Reproduce “Semi-TCL: Semi-Supervised Track Contrastive Representation Learning” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Wei Li, Yuanjun Xiong, Shuo Yang, Mingze Xu, Yongxin Wang, Wei Xia

Semi-TCL trains matching for online multi-object tracking by comparing detections with averaged representations of short tracks rather than individual images. It combines human-labeled tracks with motion-predicted tracklets from unlabeled video as pseudo-labels. On a MOT17 validation split, it achieves 76.2 IDF1 versus 75.5 for instance-level contrastive learning, while MOTA is nearly unchanged; the method still relies on labeled data and imperfect pseudo-labels.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.