MOTIP
lastmod 2026-11-10created 2024-05-172 edits
MOTIP is a deep learning-based tracking method that solves the multiple object tracking (MOT) problem by recasting it as an object ID prediction problem. For objects found by the detector, it directly predicts which ID each belongs to by referring to past trajectory information, so association is performed by a trained model without a separate heuristic matching step. The code was released by MCG-NJU.
Multiple Object Tracking as ID Prediction