D2HNet

Official PyTorch Implementation of the ECCV 2022 Paper

1 Introduction

This project is a night image restoration framework called D2HNet by jointly denoising and deblurring successively captured long- and short-exposure images. To train and benchmark D2HNet, we create a large-scale D2-Dataset. This repo contains training and evaluation code for the following paper:

D2HNet: Joint Denoising and Deblurring with Hierarchical Network for Robust Night Image Restoration
Yuzhi Zhao¹, Yongzhe Xu², Qiong Yan², Dingdong Yang², Xuehui Wang³, Lai-Man Po¹
¹City University of Hong Kong, ²SenseTime Research and Tetras.AI, ³Shanghai Jiao Tong University
European Conference on Computer Vision (ECCV) 2022

2 Dependency

This code is based on PyTorch 1.1 and CUDA 9.0. It has been tested on Ubuntu 18.04 LTS, where the machine is equipped with NVIDIA Titan GPUs.

We use Anaconda to set up the environment. Users can set up a new environment simply by:

conda env create -f environment.yml
conda activate d2hnet

Then, users need to install the deformable convolution module in the d2hnet environment by:

cd dcn
python setup.py develop

If you got an error when installing deformable convolution module, please delete the build folder first and then run:

cd dcn
rm -rf build
python setup.py build develop

3 D2-Dataset

3.1 Download links

Users can find the full D2-Dataset through the link below. Just unzip all zip files after downloading. Note that the full dataset is quite large (about 555Gb).

link：https://pan.baidu.com/s/1FunrgUvjdOyhz0hgBuW3JQ?pwd=3n6m code：3n6m

Name	Description	Resolution	Numbers	Size
original/train	synthetic `training` tuples	2560x1440	5661	192GB
original/train_blur_patch	selected patch tuples by VarmapSelection from `train`	1024x1024	9453	99.0GB
original/val	synthetic `validation` tuples from videos 1-13	2560x1440	268	9.44GB
original/val_no_overlap	synthetic `validation` tuples from videos 14-30	2560x1440	1192	38.7GB
original/val_no_overlap_blur_patch	selected patch tuples by VarmapSelection from `val_no_overlap`	1024x1024	99	1.11GB
sharpened/train	MATLAB sharpened `train` tuples	2560x1440	5661	124GB
sharpened/train_blur_patch	MATLAB sharpened `train_blur_patch` tuples	1024x1024	9453	59.0GB
sharpened/val	MATLAB sharpened `val` tuples	2560x1440	268	6.08GB
sharpened/val_no_overlap	MATLAB sharpened `val_no_overlap` tuples	2560x1440	1192	23.2GB
sharpened/val_no_overlap_blur_patch	MATLAB sharpened `val_no_overlap_blur_patch` tuples	1024x1024	99	659MB
Xiaomi_Mi_Note_10_photos	`real` captured long- and short-exposure photos	4624x3472	28	2.45GB

Please unzip it in this path. Then, rename the full path to data.

3.2 Explanation

The explanation of the data is shown in the above image. The full resolution above image can be found in the arXiv file. The naming method of D2-Dataset data can be summarized as:

Symbol in the above image	Name (keyword) of image files	Description
l	long8	the long-exposure image of approximately 80ms exposure time
l	long6	the long-exposure image of approximately 60ms exposure time
l	long4	the long-exposure image of approximately 40ms exposure time
l	long2	the long-exposure image of approximately 20ms exposure time
l_last	long_last	the last frame of the long-exposure images
v_quarter	quarter	the frame at the quartile position at vacancy (approximately 35ms), more approching to l_last
v_half	half	the frame at the center position at vacancy (approximately 35ms)
v_3quarter	3quarter	the frame at the quartile position at vacancy (approximately 35ms), more approching to s_first
s_first	short	the first frame of the short-exposure image
s	short	the short-exposure image; in this work we directly use s_first as s due to the very short exposure time

3.3 Corresponding videos

The list of all full-length original videos can be found in original_video_list.txt, and youtube-dl can be used to batch download them. Note that only part of those videos are used to create the D2-Dataset.

4 Train

4.1 Run

D2HNet has two subnets (DeblurNet and EnhanceNet) which are trained sequentially. But they share the same entering file:

python train.py

Users need to change the parameters of train.py to train a specific network, as follows:

Name	Description	Training DeblurNet	Training EnhanceNet
--opt	network to be trained	'options/tp_deblurnet_v2_002.yaml'	'options/tp_denoisenet_v2_002.yaml'
--num_gpus	numbers of GPUs to be used	2	2
--save_path	saving path to trained models	'snapshot/tp_deblurnet_v2_002'	'snapshot/tp_denoisenet_v2_002'
--log_path	saving path to tensorboard results	'log_pt/tp_deblurnet_v2_002'	'log_pt/tp_denoisenet_v2_002'

The EnhanceNet in the paper corresponds to the DenoiseNet in the code.

