Hyperparameter Optimization for Tracking With Continuous Deep Q-Learning

Xingping Dong, Jianbing Shen, Wenguan Wang, Yu Liu, Ling Shao, Fatih Porikli; The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018, pp. 518-527


Hyperparameters are numerical presets whose values are assigned prior to the commencement of the learning process. Selecting appropriate hyperparameters is critical for the accuracy of tracking algorithms, yet it is difficult to determine their optimal values, in particular, adaptive ones for each specific video sequence. Most hyperparameter optimization algorithms depend on searching a generic range and they are imposed blindly on all sequences. Here, we propose a novel hyperparameter optimization method that can find optimal hyperparameters for a given sequence using an action-prediction network leveraged on Continuous Deep Q-Learning. Since the common state-spaces for object tracking tasks are significantly more complex than the ones in traditional control problems, existing Continuous Deep Q-Learning algorithms cannot be directly applied. To overcome this challenge, we introduce an efficient heuristic to accelerate the convergence behavior. We evaluate our method on several tracking benchmarks and demonstrate its superior performance.

Related Material

author = {Dong, Xingping and Shen, Jianbing and Wang, Wenguan and Liu, Yu and Shao, Ling and Porikli, Fatih},
title = {Hyperparameter Optimization for Tracking With Continuous Deep Q-Learning},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2018}