The CERTH-ITI-VAQ700 dataset

Overview:

Our aim is to provide a comprehensive video dataset for the problem of aesthetic quality assessment consisting of user-created videos capturing moments of everyday life, such as excursions, school concerts, and training processes. We downloaded from YouTube 700 videos covering a variety of categories, such as outdoor activities, do it yourself videos, make up tutorials, lectures, and home-made videos, licensed under Creative Commons Attribution. The duration of each of these videos ranges from 1 to 6 minutes.

We conducted an annotation process that involved 12 annotators watching and evaluating the aesthetic value of these videos by assigning binary aesthetic quality ratings; 1 being assigned to videos of high aesthetic quality and 0 to videos of low aesthetic quality. Each video was assessed by 5 annotators. The final aesthetic score of each annotated video was calculated as the median score of the annotators' individual scores. As a result of the annotation process, 350 videos are rated as being of high aesthetic quality and another 350 as being of low aesthetic quality.

A comprehensive representation scheme that exploits photo- and motion-based features, motivated by photography and cinematography rules, is also provided for this dataset. For more details concerning the extracted video features, please refer to [1].

The CERTH-ITI-VAQ700 dataset includes:

• 700 YouTube videos in .mp4 format (download here)

• Video features (download here)

• Annotation (download here)

Copyright notice:

All videos were downloaded from YouTube, where they were available under the Creative Commons License. This means that the dataset is available for research, non-commersial purposes. For more details on data use and re-use please refer to CC BY license.

Publication:

If you use the CERTH-ITI-VAQ700 dataset in your research work, please cite the following paper:

[1] C. Tzelepis, E. Mavridaki, V. Mezaris, I. Patras "Video Aesthetic Quality Assessment using Kernel Support Vector Machine with Isotropic Gaussian Sample Uncertainty (KSVM-iGSU)", Proc. IEEE International Conference on Image Processing (ICIP 2016), Phoenix, Arizona, USA, September 2016