search

AliaksandrSiarohin / monkey-net

Animating Arbitrary Objects via Deep Motion Transfer
467 stars 81 forks source link

align source image with the first frame of driver video #23

Closed maliho0803 closed 4 years ago

maliho0803 commented 4 years ago

do you have any good strategy for aligning source image with the first frame of driver video

AliaksandrSiarohin commented 4 years ago

I don't know any good general stratefy. Here there is some strategy for faces https://github.com/AliaksandrSiarohin/first-order-model/blob/master/demo.py , e.g --find_best_frame

maliho0803 commented 4 years ago

thank you very much for your reply! one more question, if i predict more keypoints , will it improve the quality of the generated images?

AliaksandrSiarohin commented 4 years ago

It depends, if you predict more it improves up to some point. If you predict too much they can start to encode appearance.

maliho0803 commented 4 years ago

thank you! maybe 10 keypoints in your code for human face is the best choice