logo
CanyamAI summaries for academic research
language
MobileMobile
  • Phone FeatureWeChat Mini Program
EnglishEnglish
noticeNotice
BlogBlog
delete
arrow
Return

Bidirectional Segmentation-Aware Network for One-Shot Object Detection

delete2025-06-26
delete0
PRE
AI
Z
Zhang, Wenwen
Z
Zhenghua Chen
Y
Yongyi Su
Z
Zhiyu Xiang
H
Hangguan Shan
刘
刘而云 (Eryun Liu)
DOI：10.1016/j.neucom.2025.130761delete
deleteOriginal
deleteOriginal request for help
deleteShare
deleteSave
Abstract

Abstract

En 中文
One-shot object detection (OSOD) has recently emerged as a prominent research focus, owing to its capability to detect novel-class objects in target images using only a single query image as guidance. However, despite significant advances in existing query-guided frameworks, these models often struggle with identifying small objects and distinguishing objects from distractors due to the progressive loss of fine-grained details. Additionally, the semantic gap between query-target image pairs in cluttered scenes leads to missed detections stemming from insufficient contextual understanding. To address these limitations, we propose a novel unified framework, Bidirectional Segmentation-Aware Network (BSANet), which explores the reciprocal relationship between segmentation and detection tasks. The integration of the segmentation model, which excels at capturing fine-grained details, effectively complements and enhances detection performance. Specifically, we design an Instance-Guided Segmentation (IGS) module that performs segmentation guided by instance-level priors, aiming to introduce instance awareness to the pixel level. Furthermore, we develop a Segmentation-Aware Prediction (SAP) module to generate a segmentation-aware prototype enriched with fine-grained and homogeneous information, which is then matched with proposal features to facilitate instance-level score prediction in turn. Information from the two branches is bidirectionally fused, enhancing contextual understanding and sensitivity to fine-grained details. Extensive experiments demonstrate that our method achieves state-of-the-art performance on PASCAL VOC and COCO datasets in diverse evaluation settings.

Journal

Neurocomputing cover
Neurocomputing
IF:
6.5
Papers:
2.5W
Citations:
6.5W

Organization

No organization information available