Novel Visual Reasoning Approach Using Object-Centric Slot Attention
2025-06-08
Researchers propose a novel visual reasoning approach combining object-centric slot attention and a relational bottleneck. The method first uses a CNN to extract image features. Then, slot attention segments the image into objects, generating object-centric visual representations. The relational bottleneck restricts information flow, extracting abstract relationships between objects for understanding complex scenes. Finally, a sequence-to-sequence and algebraic machine reasoning framework transforms visual reasoning into an algebraic problem, improving efficiency and accuracy. The method excels in visual reasoning tasks like Raven's Progressive Matrices.