Convolutional Initialization for Data-Efficient Vision Transformers

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zheng, Jianqiao, Li, Xueqian, Lucey, Simon
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929219821371392
author Zheng, Jianqiao
Li, Xueqian
Lucey, Simon
author_facet Zheng, Jianqiao
Li, Xueqian
Lucey, Simon
contents Training vision transformer networks on small datasets poses challenges. In contrast, convolutional neural networks (CNNs) can achieve state-of-the-art performance by leveraging their architectural inductive bias. In this paper, we investigate whether this inductive bias can be reinterpreted as an initialization bias within a vision transformer network. Our approach is motivated by the finding that random impulse filters can achieve almost comparable performance to learned filters in CNNs. We introduce a novel initialization strategy for transformer networks that can achieve comparable performance to CNNs on small datasets while preserving its architectural flexibility.
format Preprint
id arxiv_https___arxiv_org_abs_2401_12511
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Convolutional Initialization for Data-Efficient Vision Transformers
Zheng, Jianqiao
Li, Xueqian
Lucey, Simon
Computer Vision and Pattern Recognition
Training vision transformer networks on small datasets poses challenges. In contrast, convolutional neural networks (CNNs) can achieve state-of-the-art performance by leveraging their architectural inductive bias. In this paper, we investigate whether this inductive bias can be reinterpreted as an initialization bias within a vision transformer network. Our approach is motivated by the finding that random impulse filters can achieve almost comparable performance to learned filters in CNNs. We introduce a novel initialization strategy for transformer networks that can achieve comparable performance to CNNs on small datasets while preserving its architectural flexibility.
title Convolutional Initialization for Data-Efficient Vision Transformers
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2401.12511