ReaMOT introduces a reasoning-focused multi-object tracking task, a large benchmark dataset with six scenarios, and a training-free ReaTrack framework that pairs LVLM semantic detection with SAM2 motion priors, reporting over 3x RHOTA gains on high-level reasoning cases.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
ReaMOT: A Benchmark and Framework for Reasoning-based Multi-Object Tracking
ReaMOT introduces a reasoning-focused multi-object tracking task, a large benchmark dataset with six scenarios, and a training-free ReaTrack framework that pairs LVLM semantic detection with SAM2 motion priors, reporting over 3x RHOTA gains on high-level reasoning cases.