Welcome to the devkit of the KITTI3D Instance Segmentation annotations.

The instance segmentation annotations, which are matched to the already annotated 3D bounding boxes of the KITTI3D dataset, are proveded as part of the paper: MonoCInIS: Camera Independent Monocular 3D Object Detection using Instance Segmentation

This repository contains some info on the annotation format and example code for visualising the instances.

For every image of the KITTI3D dataset (7481 training images), we manually annotated all vehicle and pedestrian instances. The annotations are provided as single channel .png files, where the pixels of each instance have a unique id. To link each instance to its corresponding KITTI3D bounding box, we use following convention:

Download the original dataset from the official KITTI website.

Download our instance segmentation annotations here (15MB).

To get you started, we provide some example code to load and visualise our instance annotations and the corresponding bbox annotations.

Check out the provided IPython Notebook to visualise our annotations.

If you find these annotations useful, please cite our paper:

MonoCInIS: Camera Independent Monocular 3D Object Detection using Instance Segmentation

Jonas Heylen, Mark De Wolf, Bruno Dawagne, Marc Proesmans, Luc Van Gool, Wim Abbeloos, Hazem Abdelkawy, Daniel Olmeda Reino

@inproceedings{heylen2021monocinis,
  title={MonoCInIS: Camera Independent Monocular 3D Object Detection using Instance Segmentation},
  author={Heylen, Jonas and De Wolf, Mark and Dawagne, Bruno and Proesmans, Marc and Van Gool, Luc and Abbeloos, Wim and Abdelkawy, Hazem and Reino, Daniel Olmeda},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={923--934},
  year={2021}
}

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.