Learning Policies for Adaptive Tracking with Deep Feature Cascades
lastmod 2026-11-10created 2019-03-123 edits
Learning Policies for Adaptive Tracking with Deep Feature Cascades (EAST) is a paper that learns a policy for object tracking so that each frame computes only as many deep features as it needs. Using reinforcement learning, it decides to process easy frames quickly with shallow features and to use deeper layers only on hard frames, increasing speed while maintaining accuracy. It belongs to the field of visual object tracking.
Learning Policies for Adaptive Tracking with Deep Feature Cascades
Documentation
- Learning Policies for Adaptive Tracking with Deep Feature Cascades
- https://arxiv.org/abs/1708.02973