Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Real-World Atmospheric Turbulence Correction via Domain Adaptation

Xijun Wang, Santiago López-Tapia, Aggelos K. Katsaggelos

TL;DR

Atmospheric turbulence distorts outdoor imagery, and models trained solely on synthetic AT fail on real-world data due to a domain gap. The authors propose Real-ATM, a domain-adaptive teacher–student framework that transfers knowledge from a supervised synthetic AT correction model to an unsupervised real-world AT correction model, using a shared generator and partially shared discriminators with a reproduce network. The generator leverages a Decoupled Dynamic Filter (DDF) conditioned on a latent code $c$ learned by a variational autoencoder, plus a degradation-parameter estimator to encode AT priors, optimized with losses $L^T_{vae}$, $L^S_{vae}$ and $L^T_{degrad}$, and the training objective combines $\mathcal{L}_g$ and $\mathcal{L}_d$. Experiments show Real-ATM improves real-world image quality and enhances downstream person-identification performance, illustrating practical impact for surveillance and outdoor vision tasks.

Abstract

Atmospheric turbulence, a common phenomenon in daily life, is primarily caused by the uneven heating of the Earth's surface. This phenomenon results in distorted and blurred acquired images or videos and can significantly impact downstream vision tasks, particularly those that rely on capturing clear, stable images or videos from outdoor environments, such as accurately detecting or recognizing objects. Therefore, people have proposed ways to simulate atmospheric turbulence and designed effective deep learning-based methods to remove the atmospheric turbulence effect. However, these synthesized turbulent images can not cover all the range of real-world turbulence effects. Though the models have achieved great performance for synthetic scenarios, there always exists a performance drop when applied to real-world cases. Moreover, reducing real-world turbulence is a more challenging task as there are no clean ground truth counterparts provided to the models during training. In this paper, we propose a real-world atmospheric turbulence mitigation model under a domain adaptation framework, which links the supervised simulated atmospheric turbulence correction with the unsupervised real-world atmospheric turbulence correction. We will show our proposed method enhances performance in real-world atmospheric turbulence scenarios, improving both image quality and downstream vision tasks.

Real-World Atmospheric Turbulence Correction via Domain Adaptation

TL;DR

Atmospheric turbulence distorts outdoor imagery, and models trained solely on synthetic AT fail on real-world data due to a domain gap. The authors propose Real-ATM, a domain-adaptive teacher–student framework that transfers knowledge from a supervised synthetic AT correction model to an unsupervised real-world AT correction model, using a shared generator and partially shared discriminators with a reproduce network. The generator leverages a Decoupled Dynamic Filter (DDF) conditioned on a latent code learned by a variational autoencoder, plus a degradation-parameter estimator to encode AT priors, optimized with losses , and , and the training objective combines and . Experiments show Real-ATM improves real-world image quality and enhances downstream person-identification performance, illustrating practical impact for surveillance and outdoor vision tasks.

Abstract

Atmospheric turbulence, a common phenomenon in daily life, is primarily caused by the uneven heating of the Earth's surface. This phenomenon results in distorted and blurred acquired images or videos and can significantly impact downstream vision tasks, particularly those that rely on capturing clear, stable images or videos from outdoor environments, such as accurately detecting or recognizing objects. Therefore, people have proposed ways to simulate atmospheric turbulence and designed effective deep learning-based methods to remove the atmospheric turbulence effect. However, these synthesized turbulent images can not cover all the range of real-world turbulence effects. Though the models have achieved great performance for synthetic scenarios, there always exists a performance drop when applied to real-world cases. Moreover, reducing real-world turbulence is a more challenging task as there are no clean ground truth counterparts provided to the models during training. In this paper, we propose a real-world atmospheric turbulence mitigation model under a domain adaptation framework, which links the supervised simulated atmospheric turbulence correction with the unsupervised real-world atmospheric turbulence correction. We will show our proposed method enhances performance in real-world atmospheric turbulence scenarios, improving both image quality and downstream vision tasks.
Paper Structure (11 sections, 14 equations, 5 figures, 3 tables)

This paper contains 11 sections, 14 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Domain adaptation framework
  4. Generator framework
  5. Training & Testing
  6. Experiments
  7. Experimental settings
  8. Results
  9. Image quality
  10. Downstream person identification task
  11. Conclusion

Figures (5)

  • Figure 1: Domain adaptation framework.
  • Figure 2: Real-world atmospheric turbulence mitigation model (Real-ATM) framework.
  • Figure 3: Generator training framework.
  • Figure 4: Testing framework.
  • Figure 5: Visual samples of real-world AT-input, and the restored results from Syn-ATM and Real-ATM. (PIQE/BRISQUE) is shown in the parenthesis, the best score is marked in bold.