KITTI Vision Benchmark Suite

The KITTI Vision Benchmark Suite leverages an autonomous driving platform to develop challenging real-world computer vision benchmarks. The datasets are captured by driving a standard station wagon, equipped with two high-resolution color and grayscale video cameras, a Velodyne laser scanner, and a GPS localization system, around the mid-size city of Karlsruhe, rural areas, and highways in Germany. This setup allows for the collection of diverse data for tasks such as stereo, optical flow, visual odometry, 3D object detection, and 3D tracking.

The platform provides all data in raw format and extracts specific benchmarks for each task, complete with evaluation metrics and an evaluation website. The goal is to reduce the bias found in laboratory-only benchmarks by offering real-world scenarios with novel difficulties, complementing existing benchmarks like Middlebury. The datasets and benchmarks are published under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 License, requiring attribution and non-commercial use. Researchers are encouraged to cite the associated publications when using the datasets.