Tech Talk - Restoring and Upscaling Damaged Photos with GFP-GAN
In this Tech Talk, we covered the novel photo restoration neural network GFP-GAN. This astounding architecture, developed just last year, is capable of restoring and upscaling images containing faces to a high level. We will start by discussing how GFP-GAN works under the hood, before moving into Gradient Notebooks to examining how this works in practice with PyTorch. Once we’ve demonstrated the efficacy of the GAN and shown how it can be run in Notebooks, we will close with a tutorial for creating a Flask app for the GAN and deploying it on Gradient with Gradient Deployments.
Link to slides: https://docs.google.com/presentation/...
Useful links:
https://blog.paperspace.com/restoring...
https://blog.paperspace.com/deploymen...
https://github.com/TencentARC/GFPGAN
https://github.com/gradient-ai/gfpgan - Our fork
https://arxiv.org/abs/2101.04061 - GFP-GAN
https://arxiv.org/abs/2107.10833 - Real-ESRGAN