Home

Awesome

SRFlow

Official SRFlow training code: Super-Resolution using Normalizing Flow in PyTorch <br><br>

[Paper] ECCV 2020 Spotlight

<br>

News: Unified Image Super-Resolution and Rescaling [code] <br> <br>

SRFlow <br> <br> <br>

Setup: Data, Environment, PyTorch Demo

<br>
git clone https://github.com/andreas128/SRFlow.git && cd SRFlow && ./setup.sh
<br>

This oneliner will:

If you want to install it manually, read the setup.sh file. (Links to data/models, pip packages)

<br> <br>

Demo: Try Normalizing Flow in PyTorch

./run_jupyter.sh

This notebook lets you:

<br><br>

Testing: Apply the included pretrained models

source myenv/bin/activate                      # Use the env you created using setup.sh
cd code
CUDA_VISIBLE_DEVICES=-1 python test.py ./confs/SRFlow_DF2K_4X.yml      # Diverse Images 4X (Dataset Included)
CUDA_VISIBLE_DEVICES=-1 python test.py ./confs/SRFlow_DF2K_8X.yml      # Diverse Images 8X (Dataset Included)
CUDA_VISIBLE_DEVICES=-1 python test.py ./confs/SRFlow_CelebA_8X.yml    # Faces 8X

For testing, we apply SRFlow to the full images on CPU.

<br><br>

Training: Reproduce or train on your Data

The following commands train the Super-Resolution network using Normalizing Flow in PyTorch:

source myenv/bin/activate                      # Use the env you created using setup.sh
cd code
python train.py -opt ./confs/SRFlow_DF2K_4X.yml      # Diverse Images 4X (Dataset Included)
python train.py -opt ./confs/SRFlow_DF2K_8X.yml      # Diverse Images 8X (Dataset Included)
python train.py -opt ./confs/SRFlow_CelebA_8X.yml    # Faces 8X

How to prepare CelebA?

1. Get HD-CelebA-Cropper

git clone https://github.com/LynnHo/HD-CelebA-Cropper

2. Download the dataset

img_celeba.7z and annotations.zip as desribed in the Readme.

3. Run the crop align

python3 align.py --img_dir ./data/data --crop_size_h 640 --crop_size_w 640 --order 3 --face_factor 0.6 --n_worker 8

4. Downsample for GT

Use the matlablike kernel to downscale to 160x160 for the GT images.

5. Downsample for LR

Downscale the GT using the Matlab kernel to the LR size (40x40 or 20x20)

6. Train/Validation

For training and validation, we use the corresponding sets defined by CelebA (Train: 000001-162770, Validation: 162771-182637)

7. Pack to pickle for training

cd code && python prepare_data.py /path/to/img_dir

<br><br>

Dataset: How to train on your own data

The following command creates the pickel files that you can use in the yaml config file:

cd code
python prepare_data.py /path/to/img_dir

The precomputed DF2K dataset gets downloaded using setup.sh. You can reproduce it or prepare your own dataset.

<br><br>

Our paper explains

<br><br>

GAN vs Normalizing Flow - Blog

Get a quick introduction to Normalizing Flow in our [Blog]. <br><br><br>

<br><br>

Wanna help to improve the code?

If you found a bug or improved the code, please do the following:

<br><br>

Paper

[Paper] ECCV 2020 Spotlight

@inproceedings{lugmayr2020srflow,
  title={SRFlow: Learning the Super-Resolution Space with Normalizing Flow},
  author={Lugmayr, Andreas and Danelljan, Martin and Van Gool, Luc and Timofte, Radu},
  booktitle={ECCV},
  year={2020}
}

<br><br>