Weakly-Supervised Domain Adaptation of Deep Regression Trackers via Reinforced Knowledge Distillation

TL;DR

This paper introduces a novel weakly-supervised domain adaptation method for deep regression trackers, combining reinforcement learning and knowledge distillation to improve accuracy and efficiency in robotic vision tasks.

Contribution

It presents the first domain adaptation approach for deep regression trackers using weak supervision, reinforcement learning, and knowledge distillation to enhance performance and stability.

Findings

Achieves real-time speed on embedded devices without GPUs.

Significantly improves tracking accuracy across five robotic domains.

Demonstrates robustness to distribution shifts and overfitting.

Abstract

Deep regression trackers are among the fastest tracking algorithms available, and therefore suitable for real-time robotic applications. However, their accuracy is inadequate in many domains due to distribution shift and overfitting. In this paper we overcome such limitations by presenting the first methodology for domain adaption of such a class of trackers. To reduce the labeling effort we propose a weakly-supervised adaptation strategy, in which reinforcement learning is used to express weak supervision as a scalar application-dependent and temporally-delayed feedback. At the same time, knowledge distillation is employed to guarantee learning stability and to compress and transfer knowledge from more powerful but slower trackers. Extensive experiments on five different robotic vision domains demonstrate the relevance of our methodology. Real-time speed is achieved on embedded devices and on machines without GPUs, while accuracy reaches significant results.