Progressive Domain Adaptation for Thermal Infrared Object Tracking

Qiao Li,Kanlun Tan,Qiao Liu,Di Yuan,Xin Li,Yunpeng Liu

2024-09-03

Abstract:Due to the lack of large-scale labeled Thermal InfraRed (TIR) training datasets, most existing TIR trackers are trained directly on RGB datasets. However, tracking methods trained on RGB datasets suffer a significant drop-off in TIR data due to the domain shift issue. To this end, in this work, we propose a Progressive Domain Adaptation framework for TIR Tracking (PDAT), which transfers useful knowledge learned from RGB tracking to TIR tracking. The framework makes full use of large-scale labeled RGB datasets without requiring time-consuming and labor-intensive labeling of large-scale TIR data. Specifically, we first propose an adversarial-based global domain adaptation module to reduce domain gap on the feature level coarsely. Second, we design a clustering-based subdomain adaptation method to further align the feature distributions of the RGB and TIR datasets finely. These two domain adaptation modules gradually eliminate the discrepancy between the two domains, and thus learn domain-invariant fine-grained features through progressive training. Additionally, we collect a largescale TIR dataset with over 1.48 million unlabeled TIR images for training the proposed domain adaptation framework. Experimental results on five TIR tracking benchmarks show that the proposed method gains a nearly 6% success rate, demonstrating its effectiveness.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the domain transfer problem in thermal infrared (TIR) target tracking. Specifically:

1. **Lack of large - scale labeled TIR training datasets**: Since there is no large enough thermal infrared image dataset with labels, most of the existing TIR trackers are directly trained on RGB datasets. However, the performance of these methods trained on RGB datasets will drop significantly on TIR data, which is caused by the domain shift issue.

2. **Performance degradation due to domain differences**: RGB images and TIR images have great differences in style due to different imaging principles, which leads to significant distribution differences between them. Therefore, when directly transferring knowledge from the RGB domain to the TIR domain, the performance of the model will drop significantly.

To solve these problems, the paper proposes a Progressive Domain Adaptation for TIR Tracking (PDAT) framework, which gradually eliminates the domain differences between RGB and TIR data through the following methods:

- **Global Domain Adaptation module**: Adopting an adversarial learning method to roughly narrow the feature distribution gap between the overall RGB and TIR domains.
- **Subdomain Adaptation module**: Based on a clustering method, further align the feature distributions of similar categories to obtain more fine - grained feature transfer.

In addition, the author also collected a large - scale dataset containing more than 1.48 million unlabeled TIR images and trained the proposed domain adaptation framework through pseudo - label generation technology. The experimental results show that this method has achieved an approximately 6% increase in success rate in five TIR tracking benchmark tests, proving its effectiveness.

In summary, the paper aims to improve the performance of TIR trackers by fully utilizing large - scale labeled RGB datasets through domain adaptation techniques, without spending a great deal of time and effort on labeling large - scale TIR datasets.