Learning Dynamic Memory Networks for Object Tracking
lastmod 2026-11-10created 2019-03-122 edits
Learning Dynamic Memory Networks for Object Tracking (MemTrack) is a visual object tracking paper that uses an external memory network to adapt to changes in the target's appearance. Unlike SiamFC-style trackers that use only a fixed template from the first frame, it reads and writes target features to memory through an addressing scheme controlled by an LSTM and updates the template dynamically, handling deformation and occlusion without online model retraining. It was published in 2018, and this page collects it as a reference for motion estimation and single-object tracking research.
Learning Dynamic Memory Networks for Object Tracking
Documentation
- Learning Dynamic Memory Networks for Object Tracking
- https://arxiv.org/abs/1803.07268