Search for a command to run...
ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness