Self-Evolving Spatial Reasoning in Vision Language Models via Geometric Logic Consistency
The Spatial Alignment via Geometric Evolution (SAGE) framework enforces logical consistency in Vision-Language Models by treating paired transformations as duality operations. It probes model inconsistencies using visual and linguistic dualities, evolves an operation pool based on performance, and integrates this consistency as a reward during training. This results in improved spatial reasoning capabilities with less training data.