jeffliulab/card-calibration-v1-data

Card Calibration Dataset (v1)

255 hand-collected photos of color calibration cards, with augmentations, YOLO annotations, and extracted feature crops.

Used to train jeffliulab/card-calibration-v1. Live demo: jeffliulab/card-calibration-v1 Space.

Contents

Archive	Files	Description
raw_photos.tar.gz	255	Original hand-collected photos
augmented_images.tar.gz	2,294	Albumentations-augmented variants (brightness/hue/blur/etc)
feature_crops.tar.gz	9,154	Center-region patch crops (4 patches × ~2,294 images)
yolo_card_dataset.tar.gz	510	YOLO stage-1 training set (card detection)
yolo_pattern_dataset.tar.gz	510	YOLO stage-2 training set (pattern detection)
yolo_labeled_card.tar.gz	510	Raw YOLO annotations (stage 1)
yolo_labeled_patterns.tar.gz	510	Raw YOLO annotations (stage 2)
generalization_test.png	1	Held-out generalization test photo
features/feature_0216.csv	1	Final training feature CSV (12-D + target)
features/GridMean_0216.csv	1	Intermediate grid mean values
features/unique_colors.csv	1	Unique color reference values

Usage

The companion repo provides a one-command download script:

git clone https://github.com/jeffliulab/Color_Calibration.git
cd Color_Calibration
python scripts/dataset_download.py

This downloads all archives and extracts them into data/ matching the layout expected by the training scripts in src/.

Manual download

from huggingface_hub import snapshot_download
snapshot_download(
    repo_id="jeffliulab/card-calibration-v1-data",
    repo_type="dataset",
    local_dir="./dataset_cache",
)

Known issues

6 augmented feature crop files are missing (gsutil transfer failures during the migration from GCS). These are non-critical center-region crops and do not affect training or evaluation.

License

MIT — both data and code are freely available for research and commercial use.