-
Minimum requirements. This project was originally developed with Python 3.6, PyTorch 1.7 and CUDA 11.1. The training requires at least a Titan X GPUs (12Gb memory each).
-
Download pre-trained models. For running the code, you need to download the two pre-trained weights of Kernel Prior (KP) and Noise Prior (NP) module. Here we use FKP as KP and VDNet as NP. The pretrained weight can be found here
First you need to place the pre-trained weight of KP and NP to anywhere you want.
Next, run the following command:
cd DIPFKP
python main.py --SR --sf 2 --dataset Set5
BEFORE RUN: please make sure that you edit the overwritting paths, DECLARE PATH AT THIS SECTION in the above file.
By default, the results should appear in the following path: ./data/log_DIPFKP/{dataset_name}x{scale}{noise_level}
We thank Jingyun Liang for releasing his code that helped in the early stages of this project.
We also thank Zongsheng Yue for the work at this which contributes to the NP module of our project.
- Python 3.6/3.7.9, PyTorch >= 1.5.1
- Requirements: opencv-python, tqdm
- Platforms: Ubuntu 20.04, cuda-10.2 & cuDNN v-7.5
This project is released under the Apache 2.0 license. The codes are based on normalizing_flows, DIP, KernelGAN and USRNet. Please also follow their licenses. Thanks for their great works.