: 0.5
585
+ cross_modal: true
586
+ absent: 0.0
587
+ whole_given: 0.3333333333333333
588
+ conditional_mode_role: generated
589
+ channel_concat: null
590
+ stage_one_without_action_expert: true
591
+ stream_only_end_blocks:
592
+ bottom: 7
593
+ top: 7
594
+ four_d_pretraining_start_weights: /scratch/shared/engs2907/logs/vgg_forecast_4d/train/flexpi_four_d_pretraining_wan22_5b_separate_streams_full_parameters_video_depth_camera_dino_and_sparse_tracking_video_from_the_15000_update_four_d_run_20261101/attempt02/run/checkpoints/adapters/step_005000.safetensors
595
+ minimum_learning_rate: 1.0e-06
596
+ training_scope: four_d_pretraining_wan22_ti2v_5b_separate_streams_full_parameters_video_depth_camera_dino_and_sparse_tracking_video
597
+ started from the 15,000-update stage-1 run's verified FP32 masters (98e01ac6b),
598
+ a fresh optimizer, 5,000 updates of the 10,000-update schedule
599
+ native_control_fps: 50
600
+ action_dimension: 14
601
+ initialization: native_wan_actiondit
602
+ track_video_encoding:
603
+ record_root: /scratch/shared/engs2907/logs/vgg_forecast_4d/data/robotwin_robot_records_all_tasks_7bf528a2_20261023
604
+ embodiment_root: /scratch/shared/engs2907/checkouts/robotwin_objects_b5230d8b_266f3aad/RoboTwin/assets/embodiments/aloha-agilex
605
+ surface_points: 8192
606
+ min_surface_points_per_body: 16
607
+ query_count:
608
+ - 64
609
+ - 256
610
+ camera_query_quota:
611
+ - 224
612
+ - 16
613
+ - 16
614
+ min_visible_queries: 16
615
+ query_draw: per_read
616
+ query_seed: 0
617
+ visibility_margin_m: 0.01
618
+ pixel_center: 0.0
619
+ colouring: frame0_head_xyz
620
+ xyz_normalization: quantile_0.02_0.98
621
+ hues: null
622
+ neighbour_px: null
623
+ colour_floor: 0.2
624
+ min_gap_px:
625
+ - 8
626
+ - 6
627
+ - 6
628
+ wrist_dense_links:
629
+ - fl_link7
630
+ - fl_link8
631
+ - fr_link7
632
+ - fr_link8
633
+ wrist_dense_points: 16384
634
+ disc_radius_px: 2
635
+ occluded_style: ring
636
+ ring_radius_px: 3
637
+ ring_width_px: 1
638
+ deepspeed_gradient_clipping: true
639
+ warmup_steps: 500
640
+ loader_prefetch_factor: 2
641
+ loader_opencv_threads: 1
642
+ weights_milestone_every: 20000
643
+ fp32_milestone_every: null
644
+ train:
645
+ trainable_parameters:
646
+ mode: full
647
+ stop_after_updates: null