Image indexing and retrieval with Yael



Yael is a library implementing computationally intensive functions used in large scale image retrieval, such as neighbor search, clustering and inverted files. The library offers interfaces for C, Python and Matlab.

The motivation of Yael is twofold. We aim at providing: 

  • core and optimized instructions and methods commonly used for large-scale multimedia retrieval systems 
  • more sophisticated functions associated with state-of-the-art methods, such as the Fisher vector, VLAD, Hamming Embedding or more generally methods based on inverted file systems, such as selective match kernels.

Yael is intended as an API and does not implement a retrieval system in an integrated manner: only a few test programs are available for key tasks such as k-means. Yet this can be done on top of it with a few dozen lines of Matlab or Python code.

Yael started as an open-source spin-off of INRIA LEAR‘s proprietary library Bigimbaz. The objective was to isolate performance-critical primitives that could be re-used in other projects. Yael’s design choices were: implemented in C for simplicity, but using an object-oriented design (structs with constructors/destructors), interface with Python as high-level language to facilitate administrative tasks. 

Yael is designed to handle dense data in float, as it is primarily used for signal processing tasks where the quality of the representation is determined by the number of dimensions rather than the precision of the components. In the Matlab interface, single matrices, and float32 in Python. Yael was designed initially to manipulate matrices in C. It was interfaced for Python using SWIG, which gives low-level access to the full library. An additional Numpy layer (ynumpy) is provided for high-level functions. The most important functions of Yael are wrapped in Mex to be callable from Matlab.

Performance is very important. Yael has computed k-means with hundreds of thousand centroids and routinely manipulate matrices that occupy more than 1/2 the machine’s RAM. This means that it has to be lightweight and 64-bit clean. The design choices of Yael are governed by efficiency concerns more than by portability. As a result, the library may work only with severely down-graded performance if instructions are not provided by the processor. In particular, Yael relies on SSE instructions such as the SSE 4.2 popcnt instruction. The library is maintained for Linux and MacOS. Yael relies on as few external libraries as possible. The only mandatory ones are BLAS/Lapack (for performance). Other libraries (Python’s C interface, Matlab’s mex, Arpack, OpenMP) are optional.

Yael and related packages are downloaded around 600 times per month. 

This article addresses the recognition of images of the same scene or object, and how Yael can perform this kind of operation. Here is an example of two images of the same scene that we would like to match:

127300 127301


We will explain how to compute descriptors (aka signatures) for the images, and how to find descriptors that are similar between images.

We are going to work on the 100 first query images of the Holidays dataset, and their associated database examples. The images and associated SIFT descriptors can be downloaded from here: Images and SIFT descriptors.

Image indexing

Imagine a user that has a large image collection with photos of buildings, with as associated metadata the GPS location of the building. Given a new photo of a building, taken with a mobile phone, the user wants to find the location where the photo was taken. This is where image indexing comes into play.

Image indexing means constructing an index referencing the images from a collection. This index has a search function that can be used to retrieve the images that are most similar to a query image. 

At build time and search time, the index is stored in RAM. This is orders of magnitude faster than disk-based implementations, such as those used in SQL database engines. However, for large datasets, this requires either a lot of RAM or a very compact representation per image. Yael provides this compact representation, so that you do not need to buy the RAM.

In combination with efficient matrix manipulation environments like Matlab and Numpy, Yael makes the process of building an index and searching in it very simple. 

Extracting image descriptors

Local image descriptors are vectors computed each on an area of the image. The areas are selected to contain strong contrast changes, with a 2D signal processing filter. Then the descriptor vector is computed from the gradient or frequency content in the area.

Local descriptors are typically designed to be invariant to some classes of transformations: translations, illumination changes, rotations, etc. At the same time, they should be discriminant enough to distinguish relevant differences on the patches, eg. different patterns on the facade of a building. There is a long line of research on designing local image features with appropriate tradeoffs in terms of invariance / discriminance / computational cost, see for example this comparison of affine covariant features.

In the images above, local descriptors extracted on the skyline ought to be very similar. Therefore, these images should be easy to match.

