Min Max Iterations

[gwyneth4320]

Is there a minimum iteration i should achieve before stopping training? Would the video be better (accurate) with higher iterations (say 100k)?

[abigflea]

There is always much discussion about this.
If you are doing "medium" quality, I personally figure 100k as a time to test convert and see how its doing.
Recently I've been hitting 300 - 600K before I feel its "finished" with any arbitrary other settings..

Original is usually done sooner because its only 64x64 and a strain forward model architecture, so 100K would be nearly done
villian is 128x128 and Im not happy with it till passed 500K (1000K?)

The main rule is until it "gets done" by "looking good".

If you want a vague idea, watch your loss graph and wait till it has gotten really flat, refer to the training guide.

[gwyneth4320]

Thanks for the reply.

Apologies, im not a tech savvy. I heard gpu runs better than cpu when doing face swapping especiialy using 'villain'. When i opened up task bar the cpu was mainly used while only 1% gpu was used. Is there a way i can change it to GPU (if this makes it better/faster)?

Thank you in advance

[abigflea]

You must be using windows.
Windows doesn't always report usage well.

If you're getting iteration at least every couple seconds it's probably running on GPU.

On CPU it is practically glacial in training speed. Is that is whats happening?