MGPNet -- PyTorch Implementation

This repository contains the PyTorch Implementation of the following paper:

Image Smoothing via Multiscale Global Perception, SPL, 2024

datasets

trainsets

SPS: https://drive.google.com/drive/folders/1inuxV8ghABOv60KVc6zY97Ccj0yyJ9uv?usp=sharing

The approach of generating the trainsets is showed in https://github.com/YidFeng/Easy2Hard

testsets

• SPS testset: [https://drive.google.com/drive/folders/1EDqgjFZt5KndZlHjtD2EjpcXvfzT878s?usp=sharing]
• NKS testset: [https://drive.google.com/drive/folders/1rsWLc7kpyM2VfGwY_Gu94TH2bnt3ywfV?usp=sharing]

Usage

The code is tested with Python 3.7, PyTorch 1.9.0 and CUDA 11.1, and is saved in codes folder.

cd codes

Training

First set a config file train.yml in options/train/, then run as following:

python train.py -opt options/train/train.yml

Test

First set a config file test_smoothing.yml in options/test/, then run as following:

python test.py -opt options/test/test_smoothing.yml

The test result will be saved in ../results folder.

Pretrained model

Pretrained model is released on ../experiments/MGPNet/models/best_G.pth.

The Contents of codes folder

Config: options/ Configure the options for data loader, network structure, model, training strategies and etc.

Data: data/ A data loader to provide data for training, validation and testing.

Model: models/ Construct models for training and testing, models/MGPNet.py construct network architectures.

Citation

@article{he2024image,
  title={Image Smoothing via Multiscale Global Perception},
  author={He, Xuyi and Quan, Yuhui and Xu, Yong and Xu, Ruotao},
  journal={IEEE Signal Processing Letters},
  year={2024},
  publisher={IEEE}
}