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/harbor/harbor.py", line 171, in _split_generators
raise DataFilesNotFoundError("No task.toml or instruction.md files found")
datasets.exceptions.DataFilesNotFoundError: No task.toml or instruction.md files found
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.
MouseGeoForge registered experiment data
Public transfer mirror for current geometry and visible-mouse segmentation reproduction. This is a byte-preserving mirror of existing registered inputs, not a new benchmark release or permission to change splits.
Completeness: do not start from a partial mirror. COMPLETION.json is published only after all archive uploads have matching remote content hashes and all 131,440 unique pinned input files are covered. The original geometry verifier checks 130,104 files; the remaining 1,336 are manual segmentation masks. Pinned inputs expand to 155,255,460,993 bytes. Archive membership can include supporting metadata beyond these pins.
Download on the cluster
This repository is currently public; no login or token is required. Download with:
hf download EtherealHorizon/MouseGeoForge-experiment-data-20261023 --repo-type dataset --local-dir /grid/hou/data/datasets/MouseGeoForge-HF
cd /grid/hou/data/datasets/MouseGeoForge-HF
sha256sum -c SHA256SUMS.txt
Do not confuse download completion with installation into the experiment workspace. Every archive has repo-relative paths, usually beginning dataset/ or output/. Each numbered part is an independent gzip tar archive; never concatenate parts. Extract into a fresh empty repository-shaped staging root:
mkdir -p /chosen/empty/MouseGeoForge-inputs
for archive in archives/*.tar.gz; do
tar -xzf "$archive" -C /chosen/empty/MouseGeoForge-inputs
done
The result contains dataset/DANNCE-MarkerlessMouse1-Core23-RealMultiview, the exact admitted synthetic and PyMouseStudio files, and output/... rendered truth/manifests. Install/link these under the existing repo's dataset/output mappings. Do not blindly extract into a dataset root and create an extra dataset/ directory. Preserve existing correct files. Use all_data_file_pins.json for the complete expected byte lengths and SHA256 hashes; original geometry pins and split manifests are also preserved inside the core archive. Re-run the original cluster verifier before experiments.
Contents and evidence
- Original nine DANNCE1 archive parts, including
sessions/markerless_mouse_1/pseudo_depth_full_v1. - Exact synthetic replay/geometry files, including both MoSeq ranges and CRIM13 replay, plus registered crossed-render RGB/truth and calibration.
- Required PyMouseStudio images, annotations and manual segmentation masks.
- Core and legacy geometry reproduction manifests, render configs, depth/pose diagnostics, and SHA256 coverage lists.
reproduction/contains the original segmentation item-level train/dev assignment and protocol, source-solver protocol, and H100 planning documents.
DANNCE1 depth is calibrated SAM2 visual-hull pseudo-depth, not measured GT or 3DGS GT. Real captures share one known calibrated physical rig and do not establish unseen-real-hardware generalization. Preserve all existing source, split and sealed/exposure rules. Copying a held-out payload does not authorize training or model scoring on it. No water-reaching or new external-data admission is introduced by this transfer.
The repository contains data, not the released VGGT weights, historical optimizer checkpoints, generated feature caches, or every old experimental output. Obtain separately the exact model dependency recorded in the experiment protocol; feature caches can be regenerated under that protocol. Upstream provenance/metadata are preserved in the original packages. No new redistribution license is asserted here.
A Hugging Face upload does not by itself copy or install files at /grid/hou/data/datasets/. Download and pin verification must still run on the cluster. This transfer starts no training.
- Downloads last month
- 513