search

zyainfal / One-Shot-Face-Swapping-on-Megapixels

One Shot Face Swapping on Megapixels.
Other
317 stars 41 forks source link

Unable to produce good results on custom images #43

Open snaeemm opened 2 years ago

snaeemm commented 2 years ago

Can someone explain why the custom images do not give correct outputs even after labelling and everything done appropriately?

I am getting the exact same results when I create my custom mask using the parse.pytorch model, and the test.py function as given in another issue reported here. These are the outputs which are identical to using the masked images from your model on the images in CelebAHQ dataset.

injectionexperiment

But when I try my custom image, the model falls apart. It seems like the model is inputting another image all of a sudden.

Yes I cropped the image to 1024*1024 before proceeding, were there any other transformations required?

injectionexperiment1

xuan240 commented 2 years ago

@snaeemm I have encountered the same problem before, and the effect of using custom images for exchange is very poor

snaeemm commented 2 years ago

Your issue should be resolved once you used images that are cropped with very little neck and shoulders and mostly the face.

On Mon, 12 Sept 2022, 06:35 xuan240, @.***> wrote:

@snaeemm https://github.com/snaeemm I have encountered the same problem before, and the effect of using custom images for exchange is very poor

— Reply to this email directly, view it on GitHub https://github.com/zyainfal/One-Shot-Face-Swapping-on-Megapixels/issues/43#issuecomment-1243111571, or unsubscribe https://github.com/notifications/unsubscribe-auth/ATFZZQNL3D3E34W2KYN2ZEDV52CHZANCNFSM56ZNFYJA . You are receiving this because you were mentioned.Message ID: @.*** .com>