Local descriptors can be extracted using any local description algorithm, as long as they can be compared with L2 distances, ie. descriptors that are far away in L2 space are also considered different in image content. For example, OpenCV provides an implementation of the SURF descriptor, and VLFeat contains a SIFT implementation. 

For this example, we will use the SIFT implementation provided along with the Holidays dataset. In the “Descriptor extraction” section of, download the executable (there is a Mac OS X version and a Linux version). 

The pre-processing applied to images before analyzing them to extract signatures can have a dramatic effect on the retrieval performance. Ideally, images should be equalized so that their luminance is similar and resized into dimensions that are not too different. This can be performed in a number of ways, eg. with Imagemagick. In our case, we’ll just use a few command-line utilities from netpbm

In total, the steps that extract the descriptors from a single image are:


# Rescaling and intensity normalization
djpeg $infile | ppmtopgm | pnmnorm -bpercent=0.01 -wpercent=0.01 -maxexpand=400 | pamscale -pixels $[1024*768] > $tmpfile

# Compute descriptors
compute_descriptors -i $tmpfile -o4 $outfile -hesaff -sift 

This should be applied to all the images that are to be indexed, and the ones that will be queried. 

The remainder of this article presents the main functions used in Yael to do image retrieval. They are implemented in the two languages supported by Yael: Python and Matlab. 

Image indexing in Python with Fisher vectors

A global image descriptor is a vector that characterizes the whole image. The Euclidean distance between the descriptors of two images should be higher for different images than for similar images. There are many popular types of global descriptors, like color histograms or GIST descriptors.

Here, we use a statistical tool derived from the Fisher kernel to aggregate the local SIFT descriptors of an image into a global image descriptor: the Fisher vector (FV). See Aggregating local image descriptors into compact codes for more details. You may also be interested in INRIA’s Fisher vector implementation which is a Matlab version of this example, on the complete Holidays dataset.

The most important functions of Yael are available in Python via the ynumpy module. They all manipulate c-compact float32 or int32 matrices. 

The FV computation relies on a training where a Gaussian Mixture Model (GMM) is fitted to a set of representative local descriptors. For simplicity, we are going to use the descriptors of the database we index. To load the database descriptors, use the ynumpy.siftgeo_read function:

for imname in image_names:
    desc, meta = ynumpy.siftgeo_read(imname)

The meta component contains the SIFT descriptor’s meta-information (location and size of the area, orientation, etc.). We do not use this information to compute the FV.

Next we sample the descriptors to reduce their dimensionality by PCA and computing a GMM. This involves some standard numpy code, and the ynumpy.gmm_learn function. For a GMM of size k (let’s set it to 64), we need about 1000*k training descriptors

k = 64
n_sample = k * 1000

# choose n_sample descriptors at random
sample_indices = np.random.choice(all_desc.shape[0], n_sample)
sample = all_desc[sample_indices]

# train GMM
gmm = ynumpy.gmm_learn(sample, k)

The GMM is a tuple containing the a-priori weights per mixture component, the mixture centres and the diagonal of the component covariance matrices (the model assumes a diagonal matrix, otherwise the descriptor would be way too long).

The training is finished. The next stage is to encode the SIFTs into one vector per image: 

image_fvs = []
for image_desc in image_descs:
   # compute the Fisher vector, using only the derivative w.r.t mu
   fv = ynumpy.fisher(gmm, image_desc, include = 'mu')

All the database descriptors are stacked as lines of a single matrix image_fvs, and all queries image descriptors in another matrix query_fvs. Then the Euclidean nearest neighbors of each query (and hence the most similar images) can be retrieved with:

# get the 8 NNs for all query images in the image_fvs array
results, distances = ynumpy.knn(query_fvs, image_fvs, nnn = 8)

Now we display the search results for a few query images. There is one line per query image, which shows the image, and a row of retrieval results. The correct results have a green rectangle around them, negative ones a red rectangle. 


Note that the query image always appears as the first retrieval result, because it is included in the dataset.

