pykitti

Why we included this project

Working with the KITTI autonomous driving benchmark usually means writing glue code to unpack the raw archive. pykitti removes that step: give it a date/drive sequence and it loads calibration, timestamps, OXTS pose logs, and hands you camera frames (mono and RGB stereo pairs), Velodyne point clouds, and the 4x4 transforms that map between LiDAR, camera, and IMU frames. It stays minimal on purpose, returning plain numpy arrays and Pillow images, so it fits into whatever pipeline you already have instead of pulling in a framework. You can stream frames sequentially for visual odometry or index into the dataset for deep learning, and you can restrict loading to a specific set of frames if you only need a slice. If you're a researcher or engineer who wants to spend time on the model rather than the data plumbing, this is the tool.