Gigapixel - AI: A new facial reconstruction option

Someone in the community talked about using AI technology for facial reconstruction, and I think this technology is currently fully available and can be used in some extreme cases for face reconstruction, while the rest of the picture still uses the current scheme, but of course this technology is not necessarily suitable for all pictures, so it should have a switch. There are various open source solutions on GitHub, where GAN type solutions seem to be more feasible.

1 Like

It’s good. At the moment I use the Remini. When the Topaz Labs makes just a GAN software I will buy it.

1 Like

Agree. I ve tried GFPGAN in conda environment on Windows PC. Reuslts are quite impressive. But not perfect yet. For ex. totally blurry faces are reconstructed randomly without logic, like ghosts faces. Except that, this open source solution gives prety decent results.Would like if Topaz could implement similiar technique in Gpx AI. Like successfull Chronos model in Video Enhance AI that surpassed open source solution called RIFE in terms of interpolation. I think that Topaz is capable doing that. And maybe to focus& develop another individual product based on this themathics why not. To not be integrated in Gpx AI. such a piece of software would probably be extremely popular and attract a lot of users. Who knows :wink:

2 Likes

I agree.
Many open source AI produce very good face reconstruct result especially when the face is looking straight forward. Not perfect yet, it may fail to reconstruct the face if looking sideward.

Here are some examples I tried,


Left: Original low resolution image
Right: After 4x upscale with AI face reconstruction


Left: Original low resolution image
Right: After 4x upscale with AI face reconstruction


Left: Original low resolution image
Right: After 4x upscale with AI face reconstruction


Left: Original low resolution image
Right: After 4x upscale with AI face reconstruction

Original Face Image:
test_05_s test_06_s test_03_s test_08_s

1 Like

Which program did you test? I would like to use it on an M1 Mac. On Linux I used programs from source but I don’t want to do it on Mac.
Do you have an idea?

The images above were tested with GFPGAN by TencentARC (Applied Research Center).
You can find the source code in Github.

1 Like

Thank you!
At the moment I use the Remini but I think not that the best in the AI face reconstruction.

It looks incredible! the whole head was reconstructed including the hair, the end result is a very logical restoration of pixel decay