Awesome
<!--A curated list of resources for Deep Face Restoration--> <!-- PROJECT LOGO --> <p align="center"> <h3 align="center">Deep Face Restoartion: Denoise, Super-Resolution, Deblur and Artifact Removal </h3> <p align="center">A comprehensive list of resources for Deep Face Restoartion <br /> <br /> <br /> <a href="https://github.com/TaoWangzj/Awesome-Face-Restoration/issues/new">Suggest new item</a> <br /> <a href="https://github.com/TaoWangzj/Awesome-Face-Restoration/issues/new">Report Bug</a> </p> </p>This repository provides a summary of deep learning-based face restoration algorithms. Our classification is based on the review paper "A Survey of Deep Face Restoration: Denoise, Super-Resolution, Deblur, Artifact Removal".
<img src="./figures/full.gif" height=224 width=224 alt="Blind face restoration"> | <img src="./figures/LR.gif" width=224 height=224 alt="Face super-resolution"> | <img src="./figures/blur.gif" height=224 width=224 alt="Face deblurring"> | <img src="./figures/noise.gif" height=224 width=224 alt="Face denoising"> | <img src="./figures/jpeg.gif" height=224 width=224 alt="Face artifact removal"> |
---|---|---|---|---|
Blind Face Restoration | Face Super-Resolution | Face Deblurring</font> | Face Denoising | Face Artifact Removal |
Survey paper
T. Wang, k. Zhang, X. Chen, W. Luo, J. Deng, T. Lu, X. Cao, W. Liu, H. Li, and S. Zafeiriou “A Survey of Deep Face Restoration: Denoise, Super-Resolution, Deblur, Artifact Removal,” arXiv preprint arXiv:2211.02831, 2022. [pdf]
@article{wang2022survey
title={A Survey of Deep Face Restoration: Denoise, Super-Resolution, Deblur, Artifact Removal},
author={Wang, Tao, and Zhang, kaihao, and Chen, Xuanxi and Luo, Wenhan and Deng, Jiankang and Lu, Tong and Cao, Xiaochun and Liu, Wei and Li, Hongdong and Zafeiriou, Stefanos},
journal={arXiv preprint arXiv:2211.02831},
year={2022}
}
See our paper for more details. If you have any suggestions, feel free to contact me (e-mail: taowangzj@gmail.com). Thanks.
Table of contents
<!-- - [Survey paper](#survey-paper) - [Table of contents](#table-of-contents) -->- Surveys
- Deep Blind Face Restoration
- Deep Face Super-Resolution
- Deep Face Deblurring
- Deep Face Denoising
- Deep Face Artifact Removal
- Other Related Works
- Image Quality Assessment
- Benchmark Datasets
Surveys
<!-- |year|Pub|Title|Link| |:----:|:----:|:----:|:----:| <!-- |2020|TPAMI|Single image deraining: From model-based to data-driven and beyond|\[[paper]()\]| |2020|NN|Deep learning on image denoising: An overview|\[[paper]()\]| |2020|TPAMI|Deep learning for image superresolution: A survey|\[[paper]()\]| |2022|IJCV|Deep image deblurring: A survey|\[[paper]()\]| |2022|Neuro-<br>computing|A survey of deep learning approaches to image restoration|\[[paper]()\]| -->year | Pub | Title | Link |
---|---|---|---|
2022 | Arxiv | A Survey of Deep Face Restoration: Denoise, Super-Resolution, Deblur, Artifact Removal | [paper] |
Deep Blind Face Restoration
Year<div style="width:20px"> | Pub<div style="width:60px"> | Title<div style="width:600px"> | Links<div style="width:100px"> | Arch<div style="width:45px"> |
---|---|---|---|---|
2018 | ECCV | Learning Warped Guidance for Blind Face Restoration | [paper][code] | CNN |
2018 | CVPR | FSRNet: End-to-End Learning Face Super-Resolution with Facial Priors | [paper][code] | CNN |
2019 | IJCV | Identity-preserving Face Recovery from Stylized Portraits | [paper][code] | GAN |
2020 | ECCV | Blind Face Restoration via Deep Multi-scale Component Dictionaries | [paper][code] | CNN |
2020 | CVPR | Enhanced Blind Face Restoration with Multi-Exemplar Images <br>and Adaptive Spatial Feature Fusion | [paper][code] | CNN |
2020 | MM | HiFaceGAN: Face Renovation via Collaborative Suppression and Replenishment | [paper][code] | GAN |
