COCO dataset

COCO (Common Objects in Context) is the gold-standard benchmark for large-scale object detection, segmentation, and image captioning tasks.

COCO is a massive, high-quality computer vision dataset (Common Objects in Context) used to train and benchmark state-of-the-art models. The dataset features over 330,000 images, with more than 200,000 labeled for annotation tasks. It provides rich annotations: 1.5 million object instances across 80 categories, plus 91 'stuff' categories for panoptic segmentation. Each image also includes five captions, and keypoints are provided for 250,000 person instances, making it highly versatile for diverse research and production applications.

https://cocodataset.org/