Synthetic SPAD Depth Sensing Dataset

The synthetic SPAD depth sensing dataset is specifically designed to train and evaluate neural network architectures for histogram-less single-photon depth estimation tasks. It simulates diverse real-world conditions with comprehensive parameter combinations, including target distances, background illumination, target reflectivity, and atmospheric conditions. The dataset consists of two parts: a test set with fixed parameter combinations for consistent evaluation, and a training/validation set with randomly generated parameters to enhance model generalization. The dataset provides raw timestamps rather than processed point clouds. Each parameter combination contains 2000 independent timestamp sequences that have undergone basic coincidence detection. For detailed specifications, please refer to the paper and the supplementary material.