Taming Transformers for High-Resolution Image Synthesis

Open paper
Benchmark Model Rank Results
text-to-image-generation-on-lhqcTaming#3Block-FID: 38.89
image-outpainting-on-lhqcTaming#4Block-FID (Right Extend): 22.53Block-FID (Down Extend): 26.38
text-to-image-generation-on-conceptual-captionsVQ-GAN#4FID: 28.86
image-reconstruction-on-ultra-high-resolution-image-reconstructionVQGAN (16x16)#5rFID: 5.95PSNR: 22.91
image-to-image-translation-on-coco-stuff-labels-to-photosVQGAN+Transformer#5FID: 22.4
deepfake-detection-on-fakeavcelebVQGAN#9ROC AUC: 51.8AP: 55.0
image-generation-on-celeba-hq-256x256VQGAN+Transformer#11FID: 10.2
image-to-image-translation-on-ade20k-labels-to-photosVQGAN+Transformer#11FID: 35.5
image-generation-on-celeba-256x256VQGAN#12FID: 10.2
image-reconstruction-on-imagenetTaming-VQGAN (16x16)#22FID: 3.64LPIPS: 0.177PSNR: 19.93SSIM: 0.542
image-generation-on-ffhq-256-x-256VQGAN+Transformer#35FID: 9.6
image-generation-on-imagenet-256x256VQGAN+Transformer (k=600, p=1.0, a=0.05)#186FID: 5.2
image-generation-on-imagenet-256x256VQGAN+Transformer (k=mixed, p=1.0, a=0.005)#188FID: 6.59

Rank counts only results with a code link.