14,197,122 images organized into 21,841 WordNet synset categories for visual object recognition. The collection includes over 1 million images with hand-annotated bounding boxes and spans a diverse range of categories from natural organisms to human-made artifacts.
Use Cases
- Train deep learning models for image classification using the synset_id and image_data
- Benchmark object localization tasks using the bounding_box coordinates and class_label fields
- Research hierarchical feature learning by utilizing the WordNet parent-child relationships between categories
Strengths
- 14,197,122 total images across 21,841 distinct categories
- 1,034,908 images featuring manually verified bounding box annotations
- Organized using the WordNet hierarchy with synset IDs as primary labels