/Kernel-Modeling-Super-Resolution

Official Implementation for Kernel Modeling Super-Resolution on Real Low-Resolution Images

Primary LanguageMATLAB

Kernel Modeling Super-Resolution on Real Low-Resolution Images

Project Page | Paper | Supplementary Material | Psychovisual Experiment

by Ruofan Zhou, Sabine Süsstrunk

Dependencies

  • Pytorch >= 0.4.0
  • OpenCV
  • NVIDIA GPU
  • HDF5 (only for training)
  • MATLAB （only for training)

Quick start (Demo)

In test_code folder, run the following command:

python demo.py

Training the network yourself

Step 1: prepare the dataset

Download DEPD dataset, prepare the patches and run training_code/kernel_estimation/getkernels.m in MATLAB.

Step 2: train a GAN on kernels

run training_code/kernel_generator/train.py.

Step 3: generate

run training_code/kernel_generator/generate.py.

Step 4: train the super-resolution network

run training_code/super-resolution/main.py.

Citations

@inproceedings{zhou2019kernel,
  title={Kernel Modeling Super-Resolution on Real Low-Resolution Images},
  author={Zhou, Ruofan and S{\"u}sstrunk, Sabine},
  booktitle={2019 International Conference on Computer Vision},
  year={2019}
}