The Blackbird unmanned aerial vehicle (UAV) dataset is a large-scale,aggressive indoor flight dataset collected using a custom-built quadrotorplatform for use in evaluation of agile perception.Inspired by the potential offuture high-speed fully-autonomous drone racing, the Blackbird dataset containsover 10 hours of flight data from 168 flights over 17 flight trajectories and 5environments at velocities up to $7.0ms^-1$. Each flight includes sensor datafrom 120Hz stereo and downward-facing photorealistic virtual cameras, 100HzIMU, $\sim190Hz$ motor speed sensors, and 360Hz millimeter-accurate motioncapture ground truth. Camera images for each flight were photorealisticallyrendered using FlightGoggles across a variety of environments to facilitateeasy experimentation of high performance perception algorithms. The dataset isavailable for download at http://blackbird-dataset. mit.edu/.