This article deals with how to restore old photos using neural networks. Applied approaches: In-Place BatchNorm, Partial Convolution, Self-Attention GAN. The whole process of restoration is described in detail:

​

1. Find all the image defects: fractures, scuffs, holes.
2. Inpaint the discovered defects, based on the pixel values around them.
3. Colorize the image.

​

Also, the website was created where any photo could be uploaded for restoration online.

​

https://habr.com/en/company/mailru/blog/459696/