This repository provides the MVVA database in our ECCV paper "Learning to Predict Salient Faces: A Novel Visual-Audio Saliency Model".

MVVA is a large-scale eye-tracking database of multiple-face video in visual-audio condition. MVVA contains Eye Movement data of 34 subjects on 300 videos, as well as the soud source annotation in frame level for all 300 videos.

This database can be used for visual-audio saliency prediction, sound source localization, activate speaker detection, speaker diration, etc. For more details, please refer to our paper.

MVVA database can be downloaded from DropBox (Click to view) or 百度云. Please feel free to contact us by clicking here so that we can give you access to the database. Then extract it with:

unzip xxx.zip

xxxs

run the following command to generate saliency maps

python demo.py

Usage Example

The multiple-face videos in our MVVA database are at diverse scenarios, and can be categorized into 6 classes, including TV play/movie, interview, video conference, variety show, music and group discussion.

The audio content covers different scenarios including quiet scenes and noisy scenes, as reported in the following table. In the noisy scenes, the background sounds contain laughter, street, music, applause, crowd and noise.

if you find this database useful for your research, please cite:

@article{liu2020visualaudio,
  title={Learning to Predict Salient Faces: A Novel Audio-Visual Saliency Model},
  author={Yufan Liu; Minglang Qiao; Mai Xu; Bing Li; Weiming Hu; Ali Borji},
  booktitle=={Proceedings of the european conference on computer vision (eccv)},
  year={2020}
}

If you have any question, please contact minglangqiao@buaa.edu.cn (or yufan.liu@ia.ac.cn), or use public issues section of this repository.