We propose to perform Image Super-Resolution via Variational AutoEncoders (SR-VAE) learning according to the conditional distribution of the high-resolution images induced by the low-resolution images.. We claim the following points:

• Conditional sampling mechanism.

• Back projection based SR-VAE network.

• Modified VGG feature based loss estimation.

Please cite our work if you use our code or dataset as,

    @InProceedings{Liu2021refvae,
        author = {Zhi-Song Liu, Wan-Chi Siu and Yui-Lam Chan},
        title = {Photo-Realistic Image Super-Resolution via Variational Autoencoders},
        booktitle = {IEEE Transactions on Circuits and Systems for Video Technology (TCSVT)},
        volume = {31},
        no = {4},
        page = {1351-1365},
        month = {April},
        year = {2021}
    }

Python 2.XXX<3.0
OpenCV 3.4.0
Caffe 
NVIDIA GPU + CUDA
Jupyter Notebook

The complete architecture is shown as follows,

1. Download pre-trained model from the following link

https://drive.google.com/drive/folders/1XIsjovqYszI9RvTa0RbyHQcEraZGTpfo?usp=sharing

2. Testing on 4x SR

run SR_VAE_4x.ipynb

3. Testing on 8x SR

run SR_VAE_8x.ipynb

1. We compared our proposed approach with state-of-the-arts face image SR approaches on objective quality by using PSNR, SSIM and PI scores as follow

2. We also compared different approaches on 4x SR.

3. We also compared different approaches on 8x SR.

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 Reference based General image SR using VAE