search

OpenTalker / DPE

[CVPR 2023] DPE: Disentanglement of Pose and Expression for General Video Portrait Editing
https://carlyx.github.io/DPE/
MIT License
417 stars 45 forks source link

paste pose #10

Closed vianseto closed 12 months ago

vianseto commented 1 year ago

hi, I really appreciate your work... it's very interesting. however, when I tried the demo_paste and examined its contents, I found that the "--face" argument was set to the default value of "exp". since I wanted to try transferring the pose from a driving video, I ended up changing it to "pose". however, the masking result wasn't accurate. can you help me with this?

on run_demo_paste line 327 I change this output_dict = self.gen(img_source, img_target, 'exp')

into this output_dict = self.gen(img_source, img_target, 'pose')

and this is the result vlcsnap-2023-06-14-11h41m18s336

Carlyx commented 1 year ago

Thanks for your attention. When the pose is transformed, the new head does not match well with the original one, so it cannot be pasted back by a simple Gaussian blur, especially when the neck is misaligned. In fact, DPE only performs pasting back operation for facial expression transfer.