CTNet

CTNet integrates spatial and channel contextual information to discover semantic contexts and improve accuracy in semantic image segmentation.

Key Features:

• Spatial Contextual Module (SCM): The SCM uncovers spatial contextual dependencies by exploring correlations between individual pixels and their associated categories to capture long-range spatial dependencies.
• Channel Contextual Module (CCM): The CCM models long-term semantic dependence between channels and generates semantic feature maps and class-specific features that serve as prior knowledge to guide the SCM.
• Adaptive Integration: CTNet adaptively integrates outputs from the SCM and CCM to improve learned representations for semantic segmentation.

Scientific Applications:

• Benchmark evaluation: CTNet was evaluated on PASCAL-Context, Cityscapes, ADE20K, and PASCAL VOC2012, demonstrating superior performance compared with several state-of-the-art semantic segmentation methods.

Methodology:

CTNet interactively leverages a Spatial Contextual Module (pixel-category correlations to capture long-range spatial dependencies) and a Channel Contextual Module (channel-wise semantic dependence producing semantic feature maps and class-specific features), with adaptive integration of their outputs to guide learning.