Partial Ring Scan: Revisiting Scan Order in Vision State Space Models
Researchers have introduced PRISMamba, a novel approach to processing images within Vision State Space Models (SSMs). Unlike traditional methods that serialize images into linear sequences, PRISMamba partitions images into concentric rings and aggregates information within each ring. This method enhances rotation robustness and improves efficiency by selectively filtering channels. PRISMamba achieves competitive accuracy on ImageNet-1K while demonstrating superior throughput and fewer FLOPs compared to existing VMamba models, particularly maintaining performance under rotational transformations. AI
IMPACT Introduces a more rotationally robust and efficient method for processing images in Vision SSMs, potentially improving performance in applications sensitive to spatial orientation.