STRAT (Spatial TRAnsformation for virtual Try-on)

Spatial TRAnsformation for virtual Try-on (STRAT) dataset contains three subdatasets: STRAT-glasses, STRAT-hat, and STRAT-tie, which correspond to "glasses try-on", "hat try-on", and "tie try-on" respectively. In each subdataset, the training set has 2000 pairs of foregrounds (accessories) and backgrounds (human faces or portrait images), while the test set has 1000 pairs of foregrounds and backgrounds. For each pairwise sample, both the vertice coordinates and warpping parameters of the foreground for each pairwise are provided for supervised learning and evaluation of spatial transformation.