Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhou, Jinxing, Li, Zhihui, Yu, Yongqiang, Zhou, Yanghao, Guo, Ruohao, Li, Guangyao, Mao, Yuxin, Han, Mingfei, Chang, Xiaojun, Wang, Meng
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2506.23271
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866915364223320064
author Zhou, Jinxing
Li, Zhihui
Yu, Yongqiang
Zhou, Yanghao
Guo, Ruohao
Li, Guangyao
Mao, Yuxin
Han, Mingfei
Chang, Xiaojun
Wang, Meng
author_facet Zhou, Jinxing
Li, Zhihui
Yu, Yongqiang
Zhou, Yanghao
Guo, Ruohao
Li, Guangyao
Mao, Yuxin
Han, Mingfei
Chang, Xiaojun
Wang, Meng
contents We present \textbf{Met}a-\textbf{T}oken \textbf{Le}arning (Mettle), a simple and memory-efficient method for adapting large-scale pretrained transformer models to downstream audio-visual tasks. Instead of sequentially modifying the output feature distribution of the transformer backbone, Mettle utilizes a lightweight \textit{Layer-Centric Distillation (LCD)} module to distill in parallel the intact audio or visual features embedded by each transformer layer into compact meta-tokens. This distillation process considers both pretrained knowledge preservation and task-specific adaptation. The obtained meta-tokens can be directly applied to classification tasks, such as audio-visual event localization and audio-visual video parsing. To further support fine-grained segmentation tasks, such as audio-visual segmentation, we introduce a \textit{Meta-Token Injection (MTI)} module, which utilizes the audio and visual meta-tokens distilled from the top transformer layer to guide feature adaptation in earlier layers. Extensive experiments on multiple audiovisual benchmarks demonstrate that our method significantly reduces memory usage and training time while maintaining parameter efficiency and competitive accuracy.
format Preprint
id arxiv_https___arxiv_org_abs_2506_23271
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Mettle: Meta-Token Learning for Memory-Efficient Audio-Visual Adaptation
Zhou, Jinxing
Li, Zhihui
Yu, Yongqiang
Zhou, Yanghao
Guo, Ruohao
Li, Guangyao
Mao, Yuxin
Han, Mingfei
Chang, Xiaojun
Wang, Meng
Computer Vision and Pattern Recognition
We present \textbf{Met}a-\textbf{T}oken \textbf{Le}arning (Mettle), a simple and memory-efficient method for adapting large-scale pretrained transformer models to downstream audio-visual tasks. Instead of sequentially modifying the output feature distribution of the transformer backbone, Mettle utilizes a lightweight \textit{Layer-Centric Distillation (LCD)} module to distill in parallel the intact audio or visual features embedded by each transformer layer into compact meta-tokens. This distillation process considers both pretrained knowledge preservation and task-specific adaptation. The obtained meta-tokens can be directly applied to classification tasks, such as audio-visual event localization and audio-visual video parsing. To further support fine-grained segmentation tasks, such as audio-visual segmentation, we introduce a \textit{Meta-Token Injection (MTI)} module, which utilizes the audio and visual meta-tokens distilled from the top transformer layer to guide feature adaptation in earlier layers. Extensive experiments on multiple audiovisual benchmarks demonstrate that our method significantly reduces memory usage and training time while maintaining parameter efficiency and competitive accuracy.
title Mettle: Meta-Token Learning for Memory-Efficient Audio-Visual Adaptation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2506.23271