Cross-modal learning with multi-modal model for video action recognition based on adaptive weight training

The canonical video action recognition methods usually label categories with numbers or one-hot vectors and train neural networks to classify a fixed set of predefined categories, thereby constraining their ability to recognise complex actions and transferable ability to unseen concepts. In contrast...

Full description

Saved in:
Bibliographic Details
Main Authors: Qingguo Zhou, Yufeng Hou, Rui Zhou, Yan Li, JinQiang Wang, Zhen Wu, Hung-Wei Li, Tien-Hsiung Weng
Format: Article
Language:English
Published: Taylor & Francis Group 2024-12-01
Series:Connection Science
Subjects:
Online Access:https://www.tandfonline.com/doi/10.1080/09540091.2024.2325474
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849691071502614528
author Qingguo Zhou
Yufeng Hou
Rui Zhou
Yan Li
JinQiang Wang
Zhen Wu
Hung-Wei Li
Tien-Hsiung Weng
author_facet Qingguo Zhou
Yufeng Hou
Rui Zhou
Yan Li
JinQiang Wang
Zhen Wu
Hung-Wei Li
Tien-Hsiung Weng
author_sort Qingguo Zhou
collection DOAJ
description The canonical video action recognition methods usually label categories with numbers or one-hot vectors and train neural networks to classify a fixed set of predefined categories, thereby constraining their ability to recognise complex actions and transferable ability to unseen concepts. In contrast, cross-modal learning can improve the performance of individual modalities. Based on the facts that a better action recogniser can be built by reading the statements used to describe actions, we exploited the recent multimodal foundation model CLIP for action recognition. In this study, an effective Vision-Language action recognition adaptation was implemented based on few-shot examples spanning different modalities. We added semantic information to action categories by treating textual and visual label as training examples for action classifier construction rather than simply labelling them with numbers. Due to the different importance of words in text and video frames, simply averaging all sequential features may result in ignoring keywords or key video frames. To capture sequential and hierarchical representation, a weighted token-wise interaction mechanism was employed to exploit the pair-wise correlations adaptively. Extensive experiments with public datasets show that cross-modal action recognition learning helps for downstream action images classification, in other words, the proposed method can train better action classifiers by reading the sentences describing action itself. The method proposed in this study not only reaches good generalisation and zero-shot/few-shot transfer ability on Out of Distribution (OOD) test sets, but also performs lower computational complexity due to the lightweight interaction mechanism with 84.15% Top-1 accuracy on the Kinetics-400.
format Article
id doaj-art-e8b24496427948bf9b08d96462eb8a69
institution DOAJ
issn 0954-0091
1360-0494
language English
publishDate 2024-12-01
publisher Taylor & Francis Group
record_format Article
series Connection Science
spelling doaj-art-e8b24496427948bf9b08d96462eb8a692025-08-20T03:21:09ZengTaylor & Francis GroupConnection Science0954-00911360-04942024-12-0136110.1080/09540091.2024.2325474Cross-modal learning with multi-modal model for video action recognition based on adaptive weight trainingQingguo Zhou0Yufeng Hou1Rui Zhou2Yan Li3JinQiang Wang4Zhen Wu5Hung-Wei Li6Tien-Hsiung Weng7School of Information Science and Engineering, Lanzhou University, Lanzhou, People's Republic of ChinaSchool of Information Science and Engineering, Lanzhou University, Lanzhou, People's Republic of ChinaSchool of Information Science and Engineering, Lanzhou University, Lanzhou, People's Republic of ChinaSchool of Information Science and Engineering, Lanzhou University, Lanzhou, People's Republic of ChinaSchool of Information Science and Engineering, Lanzhou University, Lanzhou, People's Republic of ChinaSchool of Information Science and Engineering, Lanzhou University, Lanzhou, People's Republic of ChinaDepartment of Computer Science and Information Engineering, Providence University, Taichung City, TaiwanDepartment of Computer Science and Information Engineering, Providence University, Taichung City, TaiwanThe canonical video action recognition methods usually label categories with numbers or one-hot vectors and train neural networks to classify a fixed set of predefined categories, thereby constraining their ability to recognise complex actions and transferable ability to unseen concepts. In contrast, cross-modal learning can improve the performance of individual modalities. Based on the facts that a better action recogniser can be built by reading the statements used to describe actions, we exploited the recent multimodal foundation model CLIP for action recognition. In this study, an effective Vision-Language action recognition adaptation was implemented based on few-shot examples spanning different modalities. We added semantic information to action categories by treating textual and visual label as training examples for action classifier construction rather than simply labelling them with numbers. Due to the different importance of words in text and video frames, simply averaging all sequential features may result in ignoring keywords or key video frames. To capture sequential and hierarchical representation, a weighted token-wise interaction mechanism was employed to exploit the pair-wise correlations adaptively. Extensive experiments with public datasets show that cross-modal action recognition learning helps for downstream action images classification, in other words, the proposed method can train better action classifiers by reading the sentences describing action itself. The method proposed in this study not only reaches good generalisation and zero-shot/few-shot transfer ability on Out of Distribution (OOD) test sets, but also performs lower computational complexity due to the lightweight interaction mechanism with 84.15% Top-1 accuracy on the Kinetics-400.https://www.tandfonline.com/doi/10.1080/09540091.2024.2325474Adaptive weight trainingcross-modal learningvideo action recognitionvision-Language adaptation
spellingShingle Qingguo Zhou
Yufeng Hou
Rui Zhou
Yan Li
JinQiang Wang
Zhen Wu
Hung-Wei Li
Tien-Hsiung Weng
Cross-modal learning with multi-modal model for video action recognition based on adaptive weight training
Connection Science
Adaptive weight training
cross-modal learning
video action recognition
vision-Language adaptation
title Cross-modal learning with multi-modal model for video action recognition based on adaptive weight training
title_full Cross-modal learning with multi-modal model for video action recognition based on adaptive weight training
title_fullStr Cross-modal learning with multi-modal model for video action recognition based on adaptive weight training
title_full_unstemmed Cross-modal learning with multi-modal model for video action recognition based on adaptive weight training
title_short Cross-modal learning with multi-modal model for video action recognition based on adaptive weight training
title_sort cross modal learning with multi modal model for video action recognition based on adaptive weight training
topic Adaptive weight training
cross-modal learning
video action recognition
vision-Language adaptation
url https://www.tandfonline.com/doi/10.1080/09540091.2024.2325474
work_keys_str_mv AT qingguozhou crossmodallearningwithmultimodalmodelforvideoactionrecognitionbasedonadaptiveweighttraining
AT yufenghou crossmodallearningwithmultimodalmodelforvideoactionrecognitionbasedonadaptiveweighttraining
AT ruizhou crossmodallearningwithmultimodalmodelforvideoactionrecognitionbasedonadaptiveweighttraining
AT yanli crossmodallearningwithmultimodalmodelforvideoactionrecognitionbasedonadaptiveweighttraining
AT jinqiangwang crossmodallearningwithmultimodalmodelforvideoactionrecognitionbasedonadaptiveweighttraining
AT zhenwu crossmodallearningwithmultimodalmodelforvideoactionrecognitionbasedonadaptiveweighttraining
AT hungweili crossmodallearningwithmultimodalmodelforvideoactionrecognitionbasedonadaptiveweighttraining
AT tienhsiungweng crossmodallearningwithmultimodalmodelforvideoactionrecognitionbasedonadaptiveweighttraining