Meta-Tracker: Fast and Robust Online Adaptation for Visual Object Trackers

TL;DR

This paper introduces Meta-Tracker, a meta-learning approach that enhances online adaptation in visual object trackers, resulting in faster, more accurate, and robust tracking performance.

Contribution

It presents a novel offline meta-learning method to initialize deep networks for online tracking, enabling quick adaptation and improved robustness against background clutter and noise.

Findings

Meta-Tracker improves speed and accuracy of existing trackers

Meta-learning reduces training time for online adaptation

Enhanced robustness against challenging tracking scenarios

Abstract

This paper improves state-of-the-art visual object trackers that use online adaptation. Our core contribution is an offline meta-learning-based method to adjust the initial deep networks used in online adaptation-based tracking. The meta learning is driven by the goal of deep networks that can quickly be adapted to robustly model a particular target in future frames. Ideally the resulting models focus on features that are useful for future frames, and avoid overfitting to background clutter, small parts of the target, or noise. By enforcing a small number of update iterations during meta-learning, the resulting networks train significantly faster. We demonstrate this approach on top of the high performance tracking approaches: tracking-by-detection based MDNet and the correlation based CREST. Experimental results on standard benchmarks, OTB2015 and VOT2016, show that our meta-learned versions of both trackers improve speed, accuracy, and robustness.