Paper
30 November 2022 DM-CATN: Deep Modular Co-Attention Transformer Networks for image captioning
Xingjian Wang, Xiaolong Fang, You Yang
Author Affiliations +
Proceedings Volume 12456, International Conference on Artificial Intelligence and Intelligent Information Processing (AIIIP 2022); 124562O (2022) https://doi.org/10.1117/12.2659580
Event: International Conference on Artificial Intelligence and Intelligent Information Processing (AIIIP 2022), 2022, Qingdao, China
Abstract
The image features are directly input into the decoding part of the model, which leads to the insufficient use of feature information and makes it difficult for the model to better express the image information. We introduce a Modular Co- Attention Transformer Layer (M-CATL) to efficiently model high-order intra-feature and inter-feature interactions for single and multiple input features to mine the details of image features. And construct a Deep Modular Co-Attention Transformer Block (DM-CATB) according to M-CATL and integrated into the encoder part of the model. Furthermore, we present a Deep Modular Co-Attention Transformer Network (DM-CATN) to fully model the spatial information and position information of image features and improve the ability of features characterization, in order to provide richer image information for decoding part. Experimental results demonstrate that DM-CATN significantly outperforms the previous state-of-the-art. Our best single model delivers 133.2% in CIDEr.
© (2022) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Xingjian Wang, Xiaolong Fang, and You Yang "DM-CATN: Deep Modular Co-Attention Transformer Networks for image captioning", Proc. SPIE 12456, International Conference on Artificial Intelligence and Intelligent Information Processing (AIIIP 2022), 124562O (30 November 2022); https://doi.org/10.1117/12.2659580
Advertisement
Advertisement
RIGHTS & PERMISSIONS
Get copyright permission  Get copyright permission on Copyright Marketplace
KEYWORDS
Transformers

Visualization

Image fusion

Performance modeling

Computer programming

Matrices

Network architectures

Back to Top