search

shaoanlu / faceswap-GAN

A denoising autoencoder + adversarial losses and attention mechanisms for face swapping.
3.37k stars 843 forks source link

Video Conversion Load a still image #79

Open kazshimada opened 6 years ago

kazshimada commented 6 years ago

Hello. I appreciate the excellent code. It is a question about FaceSwap_GAN_v2.1_train.ipynb（https://github.com/shaoanlu/faceswap-GAN/blob/master/FaceSwap_GAN_v2.1_train.ipynb) I want to load a still image rather than videos using the logic “Video conversion functions” to recognize a face the image, to correct face, and to change face. How can I do it?

hengshan123 commented 6 years ago

How many time do you train the model using v2.1_train? I train it 1000 times and the result is like

trump
kazshimada commented 6 years ago

Actually I trained using v2 model. sorry. The result is serious !! I think the basic structure of v2 and v2.1 is the same.I usually run the training 15000~30000 iter.

shaoanlu commented 6 years ago

Not checked yet but v2 step 11 should be compatible with v2.1.