AdaptiveNet for general optical image restoration

Due to non-ideal imaging conditions such as scattering, defocusing, diffraction, and so on, the captured images may suffer from degradation and be hard to understand. Although deep learning has proven to be powerful for these image restoration tasks, it still has a deficiency in generalization when faced with various degradation models together. In this work, we propose a novel architecture combined with an adaptive encoder to realize general optical image restoration, which we term the "AdaptiveNet." It can adaptively classify the experimental conditions and learn to recover images exclusively. Extensive experiments demonstrate that AdaptiveNet achieves superior recovery results among various optical image restoration tasks.