search

Tengfei-Wang / HFGI

CVPR 2022 HFGI: High-Fidelity GAN Inversion for Image Attribute Editing
https://tengfei-wang.github.io/HFGI/
457 stars 50 forks source link

Problem about the results of pose editing #2

Closed huiqu18 closed 2 years ago

huiqu18 commented 2 years ago

Thank you for the great work! I have tried the inference code with the pretrained checkpoint for pose editing, but there are obvious artifacts in the edited images. Could you please double check that the checkpoint is correct? BTW, why the pose editing is not included in the inference code or playground notebook?

Tengfei-Wang commented 2 years ago

Hi, May I know whether the editing results on other attributes are correct? Since pose editing suffers larger misalignment than others, we adopt a pre-alignment prior to ADA (as the augmented misalignment in the ADA training does cover largely misaligned cases.) We have given the details and some failure cases in the supplementary materials, but I am sure whether the current arxiv version includes this part. There are so many files (e.g., images) in the source file, exceeding the limitation of maximum files of arxiv, so I did not upload the full Appendix on it. I am considering putting the compiled appendix on my webpage latter.

huiqu18 commented 2 years ago

The other editing results are correct. But I didn't find the details about pre-alignment in the Appendix. I am interested in how the pre-alignment is performed as the large pose editing is the most difficult case.