Datasets:
IPN HandS browser samples
588 paired views from 196 of the 200 source videos. This is a visualization subset, not the full IPN HandS release and not a new dataset or official benchmark. Four videos are excluded because the decoded fps=30 timeline differs from the official annotations. Each remaining video contributes one labeled frame from each temporal third, selecting the nearest usable hand annotation to its center.
Source and credit: Gibran Benitez-Garcia, Jesus Olivares-Mercado, Gabriel Sanchez-Perez and Hiroki Takahashi, IPN HandS: Efficient Annotation Tool and Dataset for Skeleton-Based Hand Gesture Recognition, Applied Sciences 2025. Original IPN Hand: Gibran Benitez-Garcia, Jesus Olivares-Mercado, Gabriel Sanchez-Perez and Keiji Yanai, ICPR 2020/2021.
Changes: frames extracted at 30 FPS using the official timeline; hand x/y labels drawn on the same RGB frame with finger colors. Both views saved as lossless WebP. No new inference, 3D projection, label correction or inferred temporal offsets. Only hand x/y is manually corrected; relative z and auxiliary body pose are not manually verified and are not visualized. Bad annotations and image limitations are retained. Source frame IDs, labels and file hashes are in sample_provenance.json.
This repository hosts the images for the author's independent research browser. It is not affiliated with or endorsed by the dataset authors.
- Downloads last month
- 148