Artem Ivanov comments

Results 8 comments of


                                            Artem Ivanov

deepfakes/faceswap state-of-the-art

@Kirin-kun @agilebean I want to note that below is no official FaceSwap view on the subject but my subjective advices based on personal experience with FS and its forks. 1)...

deepfakes/faceswap state-of-the-art

@kellurian 1. I think whatever works for you is good. I wouldn't use too many pics for `set B` and I probably wouldn't favor loss values over preview because loss...

deepfakes/faceswap state-of-the-art

@kellurian Never heard of "red screen of death" and I am pretty sure I trained the model for days straight with 980Ti. Basically your `ENCODER_DIM` is quite low though I...

deepfakes/faceswap state-of-the-art

@kellurian `ENCODER_DIM` is the size of the encoder's Deep interconnected ANN. The bigger the number the more clarity you get. You must find a balance balance between the output size...

deepfakes/faceswap state-of-the-art

@kellurian Noted, thanks. Shall fix in the following PR.

deepfakes/faceswap state-of-the-art

@agilebean Its a rather long summary, perhaps you should compose some kind of WIKI page. > Sharp Just leave the option undefined >How would you increase the image shape in...

deepfakes/faceswap state-of-the-art

@gessyoo that's pretty much what I said about batch sizes. Well almost. If you know just a little python I'd tweak the model a bit to better utilize your memory....

deepfakes/faceswap state-of-the-art

@kellurian @gessyoo To avoid Windows hogging all the RAM you could switch to your integrated video card and do FS at your dedicated one. If that's not an option close...