Real-World Blur Dataset for Learning and Benchmarking Deblurring Algorithms


ECCV 2020

Abstract


In this work, we present a large-scale dataset of real-world blurred images and ground truth sharp images for learning and benchmarking single image deblurring methods. To collect our dataset, we build an image acquisition system to simultaneously capture geometrically aligned pairs of blurred and sharp images, and develop a postprocessing method to produce high-quality ground truth images. We analyze the effect of our postprocessing method and the performance of existing deblurring methods. Our analysis shows that our dataset significantly improves deblurring quality for real-world blurred images.

...

Dataset


RealBlur

Real-world blur dataset acquired from wide-angle lens

RealBlur-Tele

Real-world blur dataset acquired from telephoto lens

BSD-B

Synthetic uniform blur dataset using BSD dataset

Trained Model

SRN-DeblurNet and DeblurGAN-v2 trained with RealBlur dataset

RealBlur

training set : 3,758 image pairs of 182 different scenes.
test set : 980 image pairs of 50 different scenes.

RealBlur-Tele

test set : 996 image pairs of 50 different scenes.

BSD-B

training set : 20,000 image pairs of synthetic uniform blurred images and sharp images.

Code

Training and Test code
Post-processing code
Evaluation code

Benchmark


Benchmark of state-of-the-art deblurring methods on real-world blurred images. Purple: traditional optimization-based methods. Black: deep learning-based methods. Blue*: models trained with our dataset. Methods are sorted in the descending order with respect to PSNR.

RealBlur-J RealBlur-R
Methods PSNR/SSIM Methods PSNR/SSIM
SRN-DeblurNet* 31.38/0.9091 SRN-DeblurNet* 38.65/0.9652
DeblurGAN-v2* 29.69/0.8703 DeblurGAN-v2* 36.44/0.9347
DeblurGAN-v2 28.70/0.8662 J.Zhang et al. 35.70/0.9481
SRN-DeblurNet 28.56/0.8674 SRN-DeblurNet 35.66/0.9472
J.Zhang et al. 28.42/0.8596 H.Zhang et al. 35.48/0.9466
DeblurGAN 27.97/0.8343 DeblurGAN-v2 35.26/0.9440
Nah et al. 27.87/0.8274 Xu et al. 34.46/0.9368
H.Zhang et al. 27.80/0.8472 Pan et al. 34.01/0.9162
Pan et al. 27.22/0.7901 DeblurGAN 33.79/0.9034
Xu et al. 27.14/0.8303 Hu et al. 33.67/0.9158
Hu et al. 26.41/0.8028 Nah et al. 32.51/0.8406

Bibtex


@inproceedings{rim_2020_ECCV,
 title={Real-World Blur Dataset for Learning and Benchmarking Deblurring Algorithms},
 author={Jaesung Rim, Haeyun Lee, Jucheol Won, Sunghyun Cho},
 booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
 year={2020},
}

Let's Get In Touch!


Please feel free to contact us with any feedback, questions, or comments