SAM-Deep-EIoU: Selective Mask Propagation for Multi-Object Tracking
Researchers have developed SAM-Deep-EIoU, a novel approach to multi-object tracking that selectively employs a more powerful video object segmentation (VOS) model only when a base tracker encounters uncertainty. This method, which is training-free and treats existing trackers as black boxes, aims to improve identity preservation in challenging frames. When tested on the SportsMOT benchmark, SAM-Deep-EIoU achieved state-of-the-art performance with an 86.8 HOTA score. AI
IMPACT Enhances object tracking accuracy by intelligently leveraging more powerful segmentation models, potentially improving applications in sports analytics and video surveillance.