https://github.com/vita-group/deblurganv2

[ICCV 2019] "DeblurGAN-v2: Deblurring (Orders-of-Magnitude) Faster and Better" by Orest Kupyn, Tetiana Martyniuk, Junru Wu, Zhangyang Wang
https://github.com/vita-group/deblurganv2

deblurgan deep-learning generative-adversarial-network iccv iccv2019 low-level-vision pytorch ukraine

Last synced: 5 months ago
JSON representation

[ICCV 2019] "DeblurGAN-v2: Deblurring (Orders-of-Magnitude) Faster and Better" by Orest Kupyn, Tetiana Martyniuk, Junru Wu, Zhangyang Wang

• Host: GitHub
• URL: https://github.com/vita-group/deblurganv2
• Owner: VITA-Group
• License: other
• Created: 2019-08-10T09:02:40.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2022-07-14T15:36:09.000Z (over 3 years ago)
• Last Synced: 2025-05-16T09:04:15.121Z (5 months ago)
• Topics: deblurgan, deep-learning, generative-adversarial-network, iccv, iccv2019, low-level-vision, pytorch, ukraine
• Language: Python
• Homepage:
• Size: 64.5 MB
• Stars: 1,100
• Watchers: 28
• Forks: 279
• Open Issues: 69
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# DeblurGAN-v2: Deblurring (Orders-of-Magnitude) Faster and Better

Code for this paper [DeblurGAN-v2: Deblurring (Orders-of-Magnitude) Faster and Better](https://arxiv.org/abs/1908.03826)

Orest Kupyn, Tetiana Martyniuk, Junru Wu, Zhangyang Wang

In ICCV 2019

## Overview

We present a new end-to-end generative adversarial network (GAN) for single image motion deblurring, named 

DeblurGAN-v2, which considerably boosts state-of-the-art deblurring efficiency, quality, and flexibility. DeblurGAN-v2 

is based on a relativistic conditional GAN with a double-scale discriminator. For the first time, we introduce the 

Feature Pyramid Network into deblurring, as a core building block in the generator of DeblurGAN-v2. It can flexibly 

work with a wide range of backbones, to navigate the balance between performance and efficiency. The plug-in of 

sophisticated backbones (e.g., Inception-ResNet-v2) can lead to solid state-of-the-art deblurring. Meanwhile, 

with light-weight backbones (e.g., MobileNet and its variants), DeblurGAN-v2 reaches 10-100 times faster than 

the nearest competitors, while maintaining close to state-of-the-art results, implying the option of real-time 

video deblurring. We demonstrate that DeblurGAN-v2 obtains very competitive performance on several popular 

benchmarks, in terms of deblurring quality (both objective and subjective), as well as efficiency. Besides, 

we show the architecture to be effective for general image restoration tasks too.

![](./doc_images/kohler_visual.png)

![](./doc_images/restore_visual.png)

![](./doc_images/gopro_table.png)

![](./doc_images/lai_table.png)

## DeblurGAN-v2 Architecture

![](./doc_images/pipeline.jpg)

 

## Datasets

The datasets for training can be downloaded via the links below:

- [DVD](https://drive.google.com/file/d/1bpj9pCcZR_6-AHb5aNnev5lILQbH8GMZ/view)

- [GoPro](https://drive.google.com/file/d/1KStHiZn5TNm2mo3OLZLjnRvd0vVFCI0W/view)

- [NFS](https://drive.google.com/file/d/1Ut7qbQOrsTZCUJA_mJLptRMipD8sJzjy/view)

## Training

#### Command

```python train.py```

training script will load config under config/config.yaml

#### Tensorboard visualization

![](./doc_images/tensorboard2.png)

## Testing

To test on a single image,

```python predict.py IMAGE_NAME.jpg```

By default, the name of the pretrained model used by Predictor is 'best_fpn.h5'. One can change it in the code ('weights_path' argument). It assumes that the fpn_inception backbone is used. If you want to try it with different backbone pretrain, please specify it also under ['model']['g_name'] in config/config.yaml.

## Pre-trained models

    

        Dataset

        G Model

        D Model

        Loss Type

        PSNR/ SSIM

        Link

    

    

        GoPro Test Dataset

        InceptionResNet-v2

        double_gan

        ragan-ls

        29.55/ 0.934

        fpn_inception.h5

    

    

        MobileNet

        double_gan

        ragan-ls

        28.17/ 0.925

        fpn_mobilenet.h5

    

    

        MobileNet-DSC

        double_gan

        ragan-ls

        28.03/ 0.922

        

    

## Parent Repository

The code was taken from https://github.com/KupynOrest/RestoreGAN . This repository contains flexible pipelines for different Image Restoration tasks.

## Citation

If you use this code for your research, please cite our paper.

```

​```

@InProceedings{Kupyn_2019_ICCV,

author = {Orest Kupyn and Tetiana Martyniuk and Junru Wu and Zhangyang Wang},

title = {DeblurGAN-v2: Deblurring (Orders-of-Magnitude) Faster and Better},

booktitle = {The IEEE International Conference on Computer Vision (ICCV)},

month = {Oct},

year = {2019}

}

​```

```