Image indexing based on global descriptors like the Fisher Vector is very efficient and easy to implement using Yael. For larger datasets (more than a few tens of thousand images), it is useful to use vector quantization or hashing techniques to perform the nearest-neighbor search faster. 

Image indexing in Matlab with inverted files

In this chapter, we directly index all the local SIFT descriptors of the database images into an indexing structure in RAM called the inverted file. Each SIFT descriptor is assigned an index in [1,k] using a quantization function. The inverted file contains k lists, one per possible index. When a SIFT from an image is assigned to an index 1 ≤ i ≤ k, the id of this image is added to the list i.

In the example below, we show how to use an inverted file of Yael from Matlab. More specifically, the inverted file we consider supports binary signatures, as proposed in the Hamming Embedding approach described in this paper.

Before launching the code, please ensure that

  • You have a working and compiled version of Yael’s matlab interface
  • The corresponding directory (‘YAELDIR/matlab’) is in your matlab Path. If not, use the addpath(‘YAELDIR/matlab’) to add it.

To start with, we define the parameters of the indexing method. Here, we choose a vocabulary of size k=1024. We also set some parameters specific to Hamming embedding.

k = 1024;                            % Vocabulary size
dir_data = './holidays_100/';        % data directory

% Parameters For Hamming Embedding
nbits = 128;                         % Typical values are 32, 64 or 128 bits
ht = floor(nbits*24/64);             % Hamming Embedding threshold

Hereafter, we show how we typically load a set of images and descriptors stored in separate files. We use the standard matlab functions arrayfun and cellfun to perform operations in batch. The descriptors are assumed stored in the siftgeo format, therefore we read them with the yael ‘siftgeo_read’ function.

sifts = cell(); 

for i = 1:numel(img_list)
  [sifts_i, meta] = siftgeo_read(img_list{i}); 
  sifts{i} = sifts_i; 

Now, we are going to learn the visual vocabulary with k-means and subsequently construct the inverted file structure for Hamming Embedding. We learn it on Holidays itself to avoid requiring another dataset. But note that this should be avoided for a true system, and a proper evaluation should employ an external dataset for dictionary learning.

vtrain = [sifts{:}];
vtrain = vtrain (:, 1:2:end); tic

C = yael_kmeans (vtrain, k, 'niter', 10);

% We provide the codebook and the function that performs the assignment,
% here it is the exact nearest neighbor function yael_nn

ivfhe = yael_ivf_he (k, nbits, vtrain, @yael_nn, C);

We can add the descriptors of all the database images to the inverted file. Here, Each local descriptor receives an identifier. This is not a requirement: another possible choice would be to use directly the id of the image. But in this case we could not use this output for spatial verification. In our case, the descriptor id will be used to display the matches.

descid_to_imgid = zeros (totsifts, 1);  % desc to image conversion
imgid_to_descid = zeros (nimg, 1);      % for finding desc id
lastid = 0;

for i = 1:nimg
  ndes = nsifts(i);  % number of descriptors

  % Add the descriptors to the inverted file.
  % The function returns the visual words (and binary signatures),
  [vw,bits] = ivfhe.add (ivfhe, lastid+(1:ndes), sifts{i});
  imnorms(i) = norm(hist(vw,1:k));

  descid_to_imgid(lastid+(1:ndes)) = i;
  imgid_to_descid(i) = lastid;
  lastid = lastid + ndes;

Finally, we make some queries. We compute the number of matches n_immatches between query and database images. We invoke the standard Matlab function accumarray, which in essence compute here a histogram weighted by the match weights.

Queries = [1 13 23 42 63 83];
for q = 1:numel(Queries)
  qimg = Queries(q)

  matches = ivfhe.query (ivfhe, int32(1:nsifts(qimg)), sifts{qimg}, ht);

  % Translate to image identifiers and count number of matches per image, 
  m_imids = descid_to_imgid(matches(2,:));
  n_immatches = hist (m_imids, 1:nimg);

  % Images are ordered by descreasing score 
  [~, idx] = sort (n_immatches, 'descend');

  % Display results 

The output looks as follows. The query is the top-left image, and then the queries are displayed. The title gives the number of matches and the normalized score used to rank the images. The matches are displayed in yellow (and the non-matching descriptors in red).



