/DPSRGAN

Primary LanguagePython

DPSRGAN: Dilation Patch Super-Resolution Generative Adversarial Networks

A PyTorch implementation of DPSRGAN: Dilation Patch Super-Resolution Generative Adversarial Networks

Usage

Installation

git clone https://github.com/kushalchordiya216/Super-Resolution.git
cd Super-Resolution
pip3 install -r requirements.txt

Training

To pretrain the generator before GAN training:

python3 train.py --data_dir <path to HR images> --network SRResNet

For GAN training:

python3 train.py --data_dir <path to HR images> --network SRGAN --pretrain_gen <path to pretrained generator model file>

To view more argument descriptions:

python3 train.py --help

Testing

This will save the predicted images in the directory ./preds/

python3 --model_path <path to pretrained generator> --data_dir <path to directory containing LR images>

Architecture

Generator

Generator Architecture

Discriminator

Discriminator Architecture

Training schematic

Training schematic

Comparison and Results

Picture grid

Citation

@INPROCEEDINGS{9417903,
    author={Mirchandani, Kapil and Chordiya, Kushal},
    booktitle={2021 6th International Conference for Convergence in Technology (I2CT)},
    title={DPSRGAN: Dilation Patch Super-Resolution Generative Adversarial Networks},
    year={2021},  
    volume={},  
    number={},  
    pages={1-7},  
    doi={10.1109/I2CT51068.2021.9417903}}