CSTrack proposes a strong ReID based one-shot MOT framework. It includes a novel cross-correlation network that can effectively impel the separate branches to learn task-dependent representations, and a scale-aware attention network that learns discriminative embeddings to improve the ReID capability. This makes the data association ability of the model comparable to two-stage methods, while running faster.
Download the pre-trained model and save it to cfg/.
Download the MOT-16 dataset archive and extract it to data/MOT16/.
Open a prompt to the repository root and enter the command below:
python scripts/test_cstrack.py