Fix: Normalize category IDs across all splits (train/valid/test)

Fix metadata.jsonl: use objects.bbox + objects.categories format per docs

Add metadata.jsonl with label column for HuggingFace viewer

Remove metadata.jsonl and extra files - use COCO JSON only

Normalize: All IDs now 0-23 (COCO + metadata.jsonl unified)

Show both COCO and metadata.jsonl ID mappings

Fix: Use actual COCO category IDs (not re-indexed)

Update README with category ID to class name mapping

Clean dataset: 24 Front/Back classes only + metadata.jsonl for HF viewer

Fix: Restore accidentally deleted bill annotations (146 removed, 3,759 annotations remain)

Add annotations.parquet for native HuggingFace viewer support

Add COCO dataset loading script for proper HuggingFace viewer support

Delete 3,742 unannotated images (coins, video frames, backgrounds)

Remove unannotated images (coins, video frames) - bills only dataset

Add USD detection dataset - 3,746 annotations, 24 classes, 100% Front/Back classified

Upload README.md with huggingface_hub

initial commit