search

TMElyralab / MuseTalk

MuseTalk: Real-Time High Quality Lip Synchorization with Latent Space Inpainting
Other
1.94k stars 239 forks source link

The fidelity of the generated faces is poor #18

Open xiao-keeplearning opened 3 months ago

xiao-keeplearning commented 3 months ago

Thank your great work! I conducted some experiments and found that the fidelity of the generated faces is poor, generated person does not resemble the original video closely. Could you give me some suggestions?

zhangjinyan888 commented 3 months ago

确实如此，感觉下边换了一个人的下巴一样

itechmusic commented 3 months ago

Thank your great work! I conducted some experiments and found that the fidelity of the generated faces is poor, generated person does not resemble the original video closely. Could you give me some suggestions?

We are working on this problem. Maybe a GAN loss would help.