Research on contrastive image classification algorithm based on scalable self-attention

Xin Tian; Jian Liu; Zhenwei Cui; Yueming Guo; Yifei Xu; Yuan Sun; Kai Qiao

doi:10.1117/12.3029022

22 May 2024 Research on contrastive image classification algorithm based on scalable self-attention

Xin Tian, Jian Liu, Zhenwei Cui, Yueming Guo, Yifei Xu, Yuan Sun, Kai Qiao

Proceedings Volume 13176, Fourth International Conference on Machine Learning and Computer Application (ICMLCA 2023); 131760A (2024) https://doi.org/10.1117/12.3029022
Event: Fourth International Conference on Machine Learning and Computer Application (ICMLCA 2023), 2023, Hangzhou, China

Abstract

This study introduces a scalable self-attention based contrastive image classification algorithm, aiming to train a backbone network suitable for downstream image classification tasks through contrastive learning. It begins by addressing the challenges of high training overhead and low model accuracy encountered when applying the contrastive learning model represented by Moco V3 to image classification. Subsequently, the Momentum Contrast For Image Classification (MocoIC) algorithm is proposed as a solution to these issues. The MocoIC algorithm comprises the Scalable Shift Window Transformer (SSWT) feature extraction network, feature mapping network, prediction head, and Info NCE loss function. This model exhibits superior algorithmic performance and requires less training overhead compared to the original contrastive learning model.

(2024) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Xin Tian, Jian Liu, Zhenwei Cui, Yueming Guo, Yifei Xu, Yuan Sun, and Kai Qiao "Research on contrastive image classification algorithm based on scalable self-attention", Proc. SPIE 13176, Fourth International Conference on Machine Learning and Computer Application (ICMLCA 2023), 131760A (22 May 2024); https://doi.org/10.1117/12.3029022

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available