Traffic Video Captioning (TVC) dataset for training and testing video captioning methods for driving scenes
#Traffic Video Captioning Dataset (TVC-dataset) The existing datasets for video captioning contain videos with a variety of scenarios, and are not for ADAS traffic scenarios. For the video understanding and captioning problem, we build a dataset for ADAS traffic scenarios, called Traffic Video Captioning (TVC) dataset. The TVC dataset contains 2,000 traffic videos captured with on-board cameras. With various driving events, the TVC dataset is the first public dataset for video understanding and captioning problems.
The TVC dataset has video-clips from our captured videos with an ADAS camera and many traffic videos from other public datasets including $D^2-City$ \cite{didi}, $DADA-2000$ \cite{dada} and $Road Hazard$ \cite{rapid}. The $D^2-City$ dataset is a comprehensive collection of dashcam videos collected by vehicles on DiDi platform in five Chinese cities; all videos are 30-second clips in 720p or 1080p resolution at a frame rate of 25fps. $DADA-2000$ dataset is built for predicting driver attention in driving accident scenarios with diverse weather conditions, light conditions, and occasions, etc. $Road Hazard$ dataset is a dataset for road hazard stimuli to assess how quickly participants in the laboratory can detect and correctly respond to briefly presented hazards.
Free copies of TVC-dataset are provided for research and private study only. A copy of the database can be requested through e-mail sent to:
The e-mail message must contain a description of the envisioned use of the database and a scanned version of Agreement For TVC-dataset. Please note that the End-User named must be a legal institution, or a department or section of a named legal institution, not an individual nor a project-the person signing this Agreement has to be duly authorized by the institution for such signatures (e.g., Department or Administrative Head or similar) and shall be liable for such authorization. For academic requests, please use your university e-mail. A signature of Pof. or other regular employee of academic units (Student signature may be denied). A download link will be provided after the analysis of the request. Requests may be denied.
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 TVC-dataset 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 3.0 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.