µ-Workshop on Computer Vision

October 2nd 2014, Rennes, France


Thursday, October 2nd 2014

  8:50 Registration
  9:20 -   9:30 Introduction
  9:30 - 10:15 Jiri Matas , Czech Technical University Prague Short-term Model-Free Causal Tracking - the VOT Challenge [abstract] [slides] [video]

The Visual Object Tracking Challenge is a benchmarking activity for a certain class of trackers. The talk will introduce the challenge, its semi-automatic selection and annotation of test data and the performance metrics. We will review the results for 2013 and 2014, summarize the lessons learned and discuss changes considered for 2015.

10:15 - 11:00 Patrick Pérez, Technicolor Face2Face: Learning Metrics to Compare Faces [slides] [video]
11:00 - 11:30 Coffee break
11:30 - 12:15 Andrew Zisserman, University of Oxford Time is On My Side [abstract] [slides]

The talk will cover two aspects of encoding time in video: first, to discern the temporal direction of the video, and second, to discriminate human actions in video. For the second aspect we employ a deep Convolutional Network trained on multi-frame dense optical flow, and evaluate on the standard video human actions benchmarks of UCF-101 and HMDB-51.
Work with William Freeman, Zheng Pan, Lyndsey Pickup, Bernhard Scholkopf, YiChang Shih, Karen Simonyan, Donglai Wei, Changshui Zhang .

12:15 - 13:00 Albert Gordo, Xerox Research Center Europe Towards Text Understanding: Word Image Representation, Matching and Recognition [slides] [video]
13:00 - 14:00 Lunch
14:00 - 14:40 Yannis Avrithis, National Technical University of Athens Image Retrieval, Vector Quantization and Nearest Neighbor Search [abstract] [slides] [video]

The first part of this talk considers a family of metrics to compare images based on their local descriptors. It encompasses the VLAD descriptor and matching techniques such as Hamming embedding. Making the bridge between these approaches yields a match kernel that takes the best of existing techniques by combining an aggregation procedure with a selective match kernel.
Since image search using either local or global descriptors boils down to approximate nearest neighbor search, the second part of this talk considers this problem, focusing on vector quantization methods. A recent method is presented whereby residuals over a coarse quantizer are used to locally optimize an individual product quantizer per cell. Non-exhaustive search strategies are discussed, including an inverted multi-index.

14:40 - 15:20 Philippe-Henri Gosselin, ENSEA Cergy Scalable Learning and Indexing for Retrieval in Large Image Databases [slides] [video]
15:20 - 15:50 The Fire-Id project
15:50 - 16:20 Bye Bye Coffee

Image gallery


INRIA, Campus Universitaire de Beaulieu - Salle Métivier
35042 Rennes Cedex

Additional information and directions to the venue can be found here.


Registration deadline: Thursday September 25th 2014

There is no registration fee for this workshop. However, all participants should register prior to the event by filling this form.



Frédéric Jurie will give a talk one day before the workshop on Wednesday, October 1st 2014 between 14:00 - 15:00 in salle Métivier at Inria Rennes. The subject of his talk is Histograms of Pattern Sets for Image Classification, Object Recognition and Image re-Ranking [abstract]

This talk will summarize the work we presented in two recent papers [a,b]. We have introduced a a novel image representation capturing feature dependencies through the mining of meaningful combinations of visual features. This representation leads to a compact and discriminative encoding of images that can be used for image classification, object detection, object recognition or even image re-ranking. The method relies on (i) multiple random projections of the input space followed by local binarization of projected histograms encoded as sets of items, and (ii) the representation of images as Histograms of Pattern Sets (HoPS). The approach is validated on four publicly available datasets (Daimler Pedestrian Classification, Oxford flowers Classification, KTH Texture Categorization, PASCAL VOC2007), allowing comparisons with many recent approaches. The proposed image representation reaches state-of-the-art performance on each of these datasets. We will also present an efficient framework for image re-reanking based on the same image representation.

[a] Histograms of Pattern Sets for Image Classification and Object Recognition Winn Voravuthikunchai, Bruno Cremilleux and Frederic Jurie, IEEE Conference on Computer Vision and Pattern Recognition, 2014.
[b] Image re-ranking based on statistics of frequent patterns, Winn Voravuthikunchai, Bruno Cremilleux and Frederic Jurie, ACM International Conference on Multimedia Retrieval, 2014.

HDR defense

Hervé Jégou will defend his HDR thesis On visual recognition and similarity search on Wednesday, October 1st 2014 at 15:30 in salle Métivier at Inria Rennes. [video]

Registration is not required for these presentations.


Fire-Id Project


Hervé Jégou

Email: Herve dot Jegou at inria dot fr

Andrei Bursuc

Email: Andrei dot Bursuc at inria dot fr