wang-zidu / 3DDFA-V3

The official implementation of 3DDFA_V3 in CVPR2024 (Highlight).
MIT License
218 stars 14 forks source link

i test on video the result is a bit jittor, any method to optim the pose? #9

Open mmxuan18 opened 3 months ago

wang-zidu commented 3 months ago

Thank you for your feedback. We also have some video results for your reference. There might indeed be some jitter issues due to the limited video data in the training set.

There are two possible solutions to handle the jitter (may be more options). One straightforward way is to smooth it by recording the parameters of the previous few frames. Another approach is to enhance the model by adding more video data for further training. I believe the first solution is the simplest and most effective. I will note this issue, and if I have time, I will address this request as soon as possible (trying to resolve it through these two methods) and let you know.

jitter

emlcpfx commented 1 month ago

@mmxuan18 Can you please post the code you used to test on video?

EDIT: nevermind, I just ran it on an image sequence.