Saved in:
Bibliographic Details
Main Authors: Zhang, Yurong, Chen, Honghao, Zhang, Xinyu, Chu, Xiangxiang, Song, Li
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2407.14302
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914882360705024
author Zhang, Yurong
Chen, Honghao
Zhang, Xinyu
Chu, Xiangxiang
Song, Li
author_facet Zhang, Yurong
Chen, Honghao
Zhang, Xinyu
Chu, Xiangxiang
Song, Li
contents Parameter-efficient transfer learning (PETL) is a promising task, aiming to adapt the large-scale pre-trained model to downstream tasks with a relatively modest cost. However, current PETL methods struggle in compressing computational complexity and bear a heavy inference burden due to the complete forward process. This paper presents an efficient visual recognition paradigm, called Dynamic Adapter (Dyn-Adapter), that boosts PETL efficiency by subtly disentangling features in multiple levels. Our approach is simple: first, we devise a dynamic architecture with balanced early heads for multi-level feature extraction, along with adaptive training strategy. Second, we introduce a bidirectional sparsity strategy driven by the pursuit of powerful generalization ability. These qualities enable us to fine-tune efficiently and effectively: we reduce FLOPs during inference by 50%, while maintaining or even yielding higher recognition accuracy. Extensive experiments on diverse datasets and pretrained backbones demonstrate the potential of Dyn-Adapter serving as a general efficiency booster for PETL in vision recognition tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2407_14302
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Dyn-Adapter: Towards Disentangled Representation for Efficient Visual Recognition
Zhang, Yurong
Chen, Honghao
Zhang, Xinyu
Chu, Xiangxiang
Song, Li
Computer Vision and Pattern Recognition
Parameter-efficient transfer learning (PETL) is a promising task, aiming to adapt the large-scale pre-trained model to downstream tasks with a relatively modest cost. However, current PETL methods struggle in compressing computational complexity and bear a heavy inference burden due to the complete forward process. This paper presents an efficient visual recognition paradigm, called Dynamic Adapter (Dyn-Adapter), that boosts PETL efficiency by subtly disentangling features in multiple levels. Our approach is simple: first, we devise a dynamic architecture with balanced early heads for multi-level feature extraction, along with adaptive training strategy. Second, we introduce a bidirectional sparsity strategy driven by the pursuit of powerful generalization ability. These qualities enable us to fine-tune efficiently and effectively: we reduce FLOPs during inference by 50%, while maintaining or even yielding higher recognition accuracy. Extensive experiments on diverse datasets and pretrained backbones demonstrate the potential of Dyn-Adapter serving as a general efficiency booster for PETL in vision recognition tasks.
title Dyn-Adapter: Towards Disentangled Representation for Efficient Visual Recognition
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2407.14302