Projects on video
Event retrieval and temporal alignment
I work on video retrieval and temporal alignment, based on CTE (circulant temporal encoding), with refinements in the quantization method and the query expansion. The methods are described in these
CVPR 2013 and
ICCV 2013 papers. The temporal alignment method is introduced in this
arXiv 2015 preprint.
In this context we gathered the EVVE and Climbing datasets.
Video structuring
With
Danila Potapov, we work on video summarization. See our
ECCV 2014 paper on the subject, and this
project page that includes the MED-Summaries dataset and code for the Kernel Temporal Segmentation of Zaid Harchaoui.
Video classification
I work on video classification in the context of
Trecvid multimedia event detection in 2011-2014, together with great talented teams (with among others
Dan Oneata,
Danila Potapov,
Jérôme Revaud,
Mattis Paulin,
Heng Wang,
Lori Lamel). We got excellent results in 2012-2013, see our notebook papers in
2012,
2013,
2014.
Video indexing
For video indexing, I co-developed methods based on local descriptor matching with Hamming Embedding and on global descriptors. The methods were evaluated in the Content-based Copy Detection track of Trecvid in 2008-2010. The methods are described the
2008 and
2010 notebook papers, this
Transactions on Multimedia article and in this
ECCV 2010 paper. I also developed a
video distortion generator that was used in by NIST to generate test data for Trecvid CCD 2010.
