## 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.