9 million image URLs across training, validation, and test sets annotated with four distinct computer vision labels. The data includes image-level labels, bounding boxes, object segmentation masks, and visual relationships curated by Google LLC.
Use Cases
- Train object detection models using the bounding box annotations
- Develop instance segmentation algorithms leveraging the object segmentation masks
- Build visual relationship detection systems to identify interactions between objects based on relationship labels
- Perform large-scale image classification using the image-level labels
Strengths
- 9 million image URLs across training, validation, and test splits
- Four annotation types: image-level labels, bounding boxes, object segmentation masks, and visual relationships
- Curated by Google LLC under a CC BY 2.0 license
- Provides image URLs as the primary data source for image retrieval