Related Experiment Video
Updated: Jan 10, 2026
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
996
A lightweight transformer framework for open set anomaly segmentation in smart city applications.
M Manicka Prabha1, K Suganthi2
1School of Computer Science and Engineering, Vellore Institute of Technology, Chennai, Tamil Nadu, 600127, India.
Scientific Reports
|November 27, 2025
Summary
LightMask offers an efficient solution for open-set anomaly segmentation in urban environments. This lightweight transformer model achieves high accuracy while significantly reducing computational overhead, making it ideal for real-world infrastructure applications.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Machine Learning
Background:
- Open-set anomaly segmentation in diverse infrastructure presents significant computational and accuracy challenges.
- Existing transformer-based methods often struggle with the trade-off between computational efficiency and performance.
Purpose of the Study:
- To introduce LightMask, a novel lightweight transformer architecture for efficient and context-aware anomaly segmentation in complex urban settings.
- To address the limitations of current methods by prioritizing computational efficiency without compromising anomaly detection performance.
Main Methods:
- An optimized EfficientNet-B0 backbone for efficient feature extraction.
- An adaptive inference mechanism and separable self-attention (SSA) with linear complexity.
- A progressive multi-scale decoder featuring dynamic early termination and boundary-aware contrastive loss.
Main Results:
- LightMask demonstrates a lightweight structure with 4.29 million parameters (16.35 MB) and low computational cost (8.72 GFLOPs).
- Achieved robust performance on Cityscapes and RoadAnomaly datasets, including 91.79% precision, 93% recall, and 77.66% F1 score.
- Demonstrated strong anomaly detection capabilities with 88.28% AUC-ROC and a low false positive rate (36.24% at 95% TPR).
Conclusions:
- LightMask effectively balances computational efficiency with robust anomaly detection capabilities for open-set segmentation tasks.
- The proposed architecture provides a practical solution for real-time anomaly segmentation in resource-constrained urban infrastructure environments.
