pablovela5620/egoexo-forge-umetrack

EgoExo Forge — UmeTrack RRD

This dataset mirrors the UmeTrack split prepared for EgoExo Forge. Each .rrd file stores a single recording exported with tools/batch_raw_to_rrd.py from the simplecv workspace on November 6, 2025. The conversion preserves the Rerun data stream used by our viewers while applying the standard EgoExo Forge split logic (real/separate_hand/{training,testing}).

Contents

• 2,410 .rrd files covering the complete UmeTrack training/testing users after splitting by hand interaction.
• File names follow 00000000.rrd ordering produced by the batch exporter, making it easy to map indices back to the iterator yielded by UmeTrackSequence.iter_episode_sequences.
• Total payload size ~12 GiB stored with Git LFS.

Usage

pip install rerun-sdk  # or use the pixi "dev" environment
python -m rerun viewer 00001234.rrd

Within simplecv, point UmeTrackConfig.root_directory at the dataset directory on disk and iterate with

from simplecv.data.exoego.umetrack import UmeTrackConfig, UmeTrackSequence
cfg = UmeTrackConfig(root_directory=Path("/path/to/umetrack-split"))
for seq in UmeTrackSequence.iter_episode_sequences(cfg):
    labels = seq.load_labels()

Provenance

The source UmeTrack data lives at /mnt/8tb/data/umetrack-split in the internal storage. The exported RRDs in this repository were generated from that tree without post-processing besides Rerun serialization. For additional context on the EgoExo Forge pipeline, see the companion repository pablovela5620/egoexo-forge-egodex.