I’m currently using Face Recovery Gen 2 on 1,300 images at 400x400 resolution, and my system is barely being utilized, the GPU is not being used to its full potential. It’s barely breaking a sweat and is at 47°c.
When exporting these images, the software should be able to process in parallel, restoring faces in the background so that saving files can be instant. In my opinion, batch processing could benefit from significant improvements. In the multi-threading department.
The VRAM should be loaded and the GPU should detect the faces, and pre-render in the background. I barely break the 1.5 Go of my 24 Go VRAM.
Each image is worked one after the other. We have multiple cores on the CPU side, and thousands of cores on the GPU.
Photo AI should use everything.
In my actual case, I’m working on a 40 secondes sequence. It’s gonna take a while.
It’s a bit of a shame because the results are excellent for what we do, recovering low resolution faces, where Video AI can’t do much.
