Image Reflection Removal based on Knowledge-distilling Content Disentanglement, IEEE Signal Processing Letters, 2022

## **Abstract:**

Single image reflection removal (SIRR) is an ill-posed and challenging problem that is practically essential to image enhancement. Inspired by knowledge distillation in deep learning, we tackle the SIRR problem by proposing a knowledge-distilling-based content disentangling model that can effectively decompose the transmission and reflection layers. The experiments on benchmark SIRR datasets show that our method performs favorably against state-of-the-art SIRR methods.

Environment

• Platforms: Windows 10 / cuda8.0
  
• python: 3.7.6 / pytorch: 1.5
  

Inference

• Change the "root path" and "save image path" in main.py.
  
• Move test images to /Test.
  
• Download the pretrained weight from https://drive.google.com/drive/folders/1OINXQIFDbQVfm82UKO12FCr_Mcjr2z0N and place it in /Weighted.
  
• Do the inference by running "python main.py".