Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Benchmarking the Robustness of UAV Tracking Against Common Corruptions

Xiaoqiong Liu, Yunhe Feng, Shu Hu, Xiaohui Yuan, Heng Fan

TL;DR

This paper introduces UAV-C, a large-scale benchmark designed to evaluate UAV tracking robustness under common corruptions. Built on 193 base videos from UAV123-10fps and DBT70, UAV-C applies 18 corruptions across 4 categories (weather, blur, sensor, composite) with 3 severity levels, generating 9843 corrupted videos across 51 severity settings. Twelve trackers spanning CNN-based, CNN-Transformer-based, and Transformer-based architectures are assessed, revealing substantial degradation under corruptions, with composite and Zoom Blur corruptions posing the greatest challenges. The work provides detailed metrics, including $mS_{cor}$ and $rDrop$ variants, and concludes with insights that transformer models and corruption-aware training can bolster UAV-tracking robustness, along with a call to use UAV-C for future development and benchmarking.

Abstract

The robustness of unmanned aerial vehicle (UAV) tracking is crucial in many tasks like surveillance and robotics. Despite its importance, little attention is paid to the performance of UAV trackers under common corruptions due to lack of a dedicated platform. Addressing this, we propose UAV-C, a large-scale benchmark for assessing robustness of UAV trackers under common corruptions. Specifically, UAV-C is built upon two popular UAV datasets by introducing 18 common corruptions from 4 representative categories including adversarial, sensor, blur, and composite corruptions in different levels. Finally, UAV-C contains more than 10K sequences. To understand the robustness of existing UAV trackers against corruptions, we extensively evaluate 12 representative algorithms on UAV-C. Our study reveals several key findings: 1) Current trackers are vulnerable to corruptions, indicating more attention needed in enhancing the robustness of UAV trackers; 2) When accompanying together, composite corruptions result in more severe degradation to trackers; and 3) While each tracker has its unique performance profile, some trackers may be more sensitive to specific corruptions. By releasing UAV-C, we hope it, along with comprehensive analysis, serves as a valuable resource for advancing the robustness of UAV tracking against corruption. Our UAV-C will be available at https://github.com/Xiaoqiong-Liu/UAV-C.

Benchmarking the Robustness of UAV Tracking Against Common Corruptions

TL;DR

This paper introduces UAV-C, a large-scale benchmark designed to evaluate UAV tracking robustness under common corruptions. Built on 193 base videos from UAV123-10fps and DBT70, UAV-C applies 18 corruptions across 4 categories (weather, blur, sensor, composite) with 3 severity levels, generating 9843 corrupted videos across 51 severity settings. Twelve trackers spanning CNN-based, CNN-Transformer-based, and Transformer-based architectures are assessed, revealing substantial degradation under corruptions, with composite and Zoom Blur corruptions posing the greatest challenges. The work provides detailed metrics, including and variants, and concludes with insights that transformer models and corruption-aware training can bolster UAV-tracking robustness, along with a call to use UAV-C for future development and benchmarking.

Abstract

The robustness of unmanned aerial vehicle (UAV) tracking is crucial in many tasks like surveillance and robotics. Despite its importance, little attention is paid to the performance of UAV trackers under common corruptions due to lack of a dedicated platform. Addressing this, we propose UAV-C, a large-scale benchmark for assessing robustness of UAV trackers under common corruptions. Specifically, UAV-C is built upon two popular UAV datasets by introducing 18 common corruptions from 4 representative categories including adversarial, sensor, blur, and composite corruptions in different levels. Finally, UAV-C contains more than 10K sequences. To understand the robustness of existing UAV trackers against corruptions, we extensively evaluate 12 representative algorithms on UAV-C. Our study reveals several key findings: 1) Current trackers are vulnerable to corruptions, indicating more attention needed in enhancing the robustness of UAV trackers; 2) When accompanying together, composite corruptions result in more severe degradation to trackers; and 3) While each tracker has its unique performance profile, some trackers may be more sensitive to specific corruptions. By releasing UAV-C, we hope it, along with comprehensive analysis, serves as a valuable resource for advancing the robustness of UAV tracking against corruption. Our UAV-C will be available at https://github.com/Xiaoqiong-Liu/UAV-C.
Paper Structure (15 sections, 2 equations, 3 figures, 4 tables)

This paper contains 15 sections, 2 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. UAV Tracking Benchmark
  4. Other Corruption-Related Benchmarks
  5. Proposed UAV-C Benchmark
  6. Data Source
  7. Corruption Categories
  8. Construction of UAV-C
  9. Experiments
  10. Benchmark Dataset
  11. Evaluated Trackers
  12. Evaluation Metric
  13. Main Result
  14. Discussion
  15. Conclusion

Figures (3)

  • Figure 1: Degradation of UAV trackers under several corruptions (only partial of corruption is shown due to space limitation). The first row displays clean frames for reference. The second row illustrates the impact of Rain-D corruption, while the third row showcases the effects of Rain-D-G corruption. Best viewed in color.
  • Figure 2: Illustration of the clean frame and its 18 corrupted generated using our method.
  • Figure 3: Mean success rate of different corruptions for each tracker. Please zoom in and view in color.