Automatic video curation driven by visual content popularity

Proceedings of ACM Multimedia 2015

A. Ortis, G. M. Farinella, S. Battiato
Image Processing Laboratory - University of Catania
{ortis, gfarinella, battiato}

V. D'Amico, L. Addesso, G. Torrisi
JOL WAVE - Telecom Italia
{valeria1.damico, luca.addesso, giovanni.torrisi}

Download the paper

Download input video sequences

This archive contains both the input videos and the Ground Truth popularity score among frames.
Download the RECfusion Dataset (2015)

Download video results

Our dataset Foosball RoomMeetingSAgata
Foosball Room Meeting SAgata
Peleg's dataset [1] ConcertLectureSeminar
Concert Lecture Seminar

Experimental Results

To evaluate the performances of the proposed method on each scenario we compute two measures obtained from the following scores computed on each clustering over time:

From the above scores, we compute the weighted mean of the ratios Pa/Pr and Pg/Pr over all the segmented blocks of a video, where the weights are given by the length of the blocks (i.e., the number of frames). When Pa/Pr is close to 1 the popularity score computed by our method is similar to the ground truth popularity. When this number is greater than 1 it means that the most popular cluster obtained with our approach is affected by outliers, whereas when this number is less than 1 it means that the our method missed some element of the ground truth popular cluster. Since Pa/Pr deal just with the number of video in the popular cluster, it is useful to look also at the ratio Pg/Pr. Indeed, Pg/Pr assesses the visual content of the videos in the popular cluster (true positive). This score have to be close to 1 to indicate accuracy in the popular cluster computed by our method.

The obtained results with respect to the different scenarios are reported in the Table below.

Scenario Devices Models Pa/Pr Pg/Pr
Foosball 4 2 1.02 1
Meeting 2 2 1.01 0.99
Meeting 4 4 0.99 0.95
Meeting 5 5 0.89 0.76
SAgata 7 6 1.05 1
Concert [1] 3 1 1.06 1
Lecture [1] 3 1 1.05 0.86
Seminar [1] 3 1 0.62 0.62

[1] Y. Hoshen, G. Ben-Artzi, and S. Peleg. Wisdom of the crowd in egocentric video curation. In IEEE Conference on Computer Vision and Pattern Recognition Workshops, pages 587-593.

Other works about RECfusion:

F. L.M. Milotta, S. Battiato, F. Stanco, V. D'Amico, G. Torrisi, L. Addesso, "RECfusion: Automatic Scene Clustering and Tracking in Videos from Multiple Sources", Proceedings of EI - Mobile Devices and Multimedia: Enabling Technologies, Algorithms, and Applications 2016. (webpage)