2020 | CVPR | Image Processing Using Multi-Code GAN Prior | [paper][code] | GAN |
2021 | TPAMI | Face Restoration via Plug-and-Play 3D Facial Priors | [paper][code] | CNN |
2021 | CVPR | Progressive Semantic-Aware Style Transformation for Blind Face Restoration | [paper][code] | GAN |
2021 | CVPR | Towards Real-World Blind Face Restoration with Generative Facial Prior | [paper][code] | GAN |
2021 | CVPR | GAN Prior Embedded Network for Blind Face Restoration in the Wild | [paper][code] | GAN |
2022 | Arxiv | Multi-prior learning via neural architecture search for blind face restoration | [paper][code] | CNN |
2022 | CVPR | Blind Face Restoration via Integrating Face Shape and Generative Priors | [paper] | GAN |
2022 | CVPR | RestoreFormer: High-Quality Blind Face Restoration <br> from Undegraded Key-Value Pairs | [paper][code] | ViT |
2022 | NeurIPS | Towards Robust Blind Face Restoration with Codebook Lookup Transformer | [paper][code] | ViT |
2022 | AAAI | Panini-Net: GAN Prior Based Degradation-Aware Feature Interpolation for Face Restoration | [paper][code] | GAN |
2022 | Arxiv | FaceFormer: Scale-aware Blind Face Restoration with Transformers | [paper][code] | ViT |
2022 | Arxiv | Blind Face Restoration: Benchmark Datasets and a Baseline Model | [paper][code] | ViT |
2022 | ECCV | VQFR: Blind Face Restoration with Vector-Quantized Dictionary and Parallel Decoder | [paper][code] | CNN |
2022 | TPAMI | Learning Dual Memory Dictionaries for Blind Face Restoration | [paper][code] | CNN |
2022 | Arxiv | Difface: Blind Face Restoration with Diffused Error Contraction | [paper][code] | Diffusion |
2023 | CVPR | DR2: Diffusion-based Robust Degradation Remover for Blind Face Restoration | [paper][code] | Diffusion |
2023 | CVPR | TFRGAN: Leveraging Text Information for Blind Face Restoration with Extreme Degradation | [paper][code] | GAN |
2023 | TCSVT | DEAR-GAN: Degradation-Aware Face Restoration With GAN Prior | [paper][code] | GAN |
2023 | FSP | Degradation Learning and Skip-Transformer for Blind Face Restoration | [paper][code] | GAN |
2023 | WACV | AT-DDPM: Restoring Faces degraded by Atmospheric Turbulence using Denoising Diffusion Probabilistic Models | [paper][code] | Diffusion |
2023 | ACMMM | DiffBFR: Bootstrapping Diffusion Model for Blind Face Restoration | [paper][code] | Diffusion |
2024 | TCSVT | Towards Real-World Blind Face Restoration with Generative Diffusion Prior | [paper][code] | Diffusion |
2024 | WACV | Diffuse and Restore: A Region-Adaptive Diffusion Model for Identity-Preserving Blind Face Restoration | [paper][code] | Diffusion |
2024 | AAAI | Blind Face Restoration under Extreme Conditions: Leveraging 3D-2D Prior Fusion for Superior Structural and Texture Recovery | [paper][code] | Diffusion |
Deep Face Super-Resolution
Year<div style="width:20px"> | Pub<div style="width:60px"> | Title<div style="width:600px"> | Links<div style="width:100px"> | Arch<div style="width:45px"> |
---|---|---|---|---|
2015 | AAAI | Learning Face Hallucination in the Wild | [paper][code] | CNN |
2016 | ECCV | Deep Cascaded Bi-Network for Face Hallucination | [paper][code] | CNN |
2016 | ECCV | Ultra-Resolving Face Images by Discriminative Generative Networks | [paper][code] | GAN |
2017 | CVPR | Attention-Aware Face Hallucination via Deep Reinforcement Learning | [paper][code] | CNN |
2017 | CVPR | Hallucinating Very Low-Resolution Unaligned and Noisy Face Images by <br>Transformative Discriminative Autoencoders | [paper][code] | CNN |
2017 | ICCV | Learning to Super-Resolve Blurry Face and Text Images | [paper][code] | GAN |
2017 | AAAI | Face Hallucination with Tiny Unaligned Images <br> by Transformative Discriminative Neural Networks | [paper][code] | GAN |
2018 | ECCV | Face Super-resolution Guided by Facial Component Heatmaps | [paper][code] | CNN |
