NeRT, an unsupervised and physically grounded deep learning method, can effectively mitigate various types of turbulence, including atmospheric and water turbulence, without relying on domain-specific priors or large training datasets.
The core message of this work is to develop an efficient deep learning-based method for mitigating atmospheric turbulence in images and videos by carefully integrating insights from classical turbulence mitigation algorithms and leveraging a physics-grounded data synthesis approach.