Multi-granularity contrastive zero-shot learning model based on attribute decomposition

Page view(s)
5
Checked on Aug 09, 2025
Multi-granularity contrastive zero-shot learning model based on attribute decomposition
Title:
Multi-granularity contrastive zero-shot learning model based on attribute decomposition
Journal Title:
Information Processing & Management
Keywords:
Publication Date:
21 September 2024
Citation:
Wang, Y., Wang, J., Fan, Y., Chai, Q., Zhang, H., Li, X., & Li, R. (2025). Multi-granularity contrastive zero-shot learning model based on attribute decomposition. Information Processing & Management, 62(1), 103898. https://doi.org/10.1016/j.ipm.2024.103898
Abstract:
Zero-shot learning aims to identify new classes by transferring semantic knowledge from seen classes to unseen classes. Image classification based on zero-shot learning usually regards different attributes as the carriers of semantic knowledge, for different attributes have different effects on the task. However, existing models lack a differentiated understanding of different attributes and ignore the impact of global context information. Therefore, this paper proposes a multi-granularity contrastive zero-shot learning model based on attribute decomposition. Specifically, different attributes are initially divided into key and common attributes i.e., attribute decomposition. Then, inspired by Navon’s global-local paradigm, we work out the multigranularity contrastive learning model, which is composed of the global learning module and the local one, to further enhance the interaction between the global and local information. Finally, zero-shot image classification is achieved by training a multi-granularity contrastive learning model. The method is experimented on three public zero-shot learning benchmark data sets (i.e., AWA2, CUB and SUN). Compared with the existing model, in the CUB dataset, the harmonic mean H is improved by 1.6%; in the SUN dataset, H increased by 7.1%; in the AWA2 dataset, H increased by 2.5%, further verifying the effectiveness of this model.
License type:
Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0)
Funding Info:
This research / project is supported by the National Science and Technology Major Project - NA
Grant Reference no. : 2020AAA0106102

This research / project is supported by the the National Natural Science Foundation of China - NA
Grant Reference no. : 62176145
Description:
ISSN:
0306-4573
Files uploaded:

File Size Format Action
yuanlong-wang-final.pdf 3.83 MB PDF Request a copy