Citation

If you are using the DIV2K dataset please add a reference to the introductory dataset paper and to one of the following challenge reports.

@InProceedings{Agustsson_2017_CVPR_Workshops,
	author = {Agustsson, Eirikur and Timofte, Radu},
	title = {NTIRE 2017 Challenge on Single Image Super-Resolution: Dataset and Study},
	booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
	month = {July},
	year = {2017}
} 

@InProceedings{Timofte_2017_CVPR_Workshops,
author = {Timofte, Radu and Agustsson, Eirikur and Van Gool, Luc and Yang, Ming-Hsuan and Zhang, Lei and Lim, Bee and others},
title = {NTIRE 2017 Challenge on Single Image Super-Resolution: Methods and Results},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {July},
year = {2017}
}

@InProceedings{Timofte_2018_CVPR_Workshops,
author = {Timofte, Radu and Gu, Shuhang and Wu, Jiqing and Van Gool, Luc and Zhang, Lei and
Yang, Ming-Hsuan and Haris, Muhammad and others},
title = {NTIRE 2018 Challenge on Single Image Super-Resolution: Methods and Results},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2018}
}

@InProceedings{Timofte_2018_CVPR_Workshops,
author = {Timofte, Radu and Gu, Shuhang and Wu, Jiqing and Van Gool, Luc and Zhang, Lei and
Yang, Ming-Hsuan and Haris, Muhammad and others},
title = {NTIRE 2018 Challenge on Single Image Super-Resolution: Methods and Results},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2018}
}

@InProceedings{Ignatov_2018_ECCV_Workshops,
author = {Ignatov, Andrey and Timofte, Radu and others},
title = {PIRM challenge on perceptual image enhancement on smartphones: report},
booktitle = {European Conference on Computer Vision (ECCV) Workshops},
month = {January},
year = {2019}
}

Supplementary material (PSNR, SSIM, IFC, CORNIA results for top NTIRE 2017 challenge methods (SNU_CVLab, HelloSR, Lab402), VDSR and A+ on DIV2K, Urban100, B100, Set14, Set5)

License

Please notice that this dataset is made available for academic research purpose only. All the images are collected from the Internet, and the copyright belongs to the original owners. If any of the images belongs to you and you would like it removed, please kindly inform us, we will remove it from our dataset immediately.

 

Data overview

We are making available a large newly collected dataset -DIV2K- of RGB images with a large diversity of contents.

The DIV2K dataset is divided into:

• train data: starting from 800 high definition high resolution images we obtain corresponding low resolution images and provide both high and low resolution images for 2, 3, and 4 downscaling factors
• validation data: 100 high definition high resolution images are used for genereting low resolution corresponding images, the low res are provided from the beginning of the challenge and are meant for the participants to get online feedback from the validation server; the high resolution images will be released when the final phase of the challenge starts.
• test data: 100 diverse images are used to generate low resolution corresponding images; the participants will receive the low resolution images when the final evaluation phase starts and the results will be announced after the challenge is over and the winners are decided.

Data structure

DIV2K dataset has the following structure:

1000 2K resolution images divided into: 800 images for training, 100 images for validation, 100 images for testing

For each challenge Track (with 1. bicubic or 2. unknown downgrading operators) we have:

• the high resolution images: 0001.png, 0002.png, ..., 1000.png
• the downscaled images:
  • YYYYx2.png for downscaling factor x2; where YYYY is the image ID
  • YYYYx3.png for downscaling factor x3; where YYYY is the image ID
  • YYYYx4.png for downscaling factor x4; where YYYY is the image ID

DIV2K forder structure is as follows:

DIV2K/ -- DIV2K dataset

DIV2K/DIV2K_train_HR/ -- 0001.png, 0002.png, ..., 0800.png train HR images (provided to the participants)
DIV2K/DIV2K_train_LR_bicubic/ -- corresponding low resolution images obtained using Matlab imresize function with default settings (bicubic interpolation)
DIV2K/DIV2K_train_LR_bicubic/X2/ -- 0001x2.png, 0002x2.png, ..., 0800x2.png train LR images, downscale factor x2
DIV2K/DIV2K_train_LR_bicubic/X3/ -- 0001x3.png, 0002x3.png, ..., 0800x3.png train LR images, downscale factor x3
DIV2K/DIV2K_train_LR_bicubic/X4/ -- 0001x4.png, 0002x4.png, ..., 0800x4.png train LR images, downscale factor x4
DIV2K/DIV2K_train_LR_unknown/ -- corresponding low resolution images obtained using degradation operators kept hidden, unknown to the participants
DIV2K/DIV2K_train_LR_unknown/X2/ -- 0001x2.png, 0002x2.png, ..., 0800x2.png train LR images, downscale factor x2
DIV2K/DIV2K_train_LR_unknown/X3/ -- 0001x3.png, 0002x3.png, ..., 0800x3.png train LR images, downscale factor x3
DIV2K/DIV2K_train_LR_unknown/X4/ -- 0001x4.png, 0002x4.png, ..., 0800x4.png train LR images, downscale factor x4

