AvLab-CV/Dual_Generator_Face_Reenactment

Quality is bad

hcl14 opened this issue · 0 comments

hcl14 commented

Hi! Demo skips frames with fps=5, so I inferenced offline on my video and quality seems very bad. I used your get_arcface function for alignment of both driving video and source frames.

Here is my inference scripts and resulting video: https://drive.google.com/drive/folders/1Ow_1_eYH3yK_VEqQotHFN6ZO30_olXI5?usp=sharing

Maybe you used some another face alignment/landmark detector for training? Or am I doing something wrong?

PS. You haven't provided one of the weights and I used the one from the web, assuming it's the same: #5