TRECVID 2007 Rushes Summarization

Feng Wang and Chong-Wah Ngo

Video Retrieval Group (VIREO), City University of Hong Kong
{fwang, cwngo}



Rushes are the raw material (extra video, B-rolls footage) use to produce a video. 20 to 40 times as much material may be shot as actually becomes part of the finished product. 

The system task in rushes summarization is, given a video from the rushes test collection, to automatically create an MPEG-1 summary clip that shows the main object (animate and inanimate) and events in the rushes video to be summarized. The summary should maximize the number of frames used and present the information in way that maximizes the usability of the summary and speed of object/event recognition.

System Framework



Performance Evaluation

Seven criterions are used by TRECVID 2007 to evaluate 2 baselines made by CMU, and the 22 submissions from USA, Europe, Australia and Asia.

  1. DU - duration of the summary (sec)

  2. XD - difference between target (4% of original video size) and actual summary size (target-actual) (sec)

  3. TT - total time spent judging the inclusions of objects and events (sec)

  4. VT - total video play time (vs. pause) to judge the inclusions (sec)

  5. IN  - fraction of inclusions found in the summary (0 - 1)

  6. EA - Was the summary easy to understand: 1 strongly disagree - 5 strongly agree

  7. RE - Was there a lot of duplicate video: 1 strongly agree - 5 strongly disagree


Criterion EA RE IN DU XD TT VT
Baseline 1 3.44 3.52 0.59 62.11 -2.25 105.66 60.88
Baseline 2 3.41 3.59 0.58 60.84 -0.97 100.48 58.68
Mean of 22 teams 3.16 3.66 0.48 49.54 10.33 92.27 51.14
Our result 3.60 3.94 0.64 41.11 18.75 90.58 44.83
Our ranking 1 2 3 4 4 9 6



Demo Page:



Feng Wang (Dept. of Computer Science, CityU of HK)

Prof. C. W. Ngo (Dept. of Computer Science, CityU of HK)



TVS Workshop at ACM Multimedia 2007

VIREO Research Group at CityU of HK


Related Publications

  1. F. Wang and C. W. Ngo, "Rushes Video Summarization by Object and Event Understanding", TRECVID BBC Rushes Summarization Workshop at ACM Multimedia '07.

  2. C. W. Ngo, Z. Pan, and X. Y. Wei, "Hierarchical Hidden Markov Model for Rushes Structuring and Indexing", Int. Conf. on Image and Video Retrieval, 2006.

  3. C. W. Ngo, Z. Pan, X. Wei, X. Wu, H. K. Tan, and W. Zhao, "Motion Driven Approaches to Shot Boundary Detection, Low-Level Feature Extraction and BBC Rushes Characterization at TRECVID 2005", TRECVID Workshop, 2005. 

  4. C. W. Ngo, T. C. Pong, and R. T. Chin, "Video Partitioning through Temporal Slices Analysis", IEEE Trans. on Circuits and Systems for Video Technology, 11(8), pp. 941-953, 2001.

Last Updated: 19-03-2008