Stanford I2V Dataset

Stanford I2V is a new large-scale dataset for the evaluation of query-by-image video search. It contains 3,801 hours of news videos and 229 queries with annotated ground-truth sequences.

Jump to a section :

Online visualization | Download | Code | Paper/Citation | Authors/Contact

Figure 1: Some query images and video frames from our dataset.

Online visualization

Ground-truth visualization: Each link here points to the visualizations for a group of queries. In each visualization, we note below the video player the start and end time of ground-truth segments containing the query image:

201210, 201211, 201212, 201301, 201302, 201303, 201304, 201305, 201306, 201307, 201308, 201309, economist, time_magazine.

Frame-based visualization: This other visualization allows a quick view of the frames from ground-truth sequences. Each link below points to the visualizations for a group of queries. (Note: due to slight uncertainty in keyframe extraction, there might be a small number of frames per segment, at the beginning or at the end, which are incorrect):

201210, 201211, 201212, 201301, 201302, 201303, 201304, 201305, 201306, 201307, 201308, 201309, economist, time_magazine.

Download

The full dataset can be downloaded from this link. Make sure to read the README file for information and instructions on the best way to setup the data once it's downloaded.

Code

A standard performance evaluation script for the dataset is provided on our github repository. Please make sure to use it in case you are comparing results to previously published work.

In the current version of the repository, we also provide code for easy setup of the system, with keyframe and SIFT descriptor extraction from the videos.

Paper/Citation

Paper: pdf

If you make use of the dataset, please make sure to cite: A. Araujo, J. Chaves, D. Chen, R. Angst and B. Girod. "Stanford I2V: A News Video Dataset for Query-by-Image Experiments", in Proc. ACM Multimedia Systems, 2015

Bibtex:

@inproceedings{AraujoMMSYS2015,

author = {Araujo, A. and Chaves, J. and Chen, D. and Angst, R. and Girod, B.},

title = {{Stanford I2V: A News Video Dataset for Query-by-Image Experiments}},

booktitle = {Proc. ACM Multimedia Systems},

year = {2015},

}

 

Authors/Contact

Andre Araujo, Jason Chaves, David Chen, Roland Angst, Bernd Girod

For contact about the dataset, please reach Andre: afaraujo [AT] stanford [DOT] edu