A million examples do not automatically make a reliable benchmark: here, the scale is useful for training, but the captions should not be treated as verified ground truth. The collection is most valuable as a large pool of varied sprite concepts, with its quality caveat kept in view.
AI Dataset2026
Pixel Art Atlas — 1.4M Synthetic Sprites
Provides about 1.41 million 16×16 transparent-background game sprites with English captions for text-to-image training and experiments. The images are AI-generated, and caption-to-image agreement has not been systematically verified.
Introduction
Information
- Websitehuggingface.co
- AuthorsAngais
- Published date2026/11/06
Categories
More Items
AI Dataset2024
Provides hundreds of millions of chess positions evaluated by Stockfish, with scores and principal variations. Useful for studying engine evaluations and building chess-related machine-learning datasets.
AI Dataset2026
Turns 66,887 Objaverse assets into executable sand3d Python programs, each paired with an OpenUSD file and render. The structured geometry and articulation data support 3D program synthesis and multimodal model training.
AI Dataset2026
Aggregates terminal half-life observations from biomedical papers while preserving route, cohort, dose, and source-statistic context. Useful for pharmacokinetic analysis and biomedical information-extraction research.