The Real Blur Dataset is a collection of blurry images collected in the real world, used to learn and evaluate image deblurring algorithms. The dataset includes 4,738 pairs of images in 232 scenes. Each training set consists of 3,758 image pairs, and each test set consists of 980 image pairs. According to the different image acquisition formats, the datasets can be divided into:

• A subset of RealBlur-R in raw format
• A subset of RealBlur-J in JPEG format