KITTI-object consists of 7481 training point clouds (and images) and 7518 testing point clouds (and images).

• The 2D object detection and orientation estimation benchmark, uses 2D bounding box overlap to compute precision-recall curves for detection and computes orientation similarity to evaluate the orientation estimates in bird's eye view.
• The 3D object detection benchmark, uses 3D bounding box overlap to compute precision-recall curves.
• The bird's eye view benchmark, uses bounding box overlap in bird's eye view to compute precision-recall curves.

Data Collection

Our recording platform is a Volkswagen Passat B6, which has been modified with actuators for the pedals (acceleration and brake) and the steering wheel. The data is recorded using an eight core i7 computer equipped with a RAID system, running Ubuntu Linux and a real-time database. We use the following sensors:

• 1 Inertial Navigation System (GPS/IMU): OXTS RT 3003
• 1 Laserscanner: Velodyne HDL-64E
• 2 Grayscale cameras, 1.4 Megapixels: Point Grey Flea 2 (FL2-14S3M-C)
• 2 Color cameras, 1.4 Megapixels: Point Grey Flea 2 (FL2-14S3C-C)
• 4 Varifocal lenses, 4-8 mm: Edmund Optics NT59-917

The laser scanner spins at 10 frames per second, capturing approximately 100k points per cycle. The vertical resolution of the laser scanner is 64. The cameras are mounted approximately level with the ground plane. The camera images are cropped to a size of 1382 x 512 pixels using libdc's format 7 mode. After rectification, the images get slightly smaller. The cameras are triggered at 10 frames per second by the laser scanner (when facing forward) with shutter time adjusted dynamically (maximum shutter time: 2 ms). Our sensor setup with respect to the vehicle is illustrated in the following figure. Note that more information on calibration parameters is given in the calibration files and the development kit (see raw data section).