Places365
- 2nd generation of the Places databas
- high-level visual understanding tasks, such as scene context, object recognition, action and event prediction, and theoryof-mind inference
- more than 10 million images covering more than 400 classes and 5, 000 to 30, 000 training images per class.