Non-line-of-sight imaging autocorrelation and Ground Truth Dataset

This dataset is prepared for training the reconstruction network in non-line-of-sight imaging experiments, so that the network can recover object images from target's autocorrelation. There are two types of data in total, the image of single-digit and its corresponding autocorrelation; The image of three-digits and its corresponding autocorrelation. All images have a resolution of 128 * 128. There are 20000 data pairs for single-digit image, and 55,000 data pairs for three-digit image. The single-digit image is derived from the publicly available MNIST dataset, and three-digit images are composed of three MINST images pieced together to form one image. You can also build this dataset yourself.