LVMT: Long-Term Video Segmentation via Truncated Query Propagation
September 28, 2026
LVMT uses a lightweight GRU-based temporal propagation module and a Truncated Query Propagation (TQP) strategy to enable long-term video segmentation. This approach allows the model to process videos in chunks while maintaining object tracking through occlusions without vanishing gradients.
HOW THIS AFFECTS YOU
●
builderYou can implement more robust object tracking in long-duration video feeds.
●
researcherThe TQP strategy solves memory and gradient issues during long-sequence training.