Traffic Scenarios Event Caption (TSEC) Dataset

TSEC Dataset Usage Instructions

1. Dataset Overview

The TSEC dataset is specifically designed for research in the field of autonomous driving, aiming to support the description and analysis of key events in traffic scenarios. The dataset encompasses a variety of weather conditions, time periods, road types, and behaviors of traffic participants, providing high-quality multimodal data suitable for research on describing traffic scenario events.

2. Dataset Composition

• Original Videos: Contains 8,000 traffic scene videos, each 3 to 20 seconds long, recording diverse dynamic events.
• Semantic Segmentation Videos: Each original video corresponds to a semantic segmentation video, offering pixel-level scene analysis.
• Annotation File (CSV Format): Provides detailed textual descriptions of key events in each video, with four key event descriptions per video.

3. Data Format

• Video Files:
  • Original videos are stored in the TSEC/TSEC/ directory.
  • Semantic segmentation videos are stored in the TSEC/TSEC_MASK/ directory.
  • Videos are named with a uniform numbering system (e.g., 0001.mp4), ensuring a one-to-one correspondence between original and segmentation videos.
• Annotation File:
  • Located in the annotations.csv file.
  • The file includes the following fields:
    • Video ID: e.g., 0001.
    • Key Event Descriptions: e.g., "The traffic light at the intersection changed from red to green," "Our vehicle changed lanes to the left," "The vehicle ahead overtook another vehicle," etc.

4. Data Annotation

Annotations describe traffic events from three critical dimensions:

1. Road Environment Events: Such as red lights, green lights, obstacles appearing, snow-covered roads, etc.
2. Other Traffic Participants' Events: Such as lane changes by adjacent vehicles, the presence of pedestrians, other vehicles overtaking, or accidents involving other vehicles.
3. Ego Vehicle Events: Such as the ego vehicle overtaking, lane changing, accelerating, decelerating, or being involved in traffic accidents.

Each video clip was annotated by six annotators from the driver’s perspective. The final consistent labels were determined through statistical analysis.

5. Data Download

The TSEC dataset can be downloaded via the following Baidu Netdisk link:

• Link: https://pan.baidu.com/s/11AphlGIodKbnwT8yzDEn7Q 
• Access Code: TSEC 

Please ensure proper handling of the data after downloading and adhere to the licensing agreement.

6. Usage Notes

• Objective: The dataset is intended to support the description of key events in traffic scenarios. You can extract event information using the original videos, semantic segmentation videos, and annotation files for various analytical tasks.

7. License and Citation

This dataset is made available for academic use only. If you find your vehicle or personal information in this dataset, please contact us and we will remove the corresponding information from our dataset. We are not responsible for any actual or potential harm as the result of using this dataset.

The copyright of the TSEC-dataset is reserved by the Lab of computer vision and pattern recognition, Shandong University, China. The dataset described on this page is distributed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 License, which implies that you must: (1) Attribute the work as specified by the original authors; (2) May not use this work for commercial purposes; (3) If you alter, transform, or build upon this work, you may distribute the resulting work only under the same license. The dataset is provided "as it is" and we are not responsible for any subsequence from using this dataset.