11,530 images containing 27,450 annotated object instances across 20 standardized categories. Annotations provide bounding box coordinates, pixel-level segmentation masks, and action labels for object detection and classification.
Use Cases
- Train object detection systems using the 'bndbox' and 'class' attributes
- Perform semantic segmentation using the 'SegmentationClass' and 'SegmentationObject' image masks
- Benchmark image classification models using the 'pose' and 'truncated' attributes
Strengths
- 20 object classes including 'person', 'bicycle', 'car', and 'dog'
- XML annotations providing 'xmin', 'ymin', 'xmax', and 'ymax' bounding box coordinates
- Pixel-level segmentation masks provided for 2,913 images
- Metadata flags for 'truncated', 'occluded', and 'difficult' instances