A challenging multi-frame interpolation dataset for autonomous driving scenarios. Based on the principle of hard-sample selection and the diversity of scenarios, NL-Drive dataset contains point cloud sequences with large nonlinear movements from three public large-scale autonomous driving datasets: KITTI, Argoverse and Nuscenes. The overall dataset contains more than 20,000 LiDAR point cloud frames. The frame rate of point cloud sequence is 10Hz. And NL-Drive dataset is split into the training, validation and test set in the ratio of 14:3:3.
For the point cloud interpolation task, the point cloud frame input is selected at a given interval of frames, and the remaining point clouds as the ground truth of the interpolation frame. Particularly, each sample of NL-Drive dataset is 4 point cloud frames of 2.5Hz when there are 3 interpolation frames to predict between the middle two input frames.