Skip to content

Latest commit

 

History

History
69 lines (50 loc) · 2.78 KB

README.md

File metadata and controls

69 lines (50 loc) · 2.78 KB
Raw

RZSR: Reference-based Zero-Shot Super-Resolution with Depth Guided Self-Exemplars

Arxiv | IEEE Transactions on Multimedia

Dependencies

  • Python=3.7
  • PyTorch
  • opencv-python
  • scikit-image(skimage)

Code

Clone this repository into any place you want.

    git clone https://github.com/junsang7777/RZSR.git 
    cd RZSR

Demo

You can test our SR algorithm with your images. Place your image in "set" folder. (img - RGB, dep - Depth, ker - Kernel)

If you don't have kernel files, Change "kernels = None" in 'main.py' script

You can get the depth & kernel of the images from the repository as follows: Depth : monodepth2 & Adabins Kernel : KernelGAN

    python main.py

Set dir : img (Random gaussian blurred img) , dep (Adabins depth estimation result) , ker ( KernelGAN estimation result)

hyper-parmeter : hitogram_threshold & Number of BINs & configs.py etc... table5

Framework

framework

Historical Comaparision

NIMA

Related Work

“Zero-Shot” Super-Resolution using Deep Internal Learning | git

Robust Reference-based Super-Resolution with Similarity-Aware Deformable Convolution

Image Super-Resolution with Cross-Scale Non-Local Attention and Exhaustive Self-Exemplars Mining | git

Acknowledgment

The code is based on pytorch-ZSSR

Citation (Early access)

@ARTICLE{9868165,
    author={Yoo, Jun-Sang and Kim, Dong-Wook and Lu, Yucheng and Jung, Seung-Won},
    journal={IEEE Transactions on Multimedia}, 
    title={RZSR: Reference-based Zero-Shot Super-Resolution with Depth Guided Self-Exemplars}, 
    year={2022},
    volume={},
    number={},
    pages={1-13},
    doi={10.1109/TMM.2022.3202018}}