2018 | CVPR | Super-FAN: Integrated facial landmark localization and <br> super-resolution of real-world low resolution faces in arbitrary poses with gans | [paper][code] | GAN |
2018 | ECCV | To learn image super-resolution,<br> use a GAN to learn how to do image degradation first | [paper][code] | GAN |
2019 | CVPRW | Exemplar Guided Face Image Super-Resolution without Facial Landmarks | [paper][code] | CNN |
2019 | BMVC | Progressive Face Super-Resolution via Attention to Facial Landmark | [paper][code] | CNN |
2020 | WACV | Component Attention Guided Face Super-Resolution Network: CAGFace | [paper][code] | CNN |
2020 | TNNLS | Dual-Path Deep Fusion Network for Face Image Hallucination | [paper][code] | CNN |
2019 | NEUCOM | On potentials of regularized Wasserstein generative adversarial networks for<br> realistic hallucination of tiny faces | [paper][code] | GAN |
2020 | CVPR | PULSE: Self-Supervised Photo Upsampling via <br>Latent Space Exploration of Generative Models | [paper][code] | GAN |
2021 | TBBIS | E-ComSupResNet: Enhanced Face Super-Resolution Through Compact Network | [paper][code] | CNN |
2021 | MM | Face Hallucination via Split-Attention in Split-Attention Network | [paper][code] | CNN |
2021 | ICIP | Progressive Face Super-Resolution with Non-Parametric Facial Prior Enhancement | [paper][code] | GAN |
2022 | CVPR | GCFSR: a Generative and Controllable Face Super Resolution Method <br> Without Facial and GAN Priors | [paper][code] | GAN |
2022 | TPAMI | EDFace-Celeb-1 M: Benchmarking Face Hallucination with a Million-scale Dataset | [paper][code] | CNN |
2023 | TIP | Semi-Cycled Generative Adversarial Networks for Real-World Face Super-Resolution | [paper][code] | GAN |
2023 | AAAI | GAN Prior based Null-Space Learning for Consistent Super-Resolution | [paper][code] | GAN |
2023 | CVPR | Spatial-Frequency Mutual Learning for Face Super-Resolution | [paper][code] | CNN |
2023 | TMM | Sctanet: A Spatial Attention-guided Cnn-transformer Aggregation Network for Deep Face Image Super-Resolution | [paper][code] | ViT |
2023 | TMM | An Efficient Latent Style Guided Transformer-CNN Framework for Face Super-Resolution | [paper][code] | ViT |
2023 | TMM | Exploiting Multi-scale Parallel Self-attention and Local Variation via Dual-branch Transformer-CNN Structure for Face Super-resolution | [paper][code] | ViT |
2023 | PR | A Composite Network Model for Face Super-Resolution with Multi-Order Head Attention Facial Priors | [paper][code] | ViT |
2023 | TIP | Semi-Cycled Generative Adversarial Networks for Real-World Face Super-Resolution | [paper][code] | GAN |
2023 | AS | A Multi-Scale Deep Back-Projection Backbone for Face Super-Resolution with Diffusion Models | [paper][code] | Diffusion |
2023 | TIM | Deep HyFeat Based Attention in Attention Model for Face Super-Resolution | [paper][code] | CNN |
2023 | PRL | Attentive ExFeat based Deep Generative Adversarial Network for Noise Robust Face Super-resolution | [paper][code] | GAN |
2023 | NN | Self-attention Learning Network for Face Super-resolution | [paper][code] | CNN |
Deep Face Deblurring
Year<div style="width:20px"> | Pub<div style="width:60px"> | Title<div style="width:600px"> | Links<div style="width:100px"> | Arch<div style="width:45px"> |
---|---|---|---|---|
2017 | Arxiv | DeepDeblur: Fast one-step blurry face images restoration | [paper][code] | CNN |
2018 | CVPR | Deep Semantic Face Deblurring | [paper][code] | CNN |
2020 | IJCV | Exploiting Semantics for Face Image Deblurring | [paper][code] | CNN |
2020 | TIP | Deblurring Face Images using Uncertainty Guided Multi-Stream Semantic Networks | [paper][code] | CNN |
2020 | MM | HiFaceGAN: Face Renovation via Collaborative Suppression and Replenishment | [paper][code] | GAN |
2020 | AAAI | Learning to deblur face images via sketch synthesis | [paper][code] | CNN |