Yael is a small library that contains many primitives that are useful for image indexing, nearest-neighbor search, sorting, etc. It at the base of several state-of-the-art implementations of image indexing packages. Reference [1] describes the implementation tradeoffs of some of Yael’s main functions, and provides more references to research papers whose results were obtained with Yael.

In the code above, only the main function calls were shown, see the Yael tutorial for a fully functional version of the code, and the main Yael website for the complete documentation. 


TOMM Associate Editor of the Year Award 2015


Annually, the Editor-in-Chief of the ACM Transactions on Multimedia Computing, Communications and Applications (TOMM) honors one member of the Editorial Board with the TOMM Associate Editor of the Year Award. The purpose of the award is the distinction of excellent work for ACM TOMM and hence also for the whole multimedia community in the previous year. Criteria for the award are (1.) the amount of submissions processed in time, (2.) the performance during the reviewing process and (3.) the accurate interaction with the reviewers in order to broader the awareness for the journal. Based on the criteria mentioned above, the ACM Transactions on Multimedia Computing, Communications and Applications Associate Editor of the Year Award 2015 goes to Pradeep Atrey from State University of New York, Albany, USA. pradeep-atreyPradeep K. Atrey is an Assistant Professor at the State University of New York, Albany, NY, USA. He is also an (on-leave) Associate Professor at the University of Winnipeg, Canada and an Adjunct Professor at University of Ottawa, Canada. He received his Ph.D. in Computer Science from the National University of Singapore, M.S. in Software Systems and B.Tech. in Computer Science and Engineering from India. He was a Postdoctoral Researcher at the Multimedia Communications Research Laboratory, University of Ottawa, Canada. His current research interests are in the area of Security and Privacy with a focus on multimedia surveillance and privacy, multimedia security, secure-domain cloud-based large-scale multimedia analytics, and social media. He has authored/co-authored over 100 research articles at reputed ACM, IEEE, and Springer journals and conferences. His research has been funded by Canadian Govt. agencies NSERC and DFAIT, and by Govt. of Saudi Arabia. Dr. Atrey is on the editorial board of several journals including ACM Trans. on Multimedia Computing, Communications and Applications, ETRI Journal and IEEE Communications Society Review Letters. He was also guest editor for Springer Multimedia Systems and Multimedia Tools and Applications journals. He has been associated with over 40 international conferences/workshops in various roles such as Organizing Chair, Program Chair, Publicity Chair, Web Chair, Area Chair, Demo Chair and TPC Member. Dr. Atrey was a recipient of the Erica and Arnold Rogers Award for Excellence in Research and Scholarship (2014), ETRI Journal Best Editor Award (2012), ETRI Journal Best Reviewer Award (2009) and the three University of Winnipeg Merit Awards for Exceptional Performance (2010, 2012 and 2013). He was also recognized as “ICME 2011 – Quality Reviewer” and is invited as a Rising Star Speaker at the SIGMM Inaugural Multimedia Frontier Workshop (2015). The Editor-in-Chief Prof. Dr.-Ing. Ralf Steinmetz cordially congratulates Pradeep.

2015 ACM Transactions on Multimedia Computing, Communications and Applications (TOMM) Nicolas D. Georganas Best Paper Award


The 2015 ACM Transactions on Multimedia Computing, Communications and Applications (TOMM) Nicolas D. Georganas Best Paper Award is provided to the paper “A Quality of Experience Model for Haptic Virtual Environments” (TOMM vol.10, Issue 3) by Abdelwahab Hamam, Abdulmotaleb El Saddik and Jihad Alja’am.

The purpose of the named award is to recognize the most significant work in ACM TOMM (formerly TOMCCAP) in a given calendar year. The whole readership of ACM TOMM was invited to nominate articles which were published in Volume 10 (2014). Based on the nominations the winner has been chosen by the TOMM Editorial Board. The main assessment criteria have been quality, novelty, timeliness, clarity of presentation, in addition to relevance to multimedia computing, communications, and applications.

