人脸幻视,也就是实现对低质模糊图像的重构。paper and code.
note: utilizes the facial prior by alternately optimizing FH and dense correspondence field estimation
code: https://github.com/zhusz/ECCV16-CBN [matlab]
2. Super-fan: Integrated facial landmark localization and super-resolution of real-world low resolution faces in arbitrary poses with gans
note: guarantee the consistency of facial landmarks by end-to-end multi-task learning
code: https://github.com/jzijin/Super-FAN
note: uses not only facial landmark heatmaps but also face parsing maps as prior constraints.
code: https://github.com/cs-giung/FSRNet-pytorch
note: recovering the real iden- tity, adopts a super-identity loss function and a domain- integrated training approach to stable the joint training
code: https://github.com/SirLPS/Face-Hallucination
note: decompose deblocked face images into facial components and background, use the component landmarks to retrieve adequate HR exemplars in external datasets, perform generic SR on the background, and finally fuse them to complete HR faces.
code: https://github.com/yangchihyuan/HallucinatingCompressedFaceImages [matlab]
code: https://github.com/hhb072/WaveletSRNet/
note: utilize additional facial attribute information to perform FH with the specified attributes, based on the conditional GAN
code: https://github.com/steven413d/AACNN [tensorflow]
8. SiGAN: Siamese Generative Adversarial Network for Identity-Preserving Face Hallucination, Chih-Chung Hsu et al., TIP 2019.
code: https://github.com/jesse1029/SiGAN
code: https://github.com/Flaick/LR-License-Plate-Generation
code: https://github.com/steven413d/AACNN