RefVAE

Reference based Image Super-Resolution via Variational AutoEncoder

By Zhi-Song Liu, Li-Wen Wang and Wan-Chi Siu

This repo only provides simple testing codes, pretrained models and the network strategy demo.

We propose a Reference based Image Super-Resolution via Variational AutoEncoder (RefVAE)

We participate CVPRW Learning the Super-Resolution Space

Please check our paper

BibTex

    @InProceedings{Liu2021refvae,
        author = {Zhi-Song Liu, Wan-Chi Siu and Li-Wen Wang},
        title = {Reference based Image Super-Resolution via Variational AutoEncoder},
        booktitle = {IEEE International Conference on Computer Vision and Pattern Recognition Workshop(CVPRW)},
        month = {June},
        year = {2021}
    }

For proposed RefVAE model, we claim the following points:

• First working on using Variational AutoEncoder for reference based image super-resolution.

• Our proposed RefVAE can expand the SR space so that multiple SR images can be generated.

Dependencies

Python > 3.0
OpenCV library
Pytorch > 1.0
NVIDIA GPU + CUDA

Complete Architecture

The complete architecture is shown as follows,

Implementation

1. Quick testing

---

1. Download pre-trained from https://drive.google.com/file/d/1R3vR7PiFNT26sIBorVoq6Mf-F4pMHfmh/view?usp=sharing

then put the pre-trained models under the "models" folder.

2. Modify "test.py" and run

$ python test.py

2. Training

---

s1. Download DIV2K and Flickr2K training images from

https://data.vision.ee.ethz.ch/cvl/DIV2K/

https://github.com/LimBee/NTIRE2017

s2. Download reference images from

https://www.wikiart.org/

s3. Modify "test.py" and run

$ python main_GAN.py

---

Partial SR image comparison

1. Visualization comparison

Results on 8x image SR on DIV2K validation dataset

2. Quantitative comparison

Reference

You may check our newly work on Real image super-resolution using VAE

You may also check our work on Reference based face SR using VAE

You may also check our work on General image SR using VAE