Imagenet-1M-trained patch-based results

Display Order: Input Context, Ground Truth, Generated Inpainting