peterzhousz / exposure-fusion-shadow-removal

We propose a new method for effective shadow removal by regarding it as an exposure fusion problem.

Auto-exposure fusion for single-image shadow removal

We propose a new method for effective shadow removal by regarding it as an exposure fusion problem. Please refer to the paper for details: https://arxiv.org/abs/2103.01255

Dataset

• ISTD https://github.com/DeepInsight-PCALab/ST-CGAN
• ISTD+ https://github.com/cvlab-stonybrook/SID
• SRD

Train

Modify the corresponding path in file OE_train.sh and run the following script

sh OE_train.sh

Test

In order to test the performance of a trained model, you need to make sure that the hyper parameters in file OE_eval.sh match the ones in OE_train.sh and run the following script

sh OE_test.sh

The results reported in the paper are calculated by the matlab script used in other SOTA, please see cvlab-stonybrook/SID#1 for details. Our evaluation code will print the metrics calculated by python code and save the result images which will be used by the matlab script.

Results

• Comparsion with SOTA, see paper for details.

• Penumbra comparsion between ours and SP+M Net

• Testing result

The testing results on dataset ISTD+ is: https://drive.google.com/file/d/1tX4wDW3gqQcvN1Qo2QOZb0INdwXLymgS/view?usp=sharing

More details are coming soon

Bibtex

@inproceedings{fu2021auto,
      title={Auto-exposure Fusion for Single-image Shadow Removal}, 
      author={Lan Fu and Changqing Zhou and Qing Guo and Felix Juefei-Xu and Hongkai Yu and Wei Feng and Yang Liu and Song Wang},
      year={2021},
      booktitle={accepted to CVPR}
}