cubec commited on
Commit
a177834
·
verified ·
1 Parent(s): 9f8629b

Register filtered Omni6DPose SOPE training components

Browse files
Files changed (2) hide show
  1. README.md +66 -0
  2. training/index.json +77 -0
README.md CHANGED
@@ -313,3 +313,69 @@ Scene downloads remain separate from object targets and default task downloads.
313
 
314
  Historical MessyKitchens mask archives now reside under [_deprecated/messykitchen/original_masks/evaluation](https://huggingface.co/datasets/cubec/PartProbe3D-preprocessed/tree/main/_deprecated/messykitchen/original_masks/evaluation).
315
  Current evaluation downloads exclude these archives.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
313
 
314
  Historical MessyKitchens mask archives now reside under [_deprecated/messykitchen/original_masks/evaluation](https://huggingface.co/datasets/cubec/PartProbe3D-preprocessed/tree/main/_deprecated/messykitchen/original_masks/evaluation).
315
  Current evaluation downloads exclude these archives.
316
+
317
+ <!-- partprobe3d-omni6dpose:start -->
318
+ ## Omni6DPose SOPE
319
+
320
+ This release contains 3 selected frames per SOPE sequence.
321
+ It contains 55,848 usable views from 18,617 sequences.
322
+ The visibility requirement excludes 3 source views.
323
+ The official splits contain 44,788 training views and 11,060 test views.
324
+ All selected frames use one dataset and the same component paths.
325
+
326
+ Every retained mask covers strictly more than 0.1% of the native image.
327
+ At 640 by 480 pixels, each retained object covers at least 308 pixels.
328
+ Each view contains at least two retained objects.
329
+ The observation, object, SS, and scene components use this selection.
330
+
331
+ Native RGB, perfect metric depth, instance labels, and calibration come from SOPE.
332
+ The observation bundle stores these arrays in sample NPZ files.
333
+ Targets contain surface voxel occupancy at resolution 64.
334
+ PointCloudExtent uses near and far trim fractions of 0.0525, far margin 0.125, and padding 1.31.
335
+ Object latents and scene latents use resolution 32.
336
+ SS targets contain native latents with shape `8×16×16×16`.
337
+
338
+ | Component | Views | Transfer, GB |
339
+ | --- | ---: | ---: |
340
+ | [training/observations/omni6dpose](training/observations/omni6dpose/index.json) | 44,804 | 70.945 |
341
+ | [training/shape/32/omni6dpose](training/shape/32/omni6dpose/index.json) | 44,788 | 1.739 |
342
+ | [training/ss/omni6dpose](training/ss/omni6dpose/index.json) | 55,848 | 145.935 |
343
+ | [optional/held_out/omni6dpose](optional/held_out/omni6dpose/index.json) | 11,044 | 17.792 |
344
+ | [optional/geometry/omni6dpose](optional/geometry/omni6dpose/index.json) | 55,848 | 12.039 |
345
+ | [optional/scene_mask_area/32/omni6dpose](optional/scene_mask_area/32/omni6dpose/index.json) | 44,788 | 1.573 |
346
+
347
+ The core observations include all training views and 16 test views for monitoring.
348
+ The held-out component adds the remaining 11,044 test views.
349
+ The SS component contains all 55,848 views.
350
+ Optional geometry contains canonical GLBs and per-frame instance transforms.
351
+ Filtered scene latents occupy the same namespace as the other filtered datasets.
352
+
353
+ Restore all training components with this command:
354
+
355
+ ```bash
356
+ python -m preprocess.huggingface download --repo cubec/PartProbe3D-preprocessed --task all-training --dataset omni6dpose --cleanup
357
+ ```
358
+
359
+ Restore optional components with these commands:
360
+
361
+ ```bash
362
+ python -m preprocess.huggingface download --repo cubec/PartProbe3D-preprocessed --dataset omni6dpose --bundle-root optional/held_out --cleanup
363
+ python -m preprocess.huggingface download --repo cubec/PartProbe3D-preprocessed --dataset omni6dpose --bundle-root optional/geometry --cleanup
364
+ python -m preprocess.huggingface download --repo cubec/PartProbe3D-preprocessed --dataset omni6dpose --bundle-root optional/scene_mask_area/32 --cleanup
365
+ ```
366
+
367
+ Use project Git revision `513895b85af6e56716e1ed51d9352639a27e8e8d` or a later revision.
368
+ The committed code patch is available at [code/omni6dpose.patch](code/omni6dpose.patch).
369
+ Apply this patch to project Git revision `32d92a1e1b9e99eafaac9dda13c43e676a7aeeb3` when necessary.
370
+ Model weights remain separate from this dataset.
371
+ Component manifests preserve model hashes, source hashes, and generation settings.
372
+
373
+ The shape encoder is `TRELLIS.2 shape_enc_next_dc_f16c32_fp16`.
374
+ Its safetensors SHA-256 is `f37c5ff5b983b68e9946060000f09bc131f3e84318a2c8b7430a81e4b4636c41`.
375
+ The paired Pixal3D decoder SHA-256 is `e3b718d3e43e4f8780e9a24ac6fff231811a67e3b058e336e10fe654c911d581`.
376
+
377
+ Preview: [RGB, masks, extent, meshes, and voxels](previews/omni6dpose/extent_voxels.png).
378
+ The [release inventory](releases/omni6dpose/index.json) records source identities and component checksums.
379
+ Dataset credit: [Omni6DPose](https://jiyao06.github.io/Omni6DPose/).
380
+ Source documentation: [Omni6DPoseAPI](https://github.com/Omni6DPose/Omni6DPoseAPI).
381
+ <!-- partprobe3d-omni6dpose:end -->
training/index.json CHANGED
@@ -92,6 +92,50 @@
92
  "target_scope": "masked_objects_complete_occupancy"
93
  },
94
  "shape_excluded_views": []
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
95
  }
96
  },
97
  "optional": {
@@ -147,6 +191,39 @@
147
  "views": 42256,
148
  "resolution": 32,
149
  "source_manifest_sha256": "08bc7afbadab1fe27bf8ba1c039b428cebe582937ad18b8332f698295b2189e2"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
150
  }
151
  },
152
  "base_git_revision": "aac7b493e1e5e10512857a2582233c99fedda9c5",
 
