Multibranch Semantic Image Segmentation Model
Author Information
Author(s): Yang Zhuolin, Cao Zhen, Cao Jianfang, Chen Zhiqiang, Peng Cunhe
Primary Institution: Xinzhou Normal University, China
Hypothesis
Can an edge optimization and category-aware multibranch semantic segmentation network improve segmentation accuracy?
Conclusion
The proposed ECMNet model significantly improves semantic image segmentation performance, achieving MIoU values of 79.2% on the Cityspaces dataset and 79.6% on the CamVid dataset.
Supporting Evidence
- The ECMNet model achieved a MIoU of 79.2% on the Cityspaces dataset, outperforming existing models.
- The model's parameter count is significantly lower than that of other advanced models while maintaining high accuracy.
- ECMNet effectively segments small targets and complex edges due to its edge optimization module.
Takeaway
This study created a new model that helps computers better understand images by focusing on edges and different categories, making it easier to tell objects apart.
Methodology
The study used a multibranch network structure with attention mechanisms and edge optimization techniques to enhance image segmentation.
Limitations
The model struggles with complex postures and rich details in certain categories like pedestrians and cyclists.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website