Few-shot object detection based on multi-scale attention model

Zongbo Hao; Juncong Lu; Aoyu Luo

doi:10.1117/12.3026182

6 May 2024 Few-shot object detection based on multi-scale attention model

Zongbo Hao, Juncong Lu, Aoyu Luo

Proceedings Volume 13161, Fourth International Conference on Telecommunications, Optics, and Computer Science (TOCS 2023); 131610L (2024) https://doi.org/10.1117/12.3026182
Event: Fourth International Conference on Telecommunications, Optics and Computer Science (TOCS 2023), 2023, Xi’an, China

Abstract

Conventional deep learning based object detection methods demand substantial annotated data for training, incurring considerable time and labor costs. Conversely, few-shot object detection necessitates only limited data from novel categories, emerging as a prominent research focus. This study proposes the Attention Contrastive Network (ACNet) to address few-shot object detection challenges. ACNet incorporates an attention mechanism architecture, extracting attention values and keys from image features in both support and query sets. It compares key attention across the sets and weights query set features with attention to augment local features. Additionally, multi-scale pooling layers enhance the network's capability to identify objects across varying scales. The introduction of an attract-repel mechanism in the loss function significantly amplifies inter-class differences, thereby improving classification accuracy. ACNet's efficacy is experimentally affirmed on the PASCAL VOC and COCO datasets, yielding commendable results in few-shot detection tasks.

(2024) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Zongbo Hao, Juncong Lu, and Aoyu Luo "Few-shot object detection based on multi-scale attention model", Proc. SPIE 13161, Fourth International Conference on Telecommunications, Optics, and Computer Science (TOCS 2023), 131610L (6 May 2024); https://doi.org/10.1117/12.3026182

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available