11-30 (ICLR 2019) Imagenet-trained cnns are biased towards texture; increasing shape bias improves accuracy and robustness
01-23 (ECCV 2018) Discriminative Region Proposal Adversarial Networks for High-Quality Image-to-Image Translation