search

yuangan / EAT_code

Official code for ICCV 2023 paper: "Efficient Emotional Adaptation for Audio-Driven Talking-Head Generation".
Other
275 stars 31 forks source link

3D Keypoint Enhancement Training #29

Closed JamesLong199 closed 6 months ago

JamesLong199 commented 6 months ago

Hello, thank you for your awesome research.

I was wondering if the training code and instructions for 3D keypoint enhancement (Face-vid2vid) are included in the repo.

To my understanding, a Face-vid2vid model is first trained from scratch on the MEAD dataset following the modification specified in section 3.1.1 in the paper, before the training of A2ET. I was wondering if this understanding is correct, and if there is an ablation study on the keypoint enhancement.

Thank you in advance for your time and help!

yuangan commented 6 months ago

Hi, thank you for your attention.

The training codes for 3D keypoint enhancement are not included in this repository as they are developed using a different framework. We adapted the public implementation of OSFV(face-vid2vid) modifying the training process as described in Section 3.1.1 of our paper.

We have detailed the ablation study of the enhanced 3D keypoints in Section 4.6 and Table 3 of our paper, comparing it with the publicly available OSFV.

If you are interested in re-implementing the enhanced 3D keypoints, we would be pleased to offer any help we can.

JamesLong199 commented 6 months ago

Thank you so much for your swift response and clarification!