Video2StyleGAN: Disentangling Local and Global Variations in a Video
Primary Area: generative models
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Video Editing, face reenactment, StyleGAN
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: Image editing using a pre-trained StyleGAN generator has emerged as a powerful paradigm for facial editing, providing disentangled controls over age, expression, illumination, etc. However, the approach cannot be directly adopted for video manipulations. We hypothesize that the main missing ingredient is the lack of fine-grained and disentangled control over face location, face pose, and local facial expressions. In this work, we demonstrate that such a fine-grained control is indeed achievable using pre-trained StyleGAN by simultaneously working across multiple (latent) spaces (i.e., positional, W+, and S spaces)
and combining the optimization results. Building on this, we introduce Video2StyleGAN, which takes a target image and driving video(s) to reenact the local and global locations and expressions from the driving video in the identity of the target image. As a result, we are able to generate high-quality videos at 1024x10242 resolution without training on video data.
We evaluate the effectiveness of our method over multiple challenging scenarios and demonstrate clear improvements in terms of LPIPS over alternative approaches trained on video data (FOMM, LIA, and TPS and comparable scores in terms of FID, keypoint distance, and identity preservation.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
Supplementary Material: zip
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 2332
Loading