4.2 Hyper-parameters of the network

Users may change some parameters in the yaml files to fit their machine and requirement:

4.2.1 Training parameters:

epochs: overall training epochs
train_batch_size: training batch size, where one batch denotes that selecting one training image
num_workers: the number of workers for training dataloader
start_idx: if the users continue the training process, please enter the previous number of trained epochs. Also, remember to change finetune_path
finetune_path: the path to load pre-trained weights

4.2.2 Loss and Optimizer parameters:

VGGLoss / vgg_model_path: if users want to add the perceptual loss, please download VGG-16 officially pre-trained model: https://download.pytorch.org/models/vgg16-397923af.pth. Then, put it under pretrained_models folder. Otherwise, please comment out the VGGLoss in the yaml files
lr_g: learning rate of the network
lr_decrease_epoch / lr_decrease_factor: after every lr_decrease_epoch, the learning rate multiplies with lr_decrease_factor

4.2.3 Dataset parameters:

train_path: path to original synthetic training set (original/train)
val_path: path to original synthetic validation set (original/val_no_overlap)
train_sharp_path: path to sharpened training set (sharpened/train)
val_sharp_path: path to sharpened validation set (sharpened/val_no_overlap)
deblur_crop_size: size of the cropped square from a full-resolution image for DeblurNet
deblur_patch_per_image: the numbers of cropped patches from a full-resolution image for DeblurNet
deblur_size: the processing resolution of DeblurNet at training
denoise_crop_size: the processing resolution of EnhanceNet at training, cropped from DeblurNet results
denoise_patch_per_image: the numbers of cropped patches for EnhanceNet
cutnoise: CutNoise-related hyper-parameters
illum_adjust: Illumination Adjustment (IA)-related hyper-parameters
color_adjust: Color Adjustment (CA)-related hyper-parameters

4.3 D2HNet pre-trained models

Users can download pre-trained models via the link below:

link：https://pan.baidu.com/s/1bLyM1aKbFBVvYwmEJyBv2w?pwd=t7wq code：t7wq

Please unzip it in this path, you will get a folder named snapshot.

5 Validation and Testing

5.1 Validation

Noisy long- and short-exposure validation pairs with different resolutions (e.g., 1440x2560 or 2880x5120) are pre-generated. Users may generate original/val_no_overlap_noisy_1440p or original/val_no_overlap_noisy_2880p by adding noises to original/val_no_overlap for running the script. Please change --val_path and --val_sharp_path to the paths to input (e.g., original/val_no_overlap_noisy_1440p) and ground truth (sharpened/val_no_overlap), respectively. Users can test the provided models using the following script:

python validation.py

5.2 Testing (on Real-world Photos)

Noisy long- and short-exposure real-world pairs are provided. Please change --src_path to the data path (Xiaomi_Mi_Note_10_photos). Users can test the provided models using the following script:

python realworld.py

Resulting images will be automatically saved in the results_real_photo.

6 Image capturing tool (Andriod apk)

Users may install the image capturing tool to collect paired long- and short-exposure images on their own smartphones. But an Andriod operating system is required.

6.1 Andriod tool installation

Simply install adb tool on the Ubuntu:

sudo apt-get install adb

Check whether the adb tool is successfully installed:

adb version

6.2 Apk installation

6.2.1 Preparation

Firstly, make sure that your smartphone is connected to the computer (e.g., by USB). Users may plug and unplug the USB cable and find which device is your smartphone:

lsusb
(plug your smartphone)
lsusb

For instance, you will find your smartphone corresponds to the following information:

6.2.2 Configuration

Create the ini file and write the id into it:

echo 0xfirst_part_of_your_smartphone_name > ~/.andriod/adb_usb.ini

For instance:

echo 0x1f3a > ~/.andriod/adb_usb.ini

Then, add a configuration file on your computer:

nano /etc/udev/rules.d/51-andriod.rules

Enter the following information:

SUBSYSTEM=="usb", ATTRS{idVendor}=="first_part_of_your_smartphone_name" ATTRS{idProduct}=="second_part_of_your_smartphone_name", MODE="0666"

For instance:

SUBSYSTEM=="usb", ATTRS{idVendor}=="1f3a" ATTRS{idProduct}=="6001", MODE="0666"

Change the access permissions to the file:

sudo chmod 777 /etc/udev/rules.d/51-andriod.rules

Next, restart the USB service:

sudo service udev restart

Finally, restart the adb service and check the device:

adb kill-server
sudp adb start-server
adb devices

6.2.3 Installation

Users may download the provided apk via the link below:

link：https://pan.baidu.com/s/17exe6zbYzJvu-adDyjRznw?pwd=qhkl code：qhkl

Then, install the apk to your smartphone by running:

adb install path_to_apk

7 Citation

If you find this work useful for your research, please cite:

@inproceedings{zhao2022d2hnet,
  title={D2hnet: Joint denoising and deblurring with hierarchical network for robust night image restoration},
  author={Zhao, Yuzhi and Xu, Yongzhe and Yan, Qiong and Yang, Dingdong and Wang, Xuehui and Po, Lai-Man},
  booktitle={17th European Conference on Computer Vision (ECCV)},
  pages={91--110},
  year={2022},
  organization={Springer}
}

8 Acknowledgement

This work has gone through ICCV, AAAI, and ECCV. We appreciate the works and suggestions from all reviewers and area chairs. Part of the code is based on MMEditing and LSFNet. Thanks for their great works!

We thank Chao Wang in the SenseTime Research and Tetras.AI for helping capture the real-world photos. We thank Weile Liang in the SenseTime Research and Tetras.AI for developing the Andriod image capturing tool. We thank Dr. Qinbin Li and researchers in the AI imaging group of the SenseTime Research and Tetras.AI for early reviewing the paper.

Please contact yzzhao2-c@my.cityu.edu.hk for further questions.

zhaoyuzhi/D2HNet