Add radr.py datasets loader (referenced in README)

Update datacard: WACV 2027, corrected baselines/tasks, Croissant 1.0

Update datacard: WACV 2027, corrected baselines/tasks, Croissant 1.0

Add yaw2 raw exemplar (10 deg misalignment, ~37 GB) for reviewer inspection of fault data + TI cascade .bin format

Remove healthy raw exemplar; will replace with yaw2 (more representative of fault induction)

Add raw exemplar capture (healthy, ~36 GB) for reviewer inspection of TI cascade .bin format

Croissant: add NeurIPS-required RAI fields (data sources, provenance, preprocessing protocol)

Pin viewer to metadata.parquet (auto-detect was picking up training_cache.h5 — wrong schema, no IMU/sensor columns)

Drop configs block; let HF auto-detect metadata.parquet (removes Subset/Split dropdowns)

Add inline RD-map PNG preview per frame; viewer renders thumbnails

Tags: tighten task_categories to image-classification; expand tags with raw-iq, complex-signal, mimo, ti-mmwcas-rf-evm

Fix task_categories: image-classification + tabular-classification (was audio + time-series)

Fix: shuffle parquet rows + correct camera spec to Intel RealSense D435 (croissant.json)

Fix: shuffle parquet rows + correct camera spec to Intel RealSense D435 (README.md)

Fix: shuffle parquet rows + correct camera spec to Intel RealSense D435 (metadata.parquet)

Remove radr.py loader; HF viewer rejects custom scripts (uses configs/Parquet instead)

Enrich metadata.parquet with per-frame IMU/temp/GPS sensor traces

README: point viewer at metadata.parquet

Add metadata.parquet for HF dataset viewer (labels-only, 1800 rows)

Add training_cache.h5 (training cache: 1800 frames RD + raw IQ)

Initial v1.0 dataset card + sensor-sync HDF5s + Croissant metadata

initial commit