oawiles / X2Face

Pytorch code for ECCV 2018 paper

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

How to training?

cscyangj opened this issue · comments

I got following result.

image

It seems that i need to training for new driving image? so how to do that?

thanks

Hi. I'd say that the problem is probably that your crops are too different. First try recropping the images so that the face is in approximately the same place as the sample ones. This has a pretty big effect on the results. If you want to use your own crops you'd have to have access to a pretty big dataset which may or may not be feasible...

However, if the re-cropping doesn't improve things, let me know.

@oawiles thanks, I will try and feedback to you.

@oawiles looks much better now.

image

@oawiles I got bad result here, could you give me some advise?

Hm I'm not sure: it seems to me that the person isn't changing too much (or at all?). Hence why you're getting the same output?