Related Experiment Video
Updated: Jul 8, 2026
06:37
Artificial Intelligence-Based System for Detecting Attention Levels in Students
Published on: December 15, 2023
MHS-VIT: Mamba hybrid self-attention vision transformers for traffic image detection
Xude Zhang1,2, Weihua Ou3, Xiaoping Wu2
1Engineering Research Center of Micro-Nano and Intelligent Manufacturing, Ministry of Education, Kaili University, Kaili, Guizhou, China.
Plos One
|June 30, 2025
Summary
A new Mamba Hybrid Self-Attention Vision Transformer (MHS-VIT) improves traffic image detection by combining Mamba state-space models and transformers. This hybrid model enhances efficiency and accuracy for tasks like vehicle and pedestrian detection.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Intelligent Transportation Systems
Background:
- Transformer models have advanced traffic image detection but suffer from high computational costs due to quadratic self-attention complexity.
- This limits their deployment in resource-constrained intelligent transportation systems.
- Efficient processing of complex traffic scenes requires models that balance performance with computational efficiency.
Purpose of the Study:
- To introduce a novel hybrid architecture, Mamba Hybrid Self-Attention Vision Transformers (MHS-VIT), for improved traffic image detection.
- To leverage the linear complexity of Mamba state-space models (SSMs) and the spatial dependency modeling of transformers.
- To enhance both the efficiency and accuracy of visual processing in traffic image analysis.
Main Methods:
- Developed a hybrid architecture (MHS-VIT) integrating Mamba SSMs with transformer components.
- Utilized Mamba's linear time complexity to reduce computational burden.
- Employed transformer's self-attention for capturing long-range spatial dependencies critical for traffic scenes.
Main Results:
- MHS-VIT demonstrated excellent performance in traffic image detection tasks, including vehicle detection, pedestrian detection, and traffic sign recognition.
- The model achieved accurate and rapid identification of target objects.
- Compared to same-scale backbone networks, MHS-VIT showed significant improvements in accuracy and reduced model parameter count.
Conclusions:
- The proposed MHS-VIT effectively addresses the computational limitations of traditional transformers in traffic image analysis.
- The hybrid approach successfully combines the strengths of Mamba SSMs and transformers for efficient and accurate visual perception.
- MHS-VIT offers a promising solution for advanced intelligent transportation systems requiring high-performance, resource-efficient image detection.