License agreement

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

By requesting access, you agree to the terms and conditions of original KITTI dataset license (CC-BY-SA-NC 3.0). Further, you attest that you have created account on original KITTI website and that your information is accurate. This work is for non-commercial research use only.

Log in or Sign Up to review the conditions and access this dataset content.

Dataset Card for Dataset Name

KITTI-QA is a multimodal (2D/3D) video question-answering benchmark for spatio-temporal reasoning in autonomous driving. Derived from the KITTI tracking dataset [3], it comprises 21 training and 29 testing scenes, with 3D depth maps generated via BridgeDepth [4]. Each scene features 40 safety-centric, handcrafted QA pairs that require models to integrate temporal dynamics with traffic safety knowledge.

Dataset Details

Dataset Description

  • Language(s) (NLP): English (US)
  • License: CC-BY-NC-SA 3.0

As a derivative work of the KITTI tracking dataset [3], KITTI-QA inherits and complies with the CC BY-NC-SA 3.0 licensing agreement.

Dataset Sources

  • Webpage: https://automotivesafety-lvlm.github.io
  • Paper: [IEEE IV 2026] D3VL: Understanding Driving Scenes from 3D Time Series Data and Video with Language Models
        [IEEE IV 2024] Semantic Understanding of Traffic Scenes with Large Vision Language Models

Dataset Structure

Dataset is divided into train and test datasets. Each row represents (multimodal) video-question-answer triplet.

column definition
video_name: name of the scene
left: relative path of RGB video from left camera (i.e. KITTI camera #3)
right: relative path of RGB video from left camera (i.e. KITTI camera #4)
depth: relative path of Depth video from stereo camera and BridgeDepth
question: question
answer: answer

Dataset Example




Curation Rationale

[More Information Needed]

Source Data

  • Raw stereo videos: KITTI Object Tracking Benchmark [3]
  • Stereo Depth Estimation: Done by authors [1] using BridgeDepth [4]
  • Annotation: Done by authors [1], [2]

[More Information Needed]

Citation [optional]

If you find this work is helpful, please consider citing the followings:

BibTeX:

[1] D3VL

@article{han2026d3vl,
  title={D3VL: Understanding Drive Scenes from 3D Time Series Data and Video with Language Models},
  author={Han, Heesang and Abbott, A. Lynn and Sarkar, Abhijit},
  journal={2026 IEEE Intelligent Vehicle Symposium (IV)},
  year={2026}
}

[2] Semantic Understanding of Traffic Scenes with Large Vision Language Models

@INPROCEEDINGS{10588373,
  author={Jain, Sandesh and Thapa, Surendrabikram and Chen, Kuan-Ting and Abbott, A. Lynn and Sarkar, Abhijit},
  booktitle={2024 IEEE Intelligent Vehicles Symposium (IV)}, 
  title={Semantic Understanding of Traffic Scenes with Large Vision Language Models}, 
  year={2024},
  pages={1580-1587},
  keywords={Location awareness;Visualization;Laser radar;Semantics;Transportation;Cameras;Cognition;large vision language models (LVLM);scene analysis;automated perception},
  doi={10.1109/IV55156.2024.10588373}}

[3] KITTI Tracking Dataset

@article{Geiger2013IJRR,
  author = {Andreas Geiger and Philip Lenz and Christoph Stiller and Raquel Urtasun},
  title = {Vision meets Robotics: The KITTI Dataset},
  journal = {International Journal of Robotics Research (IJRR)},
  year = {2013}
}

[4] BridgeDepth Stereo Depth Estimation (Depth Videos)

@article{guan2025bridgedepth,
    author    = {Guan, Tongfan and Guo, Jiaxin and Wang, Chen and Liu, Yun-Hui},
    title     = {BridgeDepth: Bridging Monocular and Stereo Reasoning with Latent Alignment},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2025},
    pages     = {27681-27691}
}

It has been generated using this raw template.

Downloads last month
29