Classification-prior-to-Restoration

Weather image noise (rain, snow, haze) removal with classification

Classification: Mobilenet Denoising: U-Net

Overall Architecture

Classifier: Edge and Luminance metric are used for better classification.

• Notice: Rainy and snowy days tended to have low luminance and high edge components

Restoration: Pyramidal U-Net

• U-Net has proven its power in denoising. Base on U-Net architecture, we adopted pyramid structure to enhance receptive field acceptance.

For more detail, please check "Weather Classification and Image Restoration Algorithm Attentive to Weather Conditions in Autonomous Vehicle.pdf"

Our demo is on youtube.

Dehaze: https://youtu.be/sFpYvlNCd0s

Desnow: https://youtu.be/tdBYfQB3czU

Derain: https://youtu.be/knX-aSEZKbE