The winning paper is grounded on the observation that so far there is only limited research on Quality of Experience (QoE) for Haptic-based Virtual Reality applications. In order to overcome this issue, the authors propose a human-centric taxonomy for the evaluation of QoE for haptic virtual environments. The QoE evaluation is applied through a fuzzy logic inference model. The taxonomy also gives guidelines for the evaluation of other multi-modal multimedia systems. This multi-modality was one of the main reasons for the selection of this article and TOMM members expect that it will have an impact on future QoE studies in various sub-fields of multimedia research.

The award honors the founding Editor-in-Chief of TOMM, Nicolas D. Georganas, for his outstanding contributions to the field of multimedia computing and his significant contributions to ACM. He exceedingly influenced the research and the whole multimedia community.

The Editor-in-Chief Prof. Dr.-Ing. Ralf Steinmetz and the Editorial Board of ACM TOMM cordially congratulate the winner. The award will be presented to the authors at the ACM Multimedia 2015 in Brisbane, Australia, and includes travel expenses for the winning authors.

abdelwahab-hamamAbdelwahab Hamam received his PhD in Electrical and Computer Engineering from the University of Ottawa, Canada, in 2013. He is currently a postdoctoral research scientist at Immersion in Montreal, Canada focusing in research and development of novel haptic technologies and interactions. He was previously a teaching and research assistant at the University of Ottawa from Jan 2005 to May 2013. He has more than 35 academic papers and pending patent applications. He is the recipient of the best paper award at the 2015 QoMEX workshop. He is the technical co-chair of the 2015 Haptic Audio-Visual Environments and Games (HAVE) Workshop and the co-organizer of the 2015 QoMEX workshop special session on quality of experience in haptics. His research interests include haptic applications, medical simulations, and quality of experience for multimedia haptics.

abed-el-saddikAbdulmotaleb El Saddik is Distinguished University Professor and University Research Chair in the School of Electrical Engineering and Computer Science at the University of Ottawa. He is an internationally-recognized scholar who has made strong contributions to the knowledge and understanding of multimedia computing, communications and applications. He has authored and co-authored four books and more than 450 publications. Chaired more than 50 conferences and workshop and has received research grants and contracts totaling more than $18 Mio. He has supervised more than 100 researchers. He received several international awards, among others ACM Distinguished Scientist, Fellow of the Engineering Institute of Canada, Fellow of the Canadian Academy of Engineers and Fellow of IEEE and IEEE Canada Computer Medal.

jihad-mohamed-aljaamJihad Mohamed Alja’am received the Ph.D. degree, MS. degree and BSc degree in computing from Southern University (The National Council for Scientific Research, CNRS), France. He was with IBM-Paris as Project Manager and with RTS-France as IT Consultant for several years. He is currently with the Department of Computer Science and Engineering at Qatar University. His current research interests include multimedia, assistive technology, learning systems, human–computer interaction, stochastic algorithms, artificial intelligence, information retrieval, and natural language processing. Dr. Alja’am is a member of the editorial boards of the Journal of Soft Computing, American Journal of Applied Sciences, Journal of Computing and Information Sciences, Journal of Computing and Information Technology, and Journal of Emerging Technologies in Web Intelligence. He acted as a scientific committee member of different international conferences (ACIT, SETIT, ICTTA, ACTEA, ICLAN, ICCCE, MESM, ICENCO, GMAG, CGIV, ICICS, and ICOST). He is a regular reviewer for the ACM computing review and the journal of supercomputing. He has collaborated with different researchers in Canada, France, Malaysia, and USA. He published so far 138 papers, 8 books chapters in computing and information technology which are published in conference proceedings, scientific books, and international journals. He is leading a research team in multimedia and assistive technology and collaborating in the Financial Watch and Intelligent Document Management System for Automatic Writer Identification projects.

ACM SIGMM Award for Outstanding PhD Thesis in Multimedia Computing, Communications and Applications 2015



ting-yaoACM Special Interest Group on Multimedia (SIGMM) is pleased to present the 2015 SIGMM Outstanding Ph.D. Thesis Award to Dr. Ting Yao and Honorable Mention recognition to Dr. Britta Meixner.

