search

guoqincode / Open-AnimateAnyone

Unofficial Implementation of Animate Anyone
2.85k stars 230 forks source link

I use train_hack, why stage2 background is very bad? #95

Closed VisionU closed 6 months ago

VisionU commented 6 months ago

I train it use UBC fashion dataset， stage1 test result is right, but background of stage2 result is very abnormal, could you give some advise, thanks !

Stage1:

3d133f09-383e-4eab-80c8-01aed5d45000 output

Stage2:

https://github.com/guoqincode/Open-AnimateAnyone/assets/9945763/395ccf52-f8c8-4e44-8053-b5e7acb56d45

@zhenzhiwang @guoqincode @ailingzengzzz

guoqincode commented 6 months ago
  1. You can replace poseguider's 2d conv with resnet.InflatedConv3d
  2. Notifying so many people makes no sense
VisionU commented 6 months ago
  1. You can replace poseguider's 2d conv with resnet.InflatedConv3d
  2. Notifying so many people makes no sense

Thanks for your help. OK, I understand~

kings-rgb commented 6 months ago
  1. You can replace poseguider's 2d conv with resnet.InflatedConv3d
  2. Notifying so many people makes no sense

Thanks for your help. OK, I understand~

Do you solve this question?

VisionU commented 5 months ago
  1. You can replace poseguider's 2d conv with resnet.InflatedConv3d
  2. Notifying so many people makes no sense

Thanks for your help. OK, I understand~

Do you solve this question?

NO, use zero terminal snr can fix it

newcherryberry commented 1 month ago

@VisionU Hi, is there any way I can possibly contact you about your progress? Would really appreciate it as a CS student