UCSD

The dataset contains 2,000 frames, in which the training set captures 601 to 1,400 and the testing set owns the remaining.

BibTex: