Helsinki Deblur Challenge 2021 test dataset

This dataset was primarily designed and captured to be used for the testing part of the Helsinki Deblur Challenge 2021 but it can be used for any testing and benchmarking purposes of image deblurring algorithms.

The dataset contains photographs of random strings of text (including numbers), natural images and QR codes with varying levels of blur caused by misfocusing the camera. Each photo has both a blurred and sharp version.

The images are split into 4 separate zip files each having 5 steps of different blur. Altogether there are 20 steps. Each one of the zip files contains two folders, one folder named CAM1_focused (Camera 1) with the sharp images and one named CAM2_blurred (Camera 2) with the blurred images. For each step there are 40 images of text character targets, 15 natural image targets, one QR-code target and 3 images of technical targets. Each one of the text target images is accompanied by a text file (same file name but .txt extension) containing the correct transcription of that particular text target.

As a difference to the HDC training data (https://doi.org/10.5281/zenodo.4916176) the test data character targets include also numbers.  

A detailed description of the training dataset that was acquired in the exact same way as this test dataset can be found here: http://arxiv.org/abs/2105.10233

Here is a link to the official webpage of the Helsinki Deblur Challenge 2021: https://www.fips.fi/HDC2021.php