The award committee considers Dr. Yao’s dissertation entitled “Multimedia Search by Self, External, and Crowdsourcing Knowledge” worthy of the recognition as the thesis proposes an innovative knowledge transfer framework for multimedia search which is expected to have significant impact, especially in boosting the search performance for big multimedia data.

Dr. Yao’s thesis proposes the knowledge transfer methodology in three multimedia search scenarios:

  1. Seeking consensus among multiple modalities in the context of search re-ranking,
  2. Leveraging external knowledge as a prior to be transferred to a problem that belongs to a domain different from the external knowledge, and
  3. Exploring the large user click-through data as crowdsourced human intelligence for annotation and search.

The effectiveness of the proposed framework has been successfully justified by thorough experiments. The proposed framework has substantial contributions in principled integration of multimodal data which is indispensable in multimedia search. The publications related to the thesis clearly demonstrate the major impact of this work in many research disciplines including multimedia, web, and information retrieval. The fact that parts of the proposed techniques have been and are being transferred to the commercial search service Bing further attest to the practical contributions of this thesis. Overall, the committee recognizes the significant impact and contributions presented in the thesis to the multimedia community.

Bio of Awardee

Dr. Ting Yao is an associate researcher in the Multimedia Search and Mining group at Microsoft Research, Beijing, China. His research interests are in multimedia search and computing. He completed a Ph.D. in Computer Science at City University of Hong Kong in 2014. He received the B.Sc. degree in theoretical and applied mechanics (2004), B.Eng. double degree in electronic information engineering (2004), and M.Eng. degree in signal and information processing (2008) all from the University of Science and Technology of China, Hefei, China. The system designed by him achieved the second place in the THUMOS action recognition challenge at CVPR 2015. He was also the principal designer of the image retrieval systems that achieved the third and fifth performance in the MSR-Bing image retrieval challenge at ACM MM 2014 and 2013, respectively. He received the Best Paper Award of ACM ICIMCS (2013).

Honorable Mention

britta-meixnerThe award committee is pleased to present the Honorable Mention to Dr. Britta Meixner for the thesis entitled: “Annotated Interactive Non-linear Video – Software Suite, Download and Cache Management.”

The thesis presents a fully functional software suite for authoring non-linear interactive videos with downloading and cache management mechanisms for effective video playback. The committee is significantly impressed by the thorough study presented in the thesis with extensive analysis of the properties of the software suite. The implementation which has been made available as open source software along with the thesis undoubtedly has very high potential impact to the multimedia community.

Bio of Awardee

Dr. Britta Meixner received her Master’s degree (German Diplom) in Computer Science from the University of Passau, Germany, in 2008. Furthermore, she received the First State Examination for Lectureship at Secondary Schools for the subjects Computer Science and Mathematics from the Bavarian State Ministry for Education and Culture in 2008. She received her Ph.D. degree from the University of Passau, Germany, in 2014. The title of her thesis is “Annotated Interactive Non-linear Video – Software Suite, Download and Cache Management.” She is currently a postdoctoral research fellow with the University of Passau, Germany, and will be a postdoctoral research fellow at FXPAL, Palo Alto, CA, USA, starting October 2015. Her research interest is mainly in hypermedia. She is an award winner of the 2015 Award “Women + Media Technology,” granted by Germany’s public broadcasters ARD and ZDF (ARD/ZDF Förderpreis “Frauen + Medientechnologie” 2015). She was a Reviewer for Springer Multimedia Tools and Applications (MTAP) Journal, an Organizer of the “International Workshop on Interactive Content Consumption (WSICC)” at ACM TVX in 2014 and 2015, and Associate Chair at ACM TVX2015.

