DeepCLEAR: Intelligent Atmospheric Turbulence Removal and Object Recognition

• RMFAT: Recurrent Multi-scale Feature Atmospheric Turbulence Mitigator (AAAI2026)

  [PDF] [Project] [Code]

  

  RMFAT, a Recurrent Multi-scale Feature Atmospheric Turbulence Mitigator, restores videos efficiently and consistently by using a lightweight two-input recurrent framework with multi-scale feature encoding and temporal warping to enhance spatial detail and temporal coherence.

• MAMAT: 3D Mamba-Based Atmospheric Turbulence Removal and its Object Detection Capability (AVSS2025) [PDF] [Code]

  

  MAMAT is a novel Mamba-based method in which the first module employs deformable 3D convolutions for non-rigid registration to reduce spatial shifts, while the second module enhances contrast and detail. Leveraging the advanced capabilities of the 3D Mamba architecture, experimental results demonstrate that MAMAT outperforms state-of-the-art learning-based methods.

• JDATT: A Joint Distillation Framework for Atmospheric Turbulence Mitigation and Target Detection. (BMVC2025) [PDF]

  

  JDATT is a knowledge distillation framework designed to reduce model size and improve inference speed. We introduce a joint end-to-end training strategy that preserves image quality through reconstruction loss, Channel-Wise Distillation loss, and Masked Generative Distillation loss, while maintaining detection performance via detection loss and Kullback–Leibler divergence..

• DMAT: An End-to-End Framework for Joint Atmospheric Turbulence Mitigation and Object Detection. (WACV2026)

  [PDF] [Code] [Datasets]

  

  DMAT is an end-to-end framework that jointly improves visual quality and object detection by compensating for distorted features. It enables knowledge exchange between low-level distortion correction in the atmospheric turbulence mitigator and high-level semantic features from the detector. The AT mitigator employs a 3D Mamba-based architecture to model spatio-temporal turbulence effects, and the entire system is optimised jointly.

• DeTurb: Atmospheric Turbulence Mitigation with Deformable 3D Convolutions and 3D Swin Transformers (ACCV2024)

  [PDF] [Code]

  

  The DeTurb framework combines geometric restoration with an enhancement module. Random perturbations and geometric distortions are corrected using a pyramid architecture with deformable 3D convolutions, producing aligned frames. These frames are then reconstructed into a sharp, clear image through a multi-scale 3D Swin Transformer architecture.

Research team

Core

• N. Anantrasirichai: Lead academic
• D.R. Bull and A. Achim: Co-Lead academics
• Paul Hill: Postdoctoral researcher [Review paper] [Paper1] [Paper2] [Real dataset] [Synthetic datasets]
• Zhiming Liu: Research associate [Paper1] [Paper2]

Undergrad/Postgrad projects

• Zhiming Liu (2025), Joint Distillation Framework for Turbulence Mitigation and Target Detection [Paper]
• Zhicheng (Frederick) Zou (2024), Enhancing Long-Range Imaging through Deep Learning: Mitigating Atmospheric Turbulence [Paper] [Code]
• Disen Hu (2022), Atmospheric Turbulence Object Detection in Video Using Deep Learning [Paper]
• Rachel Lin (2022), Dealing with Atmospheric Turbulence Distortion in Video Sequences
• Haziq I.B.Mohammed Shafri (2020), Multi-Input Denoising for Atmospheric Turbulence Mitigation in Video
• Jingxuan Wang (2020), Atmospheric turbulence mitigation in video using deep learning
• Jing Gao (2019), Atmospheric turbulence removal using convolutional neural network [arXiv]