Motif Channel Opened in a White-Box: Stereo Matching via Motif Correlation Graph
Stereo matching often loses important geometric details when using learning methods because they lose structural information. This work introduces MoCha-V2, a new method that uses a Motif Correlation Graph (MCG) to identify recurring textures as 'motifs.' By capturing these motifs, the system reconstructs precise geometric structures in an interpretable way.