Improvement of Vision Transformer Using Word Patches

Ayato Takama, Sota Kato, Satoshi Kamiya, Kazuhiro Hotta

2023

Abstract

Vision Transformer achieves higher accuracy on image classification than conventional convolutional neural networks. However, Vision Transformer requires more training images than conventional neural networks. Since there is no clear concept of words in images, we created Visual Words by cropping training images and clustering them using K-means like bag-of-visual words, and incorporated them into Vision Transformer as ”Word Patches” to improve the accuracy. We also try trainable words instead of visual words by clustering. Experiments were conducted to confirm the effectiveness of the proposed method. When Word Patches are trainable parameters, the accuracy was much improved from 84.16% to 87.35% on the Food101 dataset.

Paper Citation

in EndNote Style

TY - CONF

JO - Proceedings of the 18th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2023) - Volume 5: VISAPP
TI - Improvement of Vision Transformer Using Word Patches
SN - 978-989-758-634-7
AU - Takama A.
AU - Kato S.
AU - Kamiya S.
AU - Hotta K.
PY - 2023
SP - 731
EP - 736
DO - 10.5220/0011732900003417
PB - SciTePress

in Harvard Style

Takama A., Kato S., Kamiya S. and Hotta K. (2023). Improvement of Vision Transformer Using Word Patches. In Proceedings of the 18th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2023) - Volume 5: VISAPP; ISBN 978-989-758-634-7, SciTePress, pages 731-736. DOI: 10.5220/0011732900003417

in Bibtex Style

@conference{visapp23,
author={Ayato Takama and Sota Kato and Satoshi Kamiya and Kazuhiro Hotta},
title={Improvement of Vision Transformer Using Word Patches},
booktitle={Proceedings of the 18th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2023) - Volume 5: VISAPP},
year={2023},
pages={731-736},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0011732900003417},
isbn={978-989-758-634-7},
}