Deep Network Flow for Multi-Object Tracking

lastmod 2026-11-10created 2019-03-123 edits

Deep Network Flow for Multi-Object Tracking is a CVPR 2017 paper that solves multi-object tracking as a network flow optimization problem while learning the cost functions of the graph edges with deep learning. Instead of the hand-designed costs of earlier network flow-based tracking, the costs are learned end to end from data, improving how well detections are linked into trajectories. This page archives the original paper.

Documentation

Deep Network Flow for Multi-Object Tracking

http://www.nec-labs.com/uploads/images/Department-Images/MediaAnalytics/papers/cvpr17_deepnetworkflow.pdf

See also

• Motion estimation