VisDA stands for Visual Domain Adaptation, which is a dataset for domain adaptation from simulator to real environment. The three domains of training, validation and testing contain 280,000 images in 12 categories. The training images are generated by the same object in different situations, and the validation images are from MSCOCO.

This dataset aims to address the real-world problem of domain transfer, that is, training machine learning models to transfer from one domain to another and adapt without additional supervision.

Example image: