HiFaceGAN: Face Renovation via Collaborative Suppression and Replenishment

11 May 2020 Lingbo Yang Chang Liu Pan Wang Shanshe Wang Peiran Ren Siwei Ma Wen Gao

Existing face restoration researches typically relies on either the degradation prior or explicit guidance labels for training, which often results in limited generalization ability over real-world images with heterogeneous degradations and rich background contents. In this paper, we investigate the more challenging and practical "dual-blind" version of the problem by lifting the requirements on both types of prior, termed as "Face Renovation"(FR)... (read more)

PDF Abstract

Datasets


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Image Super-Resolution FFHQ 1024 x 1024 - 4x upscaling HiFaceGAN FID 1.978 # 1
MS-SSIM 0.975 # 1
PSNR 33.04 # 2
SSIM 0.875 # 2
Image Super-Resolution FFHQ 256 x 256 - 4x upscaling HiFaceGAN FID 5.36 # 1
MS-SSIM 0.971 # 1
PSNR 28.65 # 1
SSIM 0.816 # 1
Face Hallucination FFHQ 512 x 512 - 16x upscaling HiFaceGAN FID 11.389 # 1
LPIPS 0.2449 # 1
NIQE 6.767 # 1
Image Super-Resolution FFHQ 512 x 512 - 4x upscaling HiFaceGAN PSNR 30.824 # 1
SSIM 0.838 # 1
MS-SSIM 0.971 # 1
LLE 2.071 # 3
FED 0.0716 # 1
FID 1.898 # 1
LPIPS 0.0723 # 1
NIQE 6.961 # 1

Methods used in the Paper


METHOD TYPE
Adaptive Instance Normalization
Normalization