search

mcahny / Deep-Video-Inpainting

Official pytorch implementation for "Deep Video Inpainting" (CVPR 2019)
508 stars 95 forks source link

Can you tell me how to train this model? THANKS!!! #10

Closed ytongW closed 4 years ago

gonghuiyun commented 4 years ago

我也在写这个训练模型，能留一个联系方式吗

decajcd commented 4 years ago

我也在写这个训练模型，能留一个联系方式吗

When can we get train code?thanks. @mcahny @gonghuiyun

aryan461 commented 4 years ago

I tried to write the training code. But I need help with L_F and L_W losses. Could you please tell me how to define them using Flownet2? Or can you give us your losses code?

decajcd commented 4 years ago

I tried to write the training code. But I need help with L_F and L_W losses. Could you please tell me how to define them using Flownet2? Or can you give us your losses code?

Have you trained with reconstruction loss?

aryan461 commented 4 years ago

Have you trained with reconstruction loss?

I didn't consider it. Did you do this? Does it work? so what about optical flow?

mcahny commented 4 years ago

Please check the "train_git.py " which can provide some reference algorithm, not yet runnable though. Thanks.

decajcd commented 4 years ago

Please check the "train_git.py " which can provide some reference algorithm, not yet runnable though. Thanks. Can you release loss code?

mcahny commented 4 years ago

All loss functions (RECON_loss, HOLE_loss, ST_loss, LT_loss, flow_loss) are based on nn.L1Loss() , where RECON_loss includes extra ssim loss where I borrowed from https://github.com/Po-Hsun-Su/pytorch-ssim/tree/master/pytorch_ssim