SAM3-DMS: Decoupled Memory Selection for Multi-target Video Segmentation of SAM3
This addresses the problem of multi-target video segmentation in complex scenarios for computer vision applications, representing an incremental improvement over SAM3.
The paper tackles the suboptimal group-level memory selection in Segment Anything 3 (SAM3) for multi-object video segmentation by proposing SAM3-DMS, a training-free decoupled strategy that uses fine-grained memory selection on individual objects, resulting in robust identity preservation and tracking stability with advantages that increase with target density.
Segment Anything 3 (SAM3) has established a powerful foundation that robustly detects, segments, and tracks specified targets in videos. However, in its original implementation, its group-level collective memory selection is suboptimal for complex multi-object scenarios, as it employs a synchronized decision across all concurrent targets conditioned on their average performance, often overlooking individual reliability. To this end, we propose SAM3-DMS, a training-free decoupled strategy that utilizes fine-grained memory selection on individual objects. Experiments demonstrate that our approach achieves robust identity preservation and tracking stability. Notably, our advantage becomes more pronounced with increased target density, establishing a solid foundation for simultaneous multi-target video segmentation in the wild.