Environment

Pillow==9.5.0 python-version==0.0.2 scikit-image==0.19.3 tensorflow==2.11.0

Please run the following line to install enviroment

pip install -r requirements.txt

How to try

dataset

We use the synthetic method from [Zheng et. al] (https://github.com/q-zh/absorption)

place 365

RID

SIR2

Nature

Zhang et. al

ERRNET

Data_root/
         -train/
               -syn
               -t
               -r
               -estimate.txt
               ⋮
         -test/
               -syn
               -r
               -t
               ⋮
 ⋮

Training

Trainging Teacher Net -python TSmain.py

Training Student Net  -python student.py

reflection refraction estimate -python estimation.py

Run testing

Test Teacher Net -python TS_teacher_test.py

Test Student Net -python TS_student_test.py

Calculating model performance  -python compute.py

The optimizer selected is Adam, and the initial learning rate is 1e-4. The program execution environment is Tensorflow 2.11.0, python 3.7.16, and the graphics cards used are NVIDIA GeForce GTX 1080 Ti and NVIDIA GeForce RTX 3090. Input video size: 256×256

Citation

If you use this code for your research, please cite our paper :

@inproceedings{chan2025single,
  title={ Single-Image Reflection Removal via Knowledge
 Distillation Network with Deep Feature Extraction
 and Reflection-Refraction Estimation},
  author={Chan, U-In and Liu, Tsung-Jung},
  year={2024}
}