MoTE: Reconciling Generalization with Specialization for Visual-Language to Video Knowledge Transfer

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhu, Minghao, Wang, Zhengpu, Hu, Mengxian, Dang, Ronghao, Lin, Xiao, Zhou, Xun, Liu, Chengju, Chen, Qijun
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914971674214400
author Zhu, Minghao
Wang, Zhengpu
Hu, Mengxian
Dang, Ronghao
Lin, Xiao
Zhou, Xun
Liu, Chengju
Chen, Qijun
author_facet Zhu, Minghao
Wang, Zhengpu
Hu, Mengxian
Dang, Ronghao
Lin, Xiao
Zhou, Xun
Liu, Chengju
Chen, Qijun
contents Transferring visual-language knowledge from large-scale foundation models for video recognition has proved to be effective. To bridge the domain gap, additional parametric modules are added to capture the temporal information. However, zero-shot generalization diminishes with the increase in the number of specialized parameters, making existing works a trade-off between zero-shot and close-set performance. In this paper, we present MoTE, a novel framework that enables generalization and specialization to be balanced in one unified model. Our approach tunes a mixture of temporal experts to learn multiple task views with various degrees of data fitting. To maximally preserve the knowledge of each expert, we propose \emph{Weight Merging Regularization}, which regularizes the merging process of experts in weight space. Additionally with temporal feature modulation to regularize the contribution of temporal feature during test. We achieve a sound balance between zero-shot and close-set video recognition tasks and obtain state-of-the-art or competitive results on various datasets, including Kinetics-400 \& 600, UCF, and HMDB. Code is available at \url{https://github.com/ZMHH-H/MoTE}.
format Preprint
id arxiv_https___arxiv_org_abs_2410_10589
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle MoTE: Reconciling Generalization with Specialization for Visual-Language to Video Knowledge Transfer
Zhu, Minghao
Wang, Zhengpu
Hu, Mengxian
Dang, Ronghao
Lin, Xiao
Zhou, Xun
Liu, Chengju
Chen, Qijun
Computer Vision and Pattern Recognition
Transferring visual-language knowledge from large-scale foundation models for video recognition has proved to be effective. To bridge the domain gap, additional parametric modules are added to capture the temporal information. However, zero-shot generalization diminishes with the increase in the number of specialized parameters, making existing works a trade-off between zero-shot and close-set performance. In this paper, we present MoTE, a novel framework that enables generalization and specialization to be balanced in one unified model. Our approach tunes a mixture of temporal experts to learn multiple task views with various degrees of data fitting. To maximally preserve the knowledge of each expert, we propose \emph{Weight Merging Regularization}, which regularizes the merging process of experts in weight space. Additionally with temporal feature modulation to regularize the contribution of temporal feature during test. We achieve a sound balance between zero-shot and close-set video recognition tasks and obtain state-of-the-art or competitive results on various datasets, including Kinetics-400 \& 600, UCF, and HMDB. Code is available at \url{https://github.com/ZMHH-H/MoTE}.
title MoTE: Reconciling Generalization with Specialization for Visual-Language to Video Knowledge Transfer
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2410.10589