Home

Awesome

PWC PWC PWC PWC PWC PWC PWC PWC

HINet: Half Instance Normalization Network for Image Restoration

Liangyu Chen, Xin Lu, Jie Zhang, Xiaojie Chu, Chengpeng Chen

Paper: https://arxiv.org/abs/2105.06086

In this paper, we explore the role of Instance Normalization in low-level vision tasks. Specifically, we present a novel block: Half Instance Normalization Block (HIN Block), to boost the performance of image restoration networks. Based on HIN Block, we design a simple and powerful multi-stage network named HINet, which consists of two subnetworks. With the help of HIN Block, HINet surpasses the state-of-the-art (SOTA) on various image restoration tasks. For image denoising, we exceed it 0.11dB and 0.28 dB in PSNR on SIDD dataset, with only 7.5% and 30% of its multiplier-accumulator operations (MACs), 6.8 times and 2.9 times speedup respectively. For image deblurring, we get comparable performance with 22.5% of its MACs and 3.3 times speedup on REDS and GoPro datasets. For image deraining, we exceed it by 0.3 dB in PSNR on the average result of multiple datasets with 1.4 times speedup. With HINet, we won 1st place on the NTIRE 2021 Image Deblurring Challenge - Track2. JPEG Artifacts, with a PSNR of 29.70.

Network Architecture

<img src="figures/pipeline.png" alt="arch" style="zoom:100%;" />

News

2022.04.12 Our new work, Simple Baselines for Image Restoration reveals the nonlinear activation functions, e.g. ReLU, GELU, Sigmoid, and etc. are not necessary to achieve SOTA performance. The paper provide a simple baseline, NAFNet: Nonlinear Activation Free Network for Image Restoration tasks, and acheves SOTA performance on Image Denoising and Image Deblurring. The paper and the code are available at https://arxiv.org/abs/2204.04676 / https://github.com/megvii-research/NAFNet respectively.

2021.12.10 Our new work, Revisiting Global Statistics Aggregation for Improving Image Restoration, exceeds the previous SOTA restorers 0.6 dB (GoPro dataset) without re-train the model. It is accomplished by revealing the feature distribution shifts issue from training phase to testing phase. The paper and the code are available at https://arxiv.org/abs/2112.04491 / https://github.com/megvii-research/tlsc respectively.

Installation

This implementation based on BasicSR which is a open source toolbox for image/video restoration tasks.

python 3.6.9
pytorch 1.5.1
cuda 10.1

git clone https://github.com/megvii-model/HINet
cd HINet
pip install -r requirements.txt
python setup.py develop --no_cuda_ext

Quick Start (Single Image Inference)

Image Restoration Tasks

Image denoise, deblur, derain.

<details><summary>Image Denoise - SIDD dataset (Click to expand) </summary> </details> <details> <summary>Image Deblur - GoPro dataset (Click to expand) </summary> </details> <details><summary> Image Deblur - REDS dataset (Click to expand) </summary> </details> <details> <summary> Image Derain - Rain13k dataset (Click to expand) </summary> </details>

Results

Some of the following results are higher than the original paper as we optimized some hyper-parameters.

<div align="center"> <img src="./figures/NTIRE2021%20Deblur%20Track2%20Result.jpg" height="400px" alt="NTIRE2021 Deblur Track2 Result"><img src="./figures/SIDD%20Result.jpg" height="400px" alt="SIDD Result"><img src="./figures/GoPro%20Result.jpg" height="400px" alt="GoPro Result" > </div> <div align="center"> <img src="./figures/REDS%20Result.jpg" height="250px" alt="REDDS Result"><img src="./figures/Rain13k%20Result.jpg" height="300px" alt="Rain13k Result"> </div>

License

This project is under the MIT license, and it is based on BasicSR which is under the Apache 2.0 license.

Citations

If HINet helps your research or work, please consider citing HINet.

@InProceedings{Chen_2021_CVPR,
    author    = {Chen, Liangyu and Lu, Xin and Zhang, Jie and Chu, Xiaojie and Chen, Chengpeng},
    title     = {HINet: Half Instance Normalization Network for Image Restoration},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    month     = {June},
    year      = {2021},
    pages     = {182-192}
}

Contact

If you have any questions, please contact chenliangyu@megvii.com or luxin@megvii.com .