SPDA-SAM: A Self-prompted Depth-Aware Segment Anything Model for Instance Segmentation
Researchers have introduced SPDA-SAM, a novel self-prompted and depth-aware model for instance segmentation that builds upon the Segment Anything Model (SAM). This new model incorporates a Semantic-Spatial Self-prompt Module (SSSPM) to extract prompts from SAM's encoder and decoder, and a Coarse-to-Fine RGB-D Fusion Module (C2FFM) that integrates features from RGB images with estimated depth maps. The C2FFM uses depth information for both coarse-grained structural guidance and fine-grained feature fusion. SPDA-SAM reportedly outperforms existing state-of-the-art methods across twelve datasets by compensating for lost spatial information and leveraging self-generated prompts. AI
IMPACT Enhances instance segmentation capabilities by integrating depth information and self-prompting, potentially improving object boundary delineation and spatial understanding in AI vision systems.