Argoverse 2: Next Generation Datasets for Self-Driving Perception and\n Forecasting

Abstract

We introduce Argoverse 2 (AV2) - a collection of three datasets for\nperception and forecasting research in the self-driving domain. The annotated\nSensor Dataset contains 1,000 sequences of multimodal data, encompassing\nhigh-resolution imagery from seven ring cameras, and two stereo cameras in\naddition to lidar point clouds, and 6-DOF map-aligned pose. Sequences contain\n3D cuboid annotations for 26 object categories, all of which are\nsufficiently-sampled to support training and evaluation of 3D perception\nmodels. The Lidar Dataset contains 20,000 sequences of unlabeled lidar point\nclouds and map-aligned pose. This dataset is the largest ever collection of\nlidar sensor data and supports self-supervised learning and the emerging task\nof point cloud forecasting. Finally, the Motion Forecasting Dataset contains\n250,000 scenarios mined for interesting and challenging interactions between\nthe autonomous vehicle and other actors in each local scene. Models are tasked\nwith the prediction of future motion for "scored actors" in each scenario and\nare provided with track histories that capture object location, heading,\nvelocity, and category. In all three datasets, each scenario contains its own\nHD Map with 3D lane and crosswalk geometry - sourced from data captured in six\ndistinct cities. We believe these datasets will support new and existing\nmachine learning research problems in ways that existing datasets do not. All\ndatasets are released under the CC BY-NC-SA 4.0 license.\n