Tensorflow implementation for:
Robust Reflection Removal with Reflection-free Flash-only Cues
Chenyang Lei, Qifeng Chen
HKUST

in CVPR 2021

• 2022.03.15 PyTorch version is released

• Release test code
• Prepare paper and upload to arxiv
• Make project page
• Release training code
• Release dataset
• Release raw data processing code

To setup a conda environment, test on demo data:

conda env create -f environment.yml
conda activate flashrr-rfc
bash download.sh
python test.py

This code is based on tensorflow. It has been tested on Ubuntu 18.04 LTS.

Anaconda is recommended: Ubuntu 18.04 | Ubuntu 16.04

After installing Anaconda, you can setup the environment simply by

conda env create -f environment.yml

You can download the ckpt and VGG model by

bash download.sh

You can get the results for the demo data by:

python test.py

If you prepare your own dataset, note that each data sample must contains an ambient image and a flash-only iamge:

python test.py --testset /path/to/your/testset

First, download the dataset:

bash download_data.sh

Then, you can train a model by

python train.py --model YOUR_MODEL_NAME

First, download raw images on OneDrive (70MB for each iamge).

Then,

python rawdata_processing.py

Three rgb images will be saved in ./ dir. You can modify the resolutions by yourself in the code.

The reflection-free cue exploits a flash-only image obtained by subtracting the ambient image from the corresponding flash image in raw data space. The flash-only image is equivalent to an image taken in a dark environment with only a flash on. The reflection disappears in this flash-only image.

Please check our Project Page for detailed explanation.

If you find our work useful for your research, please consider citing the following papers :)

@InProceedings{Lei_2021_RFC,
     title={Robust Reflection Removal with Reflection-free Flash-only Cues}, 
     author={Chenyang Lei and Qifeng Chen},
     booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
     year = {2021}
}

If you are also interested in the polarization reflection removal, please refer to this work.

If you use the synthetic dataset, please cite these two papers since we use their data to synthesize the images:

• Flash dataset
• CoRRN

Please contact me if there is any question (Chenyang Lei, [email protected])

TBD