Register aligned scene latent components

README.md

@@ -274,7 +274,7 @@ The image encoders compute features during training.
 <!-- partprobe3d-ss-targets:end -->
 
 <!-- partprobe3d-scenes:start -->
-## Voxel-aligned 32³ scene latents
+## Voxel-aligned scene latents
 
 These optional caches contain one token per saved reference voxel.
 Overlapping objects share that token and retain separate membership labels.
@@ -282,11 +282,12 @@ The `latent_valid` flag identifies cells with a native encoder output.
 Invalid cells retain their reference coordinates and contain zero latent values.
 The archives preserve the original NPZ files, calibration transforms, and codec provenance.
 
-| Dataset | Training views | Transfer, GB |
-| --- | ---: | ---: |
-| [3dfront](optional/scene/32/3dfront/index.json) | 39,930 | 4.590 |
-| [sage](optional/scene/32/sage/index.json) | 17,456 | 1.176 |
-| [foundationpose](optional/scene/32/foundationpose/index.json) | 42,256 | 5.046 |
+| Dataset | Grid | Training views | Transfer, GB |
+| --- | ---: | ---: | ---: |
+| [3dfront](optional/scene/32/3dfront/index.json) | 32³ | 39,930 | 4.590 |
+| [sage](optional/scene/32/sage/index.json) | 32³ | 17,456 | 1.176 |
+| [foundationpose](optional/scene/32/foundationpose/index.json) | 32³ | 42,256 | 5.046 |
+| [omni6dpose](optional/scene/64/omni6dpose/index.json) | 64³ | 44,788 | 5.469 |
 
 Restore the matching observation bundles before reading scene targets.
 The `slat-seg` task downloads those observations without SS or object latent targets.
@@ -301,9 +302,10 @@ Restore the selected scene caches:
 python -m preprocess.huggingface download --repo cubec/PartProbe3D-preprocessed --dataset 3dfront --bundle-root optional/scene/32 --cleanup
 python -m preprocess.huggingface download --repo cubec/PartProbe3D-preprocessed --dataset sage --bundle-root optional/scene/32 --cleanup
 python -m preprocess.huggingface download --repo cubec/PartProbe3D-preprocessed --dataset foundationpose --bundle-root optional/scene/32 --cleanup
+python -m preprocess.huggingface download --repo cubec/PartProbe3D-preprocessed --dataset omni6dpose --bundle-root optional/scene/64 --cleanup
 ```
 
-The restored caches reside under `datasets/<dataset>/slat/32/scene_targets/`.
+The restored caches reside under `datasets/<dataset>/slat/<resolution>/scene_targets/`.
 `dataloader.slat_scene.SceneTargetDataset` reads them with their matching saved observations.
 Raw meshes and source datasets are unnecessary for this reader.
 Scene downloads remain separate from object targets and default task downloads.

training/index.json

@@ -224,6 +224,19 @@
         "comparison": "strictly greater",
         "resolution": "native"
       }
+    },
+    "scene/64/omni6dpose": {
+      "component": "optional/scene/64/omni6dpose",
+      "bytes": 5469204480,
+      "views": 44788,
+      "resolution": 64,
+      "source_manifest_sha256": "7ffa8b2b7ccd7ae0dcf2c6d966c04efb9e663930d07ac3beb63d37ac0a4e336b",
+      "mask_policy": {
+        "version": "visible-mask-area-v1",
+        "minimum_area_fraction": 0.001,
+        "comparison": "strictly greater",
+        "resolution": "native"
+      }
     }
   },
   "base_git_revision": "aac7b493e1e5e10512857a2582233c99fedda9c5",