Awesome
Usage Instruction For ReBlurSR Dataset
Dataset Using
Generate ReBlurSR-Train and ReBlurSR-Test
- Download the dataset files from the Google Drive
- Unzip the files to the corresponding folders. The structure tree of the unzipped folders is as follows:
File description for the unzipped folder:. ├── ALL_HR |── ALL_mask |── valid | ├── defocus | | ├── LR | | | └── X4 | └── motion | ├── LR | | └── X4 |── area_class.npy |── degree_class.npy |── defocus_motion_class.npy |── train_validation.npy └── train_validation_split.py
-
README.md
: This file. -
ALL_HR
: The high-resolution images of the ReBlurSR dataset, including the ReBlurSR-Train and ReBlurSR-Test subsets. -
ALL_mask
: The blur map for the HR images inALL_HR
. For each sample, the blur map value is0(0)
if the pixel is blurred, and1(255)
if the pixel is non-blurred. -
valid
: contains the LR versions of the ReBlurSR-Test subset.defocus
andmotion
subfolders contain the defocus and motion subsubsets of the ReBlurSR-Test subset, respectively. Its structure tree is as follows:valid ├── defocus │ ├── LR │ │ └── X4 └── motion └── LR └── X4
-
area_class.npy
: The category of the area of the blur region in the ReBlurSR-Test subset. Samples are divided into 3 classes: small, medium, and large.0:"large", 1:"medium", 2:"small
-
degree_class.npy
: The category of the degree of the blur region in the ReBlurSR-Test subset. Samples are divided into 3 classes: heavy, little, and middle.0:"heavy", 1:"little", 2:"middle"
-
defocus_motion_class.npy
: The category of the blur type in the ReBlurSR-Test subset. Samples are divided into 2 classes: defocus and motion.0:"defocus", 1:"motion"
-
train_validation.npy
: The category of the samples in the ReBlurSR-Train and ReBlurSR-Test subsets. Samples are divided into 2 classes: train and validation.0:"train", 1:"validation"
-
train_validation_split.py
: The script to generate the ReBlurSR-Train and ReBlurSR-Test subsets fromALL_HR
andALL_mask
folders according to thedefocus_motion_class.npy
andtrain_validation.npy
files.
-
- run the
train_validation_split.py
script to generate the ReBlurSR-Train and ReBlurSR-Test subsets.python train_validation_split.py # required packages: numpy, tqdm
- The generated ReBlurSR-Train and ReBlurSR-Test subsets are saved in the
train
andvalid
folders, respectively. The structure tree of the completetrain
folder andvalid
folder are as follows:train ├── motion │ ├── HR │ └── mask └── defocus ├── HR └── mask valid ├── motion │ ├── HR │ ├── LR | | └── X4 │ └── mask └── defocus ├── HR ├── LR | └── X4 └── mask
Generate the subsets of the ReBlurSR-Test
You can generate the subsets of the ReBlurSR-Test according to the area_class.npy
, degree_class.npy
, and defocus_motion_class.npy
files.