The verification results are poor, can you give some advice? Thank you

Boese0601 / MagicDance

[ICML 2024] MagicPose(also known as MagicDance): Realistic Human Poses and Facial Expressions Retargeting with Identity-aware Diffusion

https://boese0601.github.io/magicdance/

Other

629 stars 52 forks source link

The verification results are poor, can you give some advice? Thank you #24

Open goalinshi opened 2 months ago

goalinshi commented 2 months ago

Your work is great, I'm interested in your work, but I tested my own video image effect is very poor, and the coherence after generating the image synthesis video is also poor, the model loading (strict=True) is normal, I'm not sure if it's me doing it wrong, or if the model itself has limitations, please advise

Boese0601 commented 2 months ago

Could you please post your results here? Thanks. It would be quite difficult to tell where the error is without visualizations.

goalinshi commented 2 months ago

The coherence of the generated video is poor, and the jitter is serious, but it cannot be well displayed to you; test1 test2 test3

songzr0 commented 1 month ago

I have the same question. FaceID in Generated images does not match the person in the original image