ImageNet trained CNNs are biased towards recognising textures rather than shapesopenreview.net21 pointsgiuliomagnifico5 years ago