Closed johndpope closed 6 months ago
https://openaccess.thecvf.com/content/CVPR2021/supplemental/Wang_One-Shot_Free-View_Neural_CVPR_2021_supplemental.pdf
when I dig into where the keypoints live , we utilize a similar volumetric encoding of the appearance but instead encode the face motion implicitly, which improves cross-reenactment performance.
can rip out face_alignment from emodataset.
https://openaccess.thecvf.com/content/CVPR2021/supplemental/Wang_One-Shot_Free-View_Neural_CVPR_2021_supplemental.pdf
when I dig into where the keypoints live , we utilize a similar volumetric encoding of the appearance but instead encode the face motion implicitly, which improves cross-reenactment performance.
can rip out face_alignment from emodataset.