DOLPHINS: Dataset for Collaborative Perception Enabled Harmonious and Interconnected Self-driving

Vehicle-to-Everything (V2X) network has enabled collaborative perception in autonomous driving, which is a promising solution to the fundamental defect of stand-alone intelligence including blind zones and long-range perception. However, lack datasets severely blocked development algorithms. In this work, we release DOLPHINS: Dataset for cOllaborative Perception Harmonious INterconnected Self-driving, as new simulated large-scale various-scenario multi-view multi-modality driving dataset, provides ground-breaking benchmark platform interconnected driving. DOLPHINS outperforms current six dimensions: temporally-aligned images point clouds from both vehicles Road Side Units (RSUs) enabling Vehicle-to-Vehicle (V2V) Vehicle-to-Infrastructure (V2I) based perception; 6 typical scenarios with dynamic weather conditions make most various dataset; meticulously selected viewpoints providing full coverage key areas every object; 42376 frames 292549 objects, well corresponding 3D annotations, geo-positions, calibrations, compose largest dataset Full-HD 64-line LiDARs construct high-resolution data sufficient details; well-organized APIs open-source codes ensure extensibility DOLPHINS. We also 2D detection, tasks on The experiment results show that raw-level fusion scheme through V2X communication can help improve precision reduce necessity expensive LiDAR equipment when RSUs exist, may accelerate popularity self-driving vehicles. now available https://dolphins-dataset.net/.