2022 | TOG | Face Deblurring using Dual Camera Fusion on Mobile Phones | [paper][code] | CNN |
2022 | WACV | Deep Feature Prior Guided Face Deblurring | [paper][code] | CNN |
2022 | Arxiv | Multi-prior learning via neural architecture search for blind face restoration | [paper][code] | CNN |
2022 | Arxiv | Blind Face Restoration: Benchmark Datasets and a Baseline Model | [paper][code] | ViT |
2022 | SIGGRAPH | Face Deblurring using Dual Camera Fusion on Mobile Phones | [paper][code] | CNN |
Deep Face Denoising
Year<div style="width:20px"> | Pub<div style="width:60px"> | Title<div style="width:600px"> | Links<div style="width:100px"> | Arch<div style="width:45px"> |
---|---|---|---|---|
2020 | MM | HiFaceGAN: Face Renovation via Collaborative Suppression and Replenishment | [paper][code] | GAN |
2022 | Arxiv | Multi-prior learning via neural architecture search for blind face restoration | [paper][code] | CNN |
2022 | Arxiv | Blind Face Restoration: Benchmark Datasets and a Baseline Model | [paper][code] | ViT |
Deep Face Artifact Removal
Year<div style="width:20px"> | Pub<div style="width:60px"> | Title<div style="width:600px"> | Links<div style="width:100px"> | Arch<div style="width:45px"> |
---|---|---|---|---|
2020 | MM | HiFaceGAN: Face Renovation via Collaborative Suppression and Replenishment | [paper][code] | GAN |
2022 | Arxiv | Blind Face Restoration: Benchmark Datasets and a Baseline Model | [paper][code] | ViT |
2022 | Arxiv | Multi-prior learning via neural architecture search for blind face restoration | [paper][code] | CNN |
Other Related Works
- CG-GAN: Class-Attribute Guided Generative Adversarial Network for Old Photo Restoration, Liu et al., ACMM 2021. Link
- Old Photo Restoration via Deep Latent Space Translation, Wan et al., TPAMI 2022. Link
- Demeshnet: Blind Face Inpainting for Deep Meshface Verification, Zhang et al. TIP 2017. Link
- Face Inpainting based on High-level Facial Attributes, Jampour et al., CVIU 2017. Link
- ABPN: Adaptive Blend Pyramid Network for Real-Time Local Retouching of Ultra High-Resolution Photo, Lei et al., CVPR 2022. Link
- Autoretouch: Automatic Professional Face Retouching, Shafaei et al., WACV 2021. Link
- Network Architecture Search for Face Enhancement, Yasarla et al., Arxiv 2021. Link
- Towards Automatic Face-to-face Translation, KR, R et al., ACMMM 2019. Link
- A Sketchtransformer Network for Face Photo-sketch Synthesis, Zhu et al., IJCAI 2021. Link
Image Quality Assessment
Method | Type | Code/Ref |
---|---|---|
PSNR (Peak Signal-to-Noise Ratio) | Full-Reference | Code |
SSIM (Structural Similarity Index Measurement) | Full-Reference | Code |
MS-SSIM (Multi-scale Structural Similarity Index Measurement) | Full-Reference | Code |
LPIPS (Learned Perceptual Image Patch Similarity) | Full-Reference | Code |
NIQE (Naturalness Image Quality Evaluator) | Non-Reference | Code |
FID (Fréchet Inception Distance) | Non-Reference | Code |
PI (Perceptual Index) | Non-Reference | Code |
MOS (Mean Opinion Score) | Subject-Metric | Ref |
iPrecision | Task Driven-Metric | Ref |
iRecall | Task Driven-Metric | Ref |
LLE (Landmark Localization Error) | Task Driven-Metric | Code |
Deg (Identity Distance) | Task Driven-Metric | Code |
AFLD (Average Face Landmark Distance) | Task Driven-Metric | Code |
AFICS (Average Face ID Cosine Similarity) | Task Driven-Metric | Code |
Benchmark Datasets
Recommended Datasets
Dataset | Usage | Quantity | Type |
---|---|---|---|
EDFace-Celeb | training&testing | >1M | paired dataset |
PFHQ | training | 60,000 | paired dataset |
FFHQ | training | 70,000 | non-paired dataset |
CelebChild-Test | testing | 180 | non-paired real-world dataset |
WebPhoto-Test | testing | 407 | non-paired real-world dataset |
LFW-Test | testing | 1,711 | non-paired real-world dataset |