README.md

SUTD-TrafficQA

A challenging Video Question Answering (VQA) Benchmark based on real-world traffic scenes.

Updates:

• Jul 2021 The dataset is publicly released. You may request download now.
• Jun 2021 The dataset usage details are available now.
• May 2021 The dataset homepage is live now.
• Feb 2021 The dataset is available upon email request.

Paper

Our paper at CVPR 2021, SUTD-TrafficQA: A Question Answering Benchmark and an Efficient Network for Video Reasoning over Traffic Events, is available at: [CVF Open Access], [arXiv:2103.15538], and [ResearchGate].

Dataset

• Annotation Example examples/annotation_sample.jsonl
• Jsonl Reader Example examples/jsonl_reader.py
• Appearance Feature Preprocessing examples/preprocess_video_appearance_example.py
• Motion Feature Preprocessing examples/preprocess_video_motion_example.py
• Dataloader examples/dataloader_example.py
• Download Dataset

Citation

@InProceedings{Xu_2021_CVPR,
    author    = {Xu, Li and Huang, He and Liu, Jun},
    title     = {{SUTD-TrafficQA: A Question Answering Benchmark and an Efficient Network for Video Reasoning Over Traffic Events}},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2021},
    pages     = {9878-9888}
}

Acknowledgment

Contributors: Lin Yutian, Tran Nguyen Bao Long, Liu Renhang, Qiao Yingjie, Xun Long Ng, Koh Kai Ting, Christabel Dorothy

Code Reference: thaolmk54 / hcrn-videoqa

Contact

• li_xu [AT] mymail.sutd.edu.sg
• he_huang [AT] mymail.sutd.edu.sg