face restoration
GFPGAN is a practical blind face-restoration model that leverages pretrained StyleGAN2 priors to restore realistic, high-quality faces from low-quality images.
Solves the same problem with a different architecture or at a different layer. Expect to rewrite the integration.
choose when restoring whole images/video rather than only faces; no face-specific identity prior
Real-ESRGAN aims at developing Practical Algorithms for General Image/Video Restoration.
choose for face restoration with codebook transformer prior; handles very degraded faces and offers fidelity control
[NeurIPS 2022] Towards Robust Blind Face Restoration with Codebook Lookup Transformer
These projects were analysed and named GFPGAN among their alternatives. The relationship is not symmetric — how GFPGAN rates them is a separate judgement, made when GFPGAN is analysed in its own right.
calls GFPGAN “Specialized real-world face restoration; choose for old photos where faces are the main subject.”
Real-ESRGAN aims at developing Practical Algorithms for General Image/Video Restoration.
calls GFPGAN “Older GAN-based face restorer; choose it for legacy pipeline compatibility and smooth background blending, but it is unmaintained.”
[NeurIPS 2022] Towards Robust Blind Face Restoration with Codebook Lookup Transformer
calls GFPGAN “Choose for practical real-world face restoration with a widely-used PyTorch model instead of PaddleGAN's suite.”
PaddlePaddle GAN library, including lots of interesting applications like First-Order motion transfer, Wav2Lip, picture repair, image editing, photo2cartoon, image style transfer, GPEN, and so on.
calls GFPGAN “Choose for practical face restoration only, when you want a specialized model instead of a general-purpose toolbox.”
OpenMMLab Multimodal Advanced, Generative, and Intelligent Creation Toolbox. Unlock the magic 🪄: Generative-AI (AIGC), easy-to-use APIs, awsome model zoo, diffusion models, for text-to-image generation, image/video restoration/enhancement, etc.