/BILD

Primary LanguagePython

[Boosting Supervised Dehazing Methods via Bi-level Patch Reweighting] (ECCV 2022)

Official implementation.

Dependencies and Installation

  • matplotlib==3.4.2
  • numpy==1.19.2
  • Pillow==8.4.0
  • scikit_image==0.17.2
  • skimage==0.0
  • torch==1.7.1
  • torchvision==0.8.2

Datasets Preparation

synthetic dataset -- RESIDE: ITS, OTS, SOTS Dataset website:RESIDE

real-world dataset -- O-HAZE and NH-HAZE Dataset website:O-HAZE NH-HAZE

Performance

Usage

Test

Here, we adopt FFANet and BILD for example and you can replace your own model with code/FFA/. Trained_models are available at trained_models/

Main function is at train.py. The train() function is for model training and the test() function is for model testing. Before training or testing, make sure the file path corresponds to yours.

Performance on out_of_distribution dataset