Faceswap Forum

I am going to train using the Villain model
As you know, Villain model is 128px input, 128px output
Then, when extracting the face from the video, is there a difference between 128px and 256px, 512px?

Just leave it at 256 for Villain. You wont get any difference in results by changing that.

bryanlyon wrote: ↑Mon Jun 15, 2020 4:42 pm

Just leave it at 256 for Villain. You wont get any difference in results by changing that.

So 128Px Training speed is faster than 256Px ?

It doesn't make any difference. The model is still fed with 128px images.

In theory 256px images will be loaded from disk slower, but the bottleneck is the GPU, so in real terms there is no difference in speed.

A larger image size is also required, than model input, for coverage and padding reasons.

Just leave it at 256.