Bilateral attention decoder: A lightweight decoder for real-time semantic segmentation.

Area of Science:

• Computer Vision
• Deep Learning
• Artificial Intelligence

Background:

• Encoder-decoder networks are used for semantic segmentation, but complex decoders hinder real-time performance.
• Existing methods struggle to balance accuracy and inference speed in semantic segmentation.

Purpose of the Study:

• To propose a lightweight bilateral attention decoder for real-time semantic segmentation.
• To improve the fusion of high- and low-level feature maps for enhanced accuracy and efficiency.

Main Methods:

• A novel bilateral attention decoder with channel and spatial attention branches for feature refinement.
• A pooling fusing block for effective fusion of refined high- and low-level feature maps.
• Integration with a lightweight backbone network for real-time applications.

Main Results:

• The proposed method achieves higher inference speed compared to state-of-the-art real-time semantic segmentation methods.
• Experimental results on Cityscapes and Camvid datasets demonstrate superior segmentation performance.
• The lightweight decoder also outperforms some non-real-time semantic segmentation methods.

Conclusions:

• The lightweight bilateral attention decoder effectively addresses the real-time performance limitations of encoder-decoder models.
• The proposed attention and fusion mechanisms significantly improve semantic segmentation accuracy and efficiency.
• This method offers a promising solution for real-time semantic segmentation tasks requiring high accuracy.