TACO

TACO (Trash Annotations in Context) is a significant initiative supporting the development of artificial intelligence for environmental purposes. The project recognizes the potential of AI in areas like drone-based trash surveying, robotic litter collection, and anti-littering surveillance. Achieving accurate trash detection with deep learning models requires extensive annotated image datasets, which TACO aims to provide and continuously enhance.

All images and annotations within the TACO dataset are distributed under free copyright licenses. Specifically, the annotations are licensed under CC BY 4.0, while the images may carry different public licenses depending on their origin. Each image's license and original URL are detailed in the accompanying annotation file, with CC BY 4.0 serving as the default if no specific license is listed. The dataset encourages contributions, with submissions released weekly to its GitHub repository and subsequently reviewed for quality before official inclusion.

The project emphasizes the importance of high-quality segmentation over precise object labeling, as annotations are reviewed by experts. Users are encouraged to segment in high-resolution mode and handle occlusions by creating clean masks or multiple polygons per object when necessary. TACO also offers an exploration tool to guide contributors on segmentation quality and provides a mailing list for updates, accessible after a single annotation submission.