92
  "target_scope": "masked_objects_complete_occupancy"
93
  },
94
  "shape_excluded_views": []
95
+ },
96
+ "omni6dpose": {
97
+ "source_manifest_sha256": "51a2acd982c2a898d4ead7836709d6d59b7de549d79f107565e6e8f81988e089",
98
+ "source_views": 55848,
99
+ "views": 44804,
100
+ "resolution": 64,
101
+ "training_views": 44788,
102
+ "monitor_views": 16,
103
+ "required": [
104
+ "training/observations/omni6dpose",
105
+ "training/shape/32/omni6dpose",
106
+ "training/ss/omni6dpose"
107
+ ],
108
+ "bytes": 218618449920,
109
+ "mask_policy": {
110
+ "version": "visible-mask-area-v1",
111
+ "minimum_area_fraction": 0.001,
112
+ "comparison": "strictly greater",
113
+ "resolution": "native"
114
+ },
115
+ "shape_excluded_views": [],
116
+ "components": [
117
+ {
118
+ "path": "training/observations/omni6dpose",
119
+ "views": 44804,
120
+ "bytes": 70944645120
121
+ },
122
+ {
123
+ "path": "training/shape/32/omni6dpose",
124
+ "views": 44788,
125
+ "bytes": 1738895360
126
+ },
127
+ {
128
+ "path": "training/ss/omni6dpose",
129
+ "views": 55848,
130
+ "bytes": 145934909440
131
+ }
132
+ ],
133
+ "frames_per_sequence": 3,
134
+ "source_frames": 55851,
135
+ "git_revision": "513895b85af6e56716e1ed51d9352639a27e8e8d",
136
+ "source_sequences": 18617,
137
+ "test_views": 11060,
138
+ "encoded_objects": 628348
139
  }
140
  },
141
  "optional": {
 
191
  "views": 42256,
192
  "resolution": 32,
193
  "source_manifest_sha256": "08bc7afbadab1fe27bf8ba1c039b428cebe582937ad18b8332f698295b2189e2"
194
+ },
195
+ "held_out/omni6dpose": {
196
+ "component": "optional/held_out/omni6dpose",
197
+ "bytes": 17792389120,
198
+ "views": 11044,
199
+ "mask_policy": {
200
+ "version": "visible-mask-area-v1",
201
+ "minimum_area_fraction": 0.001,
202
+ "comparison": "strictly greater",
203
+ "resolution": "native"
204
+ }
205
+ },
206
+ "geometry/omni6dpose": {
207
+ "component": "optional/geometry/omni6dpose",
208
+ "bytes": 12039178240,
209
+ "views": 55848,
210
+ "mask_policy": {
211
+ "version": "visible-mask-area-v1",
212
+ "minimum_area_fraction": 0.001,
213
+ "comparison": "strictly greater",
214
+ "resolution": "native"
215
+ }
216
+ },
217
+ "scene_mask_area/32/omni6dpose": {
218
+ "component": "optional/scene_mask_area/32/omni6dpose",
219
+ "bytes": 1572689920,
220
+ "views": 44788,
221
+ "mask_policy": {
222
+ "version": "visible-mask-area-v1",
223
+ "minimum_area_fraction": 0.001,
224
+ "comparison": "strictly greater",
225
+ "resolution": "native"
226
+ }
227
  }
228
  },
229
  "base_git_revision": "aac7b493e1e5e10512857a2582233c99fedda9c5",