AVESFormer: Efficient Transformer Design for Real-Time Audio-Visual Segmentation
Zili Wang, Qi Yang, Linsu Shi, Jiazhong Yu, Qinghua Liang, Fei Li, Shiming Xiang
2024-08-07
Summary
This paper introduces ReSyncer, a new framework that improves the process of creating lip-synced videos by synchronizing audio and visual elements effectively.
What's the problem?
Creating videos where characters' lips move in sync with audio is challenging. Many existing methods either require long videos for training or produce noticeable errors, which can make the final video look unrealistic. This limits the ability to create high-quality virtual presenters or performers.
What's the solution?
ReSyncer addresses these issues by using a style-based generator that has been reconfigured to better synchronize audio and visual information. It utilizes a special transformer model called Style-SyncFormer to predict 3D facial movements based on audio input. This framework allows for high-quality lip-syncing and supports additional features like transferring speaking styles, personalizing performances quickly, and even swapping faces in videos.
Why it matters?
This research is significant because it enhances the technology behind creating realistic virtual characters that can speak naturally. By improving lip-syncing capabilities and enabling flexible customization, ReSyncer can lead to better applications in entertainment, education, and virtual communication, making digital interactions more engaging and lifelike.
Abstract
Recently, transformer-based models have demonstrated remarkable performance on audio-visual segmentation (AVS) tasks. However, their expensive computational cost makes real-time inference impractical. By characterizing attention maps of the network, we identify two key obstacles in AVS models: 1) attention dissipation, corresponding to the over-concentrated attention weights by Softmax within restricted frames, and 2) inefficient, burdensome transformer decoder, caused by narrow focus patterns in early stages. In this paper, we introduce AVESFormer, the first real-time Audio-Visual Efficient Segmentation transformer that achieves fast, efficient and light-weight simultaneously. Our model leverages an efficient prompt query generator to correct the behaviour of cross-attention. Additionally, we propose ELF decoder to bring greater efficiency by facilitating convolutions suitable for local features to reduce computational burdens. Extensive experiments demonstrate that our AVESFormer significantly enhances model performance, achieving 79.9% on S4, 57.9% on MS3 and 31.2% on AVSS, outperforming previous state-of-the-art and achieving an excellent trade-off between performance and speed. Code can be found at https://github.com/MarkXCloud/AVESFormer.git.