123,287 images across training and validation sets featuring object detection, segmentation, and captioning annotations. The collection includes 118,287 training images and 5,000 validation images with labels for 80 distinct object categories.
Use Cases
- Train object detection models using the bbox and category_id fields
- Develop instance segmentation algorithms utilizing the segmentation polygon or RLE data
- Fine-tune image captioning systems using the caption strings provided in the annotations
- Evaluate panoptic segmentation performance by combining stuff and thing category labels
Strengths
- 118,287 training images and 5,000 validation images
- Annotations for 80 object categories including person, bicycle, and car
- Includes per-instance segmentation masks and bounding box coordinates
- Provides five human-written captions per image for multimodal tasks