The dataset viewer is not available for this subset.
Exception: SplitsNotFoundError
Message: The split names could not be parsed from the dataset config.
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
for split_generator in builder._split_generators(
~~~~~~~~~~~~~~~~~~~~~~~~~^
StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 84, in _split_generators
raise ValueError(
...<2 lines>...
)
ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
for split in get_dataset_split_names(
~~~~~~~~~~~~~~~~~~~~~~~^
path=dataset,
^^^^^^^^^^^^^
config_name=config,
^^^^^^^^^^^^^^^^^^^
token=hf_token,
^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
info = get_dataset_config_info(
path,
...<6 lines>...
**config_kwargs,
)
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
MSR-3D Data
Processed scenes, observation pools, instance representations and evaluation data. 16 archives, covering 4 LERF scenes and 10 3DOVS scenes.
| Download | Contents |
|---|---|
archives/scenes/<dataset>/<scene>.tar.gz |
Gaussian PLY, instance labels, segmentation masks, instance renders, configuration and processed camera data. 3DOVS scenes also include processed RGB. |
archives/precomputed.tar.gz |
Observation pools, text/anchor features and instance representations for all 14 scenes. |
archives/evaluation.tar.gz |
Query lists, processed GT/masks, annotations and cached matching references. |
LERF67 has 67 queries including the Waldo Kitchen supervision scene. LERF45 is the 45-query exact-string-unseen subset across three target scenes. 3DOVS61 has 61 scene/query pairs across 10 scenes. Evaluation uses scene-macro averages; query-macro results are also retained. Details are in the extracted evaluation/cached-matcher-reference/PROTOCOL.md.
Official inputs
Obtain the following separately and retain their upstream terms:
- LERF RGB and COLMAP: expanded LERF-OVS download, from LangSplat. Segment then Splat's official release is an alternative. Use
figurines,ramen,teatime,waldo_kitchen, each withimages/andsparse/0/{cameras,images,points3D}.bin. - CLIP: official ViT-L/14 at 336 px checkpoint, from OpenAI CLIP.
- MasQCLIP: exact
base_novel.pthcheckpoint, linked by OpenSplat3D. Original model release. - Original 3DOVS collection: official download, from 3D-OVS. The processed 3DOVS scene archives already supply the RGB needed for inference.
- SAM checkpoints, for preprocessing: official model downloads. The released scene masks are already prepared.
Prepare
Download this repository to a local directory:
hf download KaiLai/MSR-3D-Data --repo-type dataset --local-dir ./MSR-3D-Data
cd MSR-3D-Data
python tools/prepare.py \
--output ../msr3d-data \
--lerf-root /path/to/lerf_ovs \
--clip-weight /path/to/ViT-L-14-336px.pt \
--masqclip-weight /path/to/base_novel.pth
Use Python 3.11+ and a new output directory. The script checks archive hashes and upstream input versions, copies external files unchanged, and automatically creates the inference code's local file inventory.
For one scene, download its scene archive and the evaluation archive, then add --scene lerf67/figurines --mode scene. For --scene 3dovs61/bed --mode scene, --lerf-root is unnecessary. Keep SHA256SUMS and tools/prepare.py with the downloads.
For the supplied pools/representations and evaluation data only:
python tools/prepare.py --mode precomputed --output ../msr3d-precomputed
Use
Use MSR-3D v1.2.0 with data revision v1.1.0. The matching weights are unchanged; 3DOVS now uses geometric top-five anchors with middle-crop features. LERF retains its three-crop anchors. With the code installed:
python -m msr3d.pipeline --data-root ../msr3d-data \
--dataset lerf67 --scene figurines --output outputs/figurines
Add --pool ../msr3d-data/precomputed/lerf67/figurines/pool to reuse a supplied observation pool.
See LICENSE.md for component terms. Please cite MSR-3D and the upstream datasets/methods linked above. Gaussian processing follows OpenSplat3D; the original LERF collection is described at lerf.io.
Reference results
Scene-macro mIoU / mF1 (%): LERF67 64.7033 / 72.6345, LERF45 63.1485 / 72.8309, 3DOVS61 80.9799 / 83.0238. The 3DOVS input revision was evaluated on the target benchmark; it is not an independent held-out configuration selection. Previous data revisions remain available in repository history.
- Downloads last month
- 72