TSEC-Dataset

TSEC‑Dataset Overview

Dataset Purpose

TSEC‑Dataset is designed to provide training and testing data for video captioning methods in driving scenarios. The dataset focuses on describing key events involving the ego vehicle, road environment, and other traffic participants.

Dataset Content

• Video Sources: On‑board camera recordings, other public dataset videos, and traffic‑accident videos downloaded from BiliBili and YouTube.
• Video Processing: Videos are divided into sub‑segments containing 1 to 3 key events.
• Video Characteristics: Diverse scenes covering various weather conditions, time periods, road conditions, and vehicle conditions, with particular emphasis on intersection scenes.
• Quantity & Duration: 8,000 video segments totaling 11.5 hours.

Dataset Annotation

• Annotation Method: Participants simulate a driving experience, watch the original videos without captions, and determine the types of key events through a voting process.
• Key Events: For each video, the four most voted key events are selected for detailed annotation.

Access Method

• Usage Restrictions: Research and personal learning use only.
• Request Procedure: Request via email (haopenghui2022@gmail.com) with a description of database usage and a signed copy of the agreement.
• User Eligibility: Must be a legitimate institution or its department; academic requests require a university email and a signature from a professor or official staff member.

Usage Terms

• Privacy Protection: If personal or vehicle information is discovered, contact the dataset provider for removal.
• Copyright: The dataset is owned by the Shandong University Computer Vision and Pattern Recognition Laboratory and is licensed under the Creative Commons Attribution‑NonCommercial‑ShareAlike 4.0 License.