KITTI 360

Overview

The original KITTI-360 is a large-scale annotated dataset recorded in the suburbs of Karlsruhe, Germany, covering 73.7 km of driving with over 320k images and 100k laser scans. It provides both 3D point clouds and 2D images with dense semantic and instance annotations across 19 classes, and maintains consistent instance IDs over time. The data, captured with a 360° sensing setup including fisheye and perspective cameras, a stereo camera, Velodyne, and SICK laser scanners, is precisely geolocalized using an IMU/GPS system.

Format description

Supported point cloud format: .bin With annotations: yes Supported annotation format: .xml Data structure: Information is provided below.

Input files structure

Format directory structure:

📦KITTI-360
├──📂calibration
│   ├──📄calib_cam_to_velo.txt
│   ├──📄perspective.txt
│   └──📄...
├──📂data_2d_raw
│   ├──📂2013_05_28_drive_{seq:04}_sync
│   │   ├──📂image_{00|01}
│   │   │   ├──📂data_rect
│   │   │   │   ├──🏞️{frame:010}.png
│   │   │   │   └──🏞️...
│   │   ├──📂image_{02|03}
│   │   │   └──📂...
├──📂data_3d_raw
│   ├──📂2013_05_28_drive_{seq:04}_sync
│   │   ├──📂velodyne_points
│   │   │   ├──📂data
│   │   │   │   ├──📄{frame:010}.bin
│   │   │   │   └──📄...
├──📂data_3d_bboxes
│   ├──📂train
│   │   ├──📄2013_05_28_drive_{seq:04}_sync.xml
│   │   └──📄...
├──📂data_poses
│   ├──📂2013_05_28_drive_{seq:04}_sync
│   │   ├──📄cam0_to_world.txt
└── └── └──📄...

Information on the KITTI-360 dataset structure can be found here

Last updated

Was this helpful?