DIV2K/DIV2K_valid_HR/ -- 0801.png, 0802.png, ..., 0900.png validation HR images (will be available to the participants at the beginning of the final evaluation phase)
DIV2K/DIV2K_valid_LR_bicubic/ -- corresponding low resolution images obtained using Matlab imresize function with default settings (bicubic interpolation)
DIV2K/DIV2K_valid_LR_bicubic/X2/ -- 0801x2.png, 0802x2.png, ..., 0900x2.png train LR images, downscale factor x2
DIV2K/DIV2K_valid_LR_bicubic/X3/ -- 0801x3.png, 0802x3.png, ..., 0900x3.png train LR images, downscale factor x3
DIV2K/DIV2K_valid_LR_bicubic/X4/ -- 0801x4.png, 0802x4.png, ..., 0900x4.png train LR images, downscale factor x4
DIV2K/DIV2K_valid_LR_unknown/ -- corresponding low resolution images obtained using degradation operators kept hidden, unknown to the participants
DIV2K/DIV2K_valid_LR_unknown/X2/ -- 0801x2.png, 0802x2.png, ..., 0900x2.png train LR images, downscale factor x2
DIV2K/DIV2K_valid_LR_unknown/X3/ -- 0801x3.png, 0802x3.png, ..., 0900x3.png train LR images, downscale factor x3
DIV2K/DIV2K_valid_LR_unknown/X4/ -- 0801x4.png, 0802x4.png, ..., 0900x4.png train LR images, downscale factor x4

DIV2K/DIV2K_test_HR/ -- 0901.png, 0902.png, ..., 1000.png test HR images (not provided to the participants, used for final evaluation and ranking)
DIV2K/DIV2K_test_LR_bicubic/ -- corresponding low resolution images obtained using Matlab imresize function with default settings (bicubic interpolation)
DIV2K/DIV2K_test_LR_bicubic/X2/ -- 0901x2.png, 0902x2.png, ..., 1000x2.png train LR images, downscale factor x2
DIV2K/DIV2K_test_LR_bicubic/X3/ -- 0901x3.png, 0902x3.png, ..., 1000x3.png train LR images, downscale factor x3
DIV2K/DIV2K_test_LR_bicubic/X4/ -- 0901x4.png, 0902x4.png, ..., 1000x4.png train LR images, downscale factor x4
DIV2K/DIV2K_test_LR_unknown/ -- corresponding low resolution images obtained using degradation operators kept hidden, unknown to the participants
DIV2K/DIV2K_test_LR_unknown/X2/ -- 0901x2.png, 0902x2.png, ..., 1000x2.png train LR images, downscale factor x2
DIV2K/DIV2K_test_LR_unknown/X3/ -- 0901x3.png, 0902x3.png, ..., 1000x3.png train LR images, downscale factor x3
DIV2K/DIV2K_test_LR_unknown/X4/ -- 0901x4.png, 0902x4.png, ..., 1000x4.png train LR images, downscale factor x4

 

Data access

 

• (NTIRE 2017) Low Res Images:
  • Train Data Track 1 bicubic downscaling x2 (LR images)
  • Train Data Track 2 unknown downgrading operators x2 (LR images)
  • Validation Data Track 1 bicubic downscaling x2 (LR images)
  • Validation Data Track 2 unknown downgrading operators x2 (LR images)
  • Train Data Track 1 bicubic downscaling x3 (LR images)
  • Train Data Track 2 unknown downgrading operators x3 (LR images)
  • Validation Data Track 1 bicubic downscaling x3 (LR images)
  • Validation Data Track 2 unknown downgrading operators x3 (LR images)
  • Train Data Track 1 bicubic downscaling x4 (LR images)
  • Train Data Track 2 unknown downgrading operators x4 (LR images)
  • Validation Data Track 1 bicubic downscaling x4 (LR images)
  • Validation Data Track 2 unknown downgrading operators x4 (LR images)
• (NTIRE 2018) Low Res Images:
  • Train Data Track 1 bicubic x8 (LR images)
  • Train Data Track 2 realistic mild x4 (LR images)
  • Train Data Track 3 realistic difficult x4 (LR images)
  • Train Data Track 4 realistic wild x4 (LR images)
  • Validation Data Track 1 bicubic x8 (LR images)
  • Validation Data Track 2 realistic mild x4 (LR images)
  • Validation Data Track 3 realistic difficult x4 (LR images)
  • Validation Data Track 4 realistic wild x4 (LR images)
• High Resolution Images:
  • Train Data (HR images)
  • Validation Data (HR images)

 

Scoring scripts

 

Matlab scoring functions used by NTIRE 2017 challenge for the evaluation of the solutions
Scoring functions used by NTIRE 2018 realistic tracks for the evaluation of the solutions