Multi-Target Pan-Class Intrinsic Relevance Driven Model for Improving Semantic Segmentation in Autonomous Driving

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Current semantic segmentation models heavily depend on deep learning for feature extraction.
• Despite high performance, refining target segmentation and overcoming limitations remain active research areas.
• A novel pan-class intrinsic relevance phenomenon, linking targets across classes, is identified.

Purpose of the Study:

• To propose a model for refining target main body segmentation using multi-target pan-class intrinsic relevance.
• To introduce a cross-class strategy distinct from existing intra-class and inter-class segmentation methods.

Main Methods:

• Establish prior knowledge of intrinsic relevance using the multi-target pan-class intrinsic relevance prior knowledge establishment (RPK-Est) module.
• Extract pan-class intrinsic relevance features via a multi-target node graph and graph convolution network in the RF-Ext module.
• Integrate intrinsic relevance and semantic features using generative adversarial learning at the gradient level in the RF-Int module.

Main Results:

• The proposed model demonstrated outstanding performance in semantic segmentation tasks.
• Evaluated on four authoritative datasets, the model surpassed current state-of-the-art methods.
• The integration strategy effectively utilized intrinsic relevance features for improved segmentation.

Conclusions:

• The developed model successfully refines target main body segmentation by exploiting pan-class intrinsic relevance.
• This approach offers a new direction for semantic segmentation, moving beyond traditional context-based methods.
• The findings highlight the potential of cross-class relevance for enhancing segmentation accuracy and robustness.