Get Started
Research questionHow can visual trackers maintain localization accuracy while using fresh but noisy target templates at deployment?An initial ground-truth template is accurate but can become stale as the target changes. Templates cropped from recent predictions reflect current appearance but may be mislocalized, making appearance adaptation unreliable.
Computer Vision
Image & Video Processing
Machine Learning
Latest papersRecent research connected to this question, newest first.Learning to Track from Privileged Target AppearancesThe evidence concerns visual trackers trained with exact current and future target crops from frame-level ground truth, then deployed using only the student tracker and past-frame templates. Results cover seven benchmarks under long- and short-term tracking protocols; privileged crops are unavailable during deployment.research paper · Sep 2, 2026
Related questions
How can aerial visual place recognition adapt across missions without catastrophic forgetting?How can detect-to-track pipelines adapt to new visual domains without target-domain labels?How can deployed vision models resist diverse backdoor triggers without model internals or auxiliary data?How can detector-free end-to-end multi-object trackers improve detection without disrupting track association?
Home
Topics
Search
Library