ImageNet-trained CNNs are not biased towards texture: Revisiting feature reliance through controlled suppression

Open in new window