MSDA-SparseInst: Real-Time Instance Segmentation via Multi-Scale Feature Fusion and Dual-Branch Attention

Read full text Summarise with VIRAT
Authors: Y. Chen
Cited by: 0
References: 27

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Real-time instance segmentation is crucial for urban road-scene understanding, where accurate pixel-level perception is required under complex backgrounds, occlusion, and large scale variation. However, existing efficient methods often struggle to balance segmentation accuracy and inference speed, especially for small distant objects and densely distributed instances. To address this issue, this paper proposes MSDA-SparseInst, a real-time instance segmentation framework based on SparseInst. Specifically, an improved backbone is adopted to enhance feature extraction, a Multi-scale Dilated Feature Aggregation (MDFA) module is introduced to strengthen cross-scale contextual modeling, and a lightweight dual-branch attention strategy composed of GCSA and GGCA is designed to refine decoder features. Experimental results on the Cityscapes validation set show that the proposed method achieves 21.8 AP, 43.4 AP50, and 18.1 AP75 at 30.7 FPS, improving the baseline SparseInst by 3.0 AP while maintaining real-time performance. The results demonstrate that MSDA-SparseInst provides a better trade-off between segmentation accuracy and efficiency for urban road-scene instance segmentation.

Fields of Study & Subjects

Advanced Neural Network Applications Domain Adaptation and Few-Shot Learning Visual Attention and Saliency Detection

Publication details

Journal / source
Journal of Computing and Electronic Information Management
Volume
20
Issue
3
Pages
79–87
Publication date
26 March 2026
Language
EN