LVIS

LVIS (Large Vocabulary Instance Segmentation) is a large-scale, high-quality dataset designed for instance segmentation tasks. Unlike traditional datasets such as COCO, LVIS contains a much larger number of object categories, specifically focusing on long-tail categories, which are underrepresented in most computer vision datasets. LVIS includes 1203 object categories, with over 2 million instances across more than 50,000 images.