AGD20K

The AGD20K dataset provides a collection of 20,061 images captured from a third-person perspective (exocentric) and 3,755 images from a first-person perspective (egocentric). These images are annotated with labels for 36 different affordances, which represent the potential object interactions.