senc-training-data / README.md
zycliao's picture
Update dataset description for public access
2a9ad2b verified
|
Raw History Blame Contribute Delete
1.48 kB
metadata
language:
  - en
tags:
  - cloth-simulation
  - senc

SENC training data backup

Training inputs found on rsm-gamma for SENC / hood_collision.

Archive prefix Contents
core SMPL motion sequences, garment templates, SMPL models and data splits
simulation-postproc Simulation inputs and material data used by supervised training
simulation-reference Body and cloth trajectories and materials used for validation and preprocessing
pretrained-models Existing baseline weights from trained_models

3837 source files, 35797294108 bytes before compression; 29469918870 bytes in 11 independent archives.

Download and restore:

hf download zycliao/senc-training-data --repo-type dataset --local-dir senc-training-data
cd senc-training-data
bash restore.sh /path/to/data

Requires tar, zstd and sha256sum. Each archive extracts independently into the same destination. Set HOOD_DATA to /path/to/data/hood_data and NC_DIR to /path/to/data/neural_cloth in utils/defaults.py. reference-configs contains the original configuration snapshot; update machine-specific paths before use.

manifest.json lists every source file and its archive. SHA256SUMS verifies the compressed archives. The optional aux_data/validation_sequences directory was absent on the source server. hood_data/hood_postproc is excluded at the account owner's request. Original VTO cloth simulations, experiment outputs, rendered videos and duplicate ZIPs are excluded.