Announcement of ACM SIGMM Rising Star Award 2015


yu-gang-jiangACM Special Interest Group on Multimedia (SIGMM) is pleased to present this year’s Rising Star Award in multimedia computing, communications and applications to Dr. Yu-Gang Jiang. The ACM SIGMM Rising Star Award recognizes a young researcher who has made outstanding research contributions to the field of multimedia computing, communication and applications during the early part of his or her career. Dr. Yu-Gang Jiang has made fundamental contributions in the area of video analysis and retrieval, especially with innovative approaches to large-scale video concept detection. He has been an active leader in exploring the bag-of-visual-words (BoW) representation for concept detection, providing influential insights on the critical representation design. He proposed the important idea of “soft-weighting” in his CIVR 2007paper, which significantly advanced the performance of visual concept detection. Dr. Jiang has proposed several important techniques for video and image search. In 2009, he proposed a novel domain adaptive concept selection method for concept-based video search. His method selects the most relevant concepts for a given query considering not only the semantic concept-to-query relatedness but also the data distribution in the target domain. Recently he proposed a method that generates query-adaptive hash codes for improved visual search, with which a finer-grained ranking of search results can be achieved compared to the traditional hashing based methods. His most recent work is in the emerging field of video content recognition by deep learning, where he proposed a comprehensive deep learning framework to model static, short-term motion and long-term temporal information in videos. Very promising results were obtained on the widely used UCF101 dataset. As a postdoctoral researcher at Columbia University and later as a faculty member at Fudan University, Dr. Jiang has devoted significant efforts to video event recognition, a problem that is receiving increasing attention in the multimedia community. His extensive contributions in this area include not only innovative algorithm design, but also large benchmark construction, system development, and survey tutorials. He devised a comprehensive system in 2010 using multimodal features, contextual concepts and temporal clues, which won the multimedia event detection (MED) task in NIST TRECVID 2010. He constructed the Columbia Consumer Video (CCV) benchmark in 2011, which has been widely used. Recently, he continues to lead major efforts in creating and sharing large-scale video datasets in critical areas (including 200+ event categories and 100,000 partially copy videos) as community resources. The high impact of his works is reflected by the high number of citations of his work. His recent paper on video search result organization received the Best Poster Paper Award at ACMMM 2014. His shared benchmark datasets and source codes have been used worldwide. In addition, he has made extensive contributions to the professional communities by serving as conference program chairs, invited speakers, and tutorial experts. In summary, Dr. Yu-Gang Jiang receives the 2015 ACM SIGMM Rising Star Award for his significant contributions in the areas of video content recognition and search.

ACM SIGMM Award for Outstanding Technical Contributions to Multimedia Computing, Communications and Applications


tatsengchuaThe 2015 winner of the prestigious ACM Special Interest Group on Multimedia (SIGMM) award for Outstanding Technical Contributions to Multimedia Computing, Communications and Applications is Prof. Dr. Tat-Seng Chua. The award is given in recognition of his pioneering contributions to multimedia, text and social media processing. Tat-Seng Chua is a leading researcher in multimedia, text and social media analysis and retrieval. He is one of the few researchers who has made substantial contributions in the fields of multimedia, information retrieval and social media. Dr. Chua’s contributions in multimedia dates back to the early 1990s, where he was among the first to work on image retrieval with relevance feedback (1991), video retrieval and sequencing by exploring metadata and cinematic rules (1995), and fine grained image retrieval at segment level (1995). These works helped shape the development of the field for many years. Given the limitation of visual content analysis, his research advocates the integration of text, metadata and visual contents coupled with domain knowledge for large-scale media analysis. He developed a multi-source, multi-modal and multi-resolution framework together with the involvement of human in the loop for such analysis and retrieval tasks. This has helped his group not only publish papers in top conferences and journals, but also achieve top positions in large-scale video evaluations when his group participated in TRECVID in 2000-2006, VideOlympics in 2007-09, as well as winning the highly competitive Star (Multimedia) Challenge in 2008. Leveraging the experience, he developed a large-scale multi-label image test set named NUS-WIDE, which has been widely used with over 600 citations. He recently started a company named ViSenze Pte Ltd ( to commercialize his research in mobile visual fashion search. In his more recent research work in multimedia question-answering (MMQA), he developed a joint text-visual model to exploit correlation between text queries, text-based answers, and visual concepts in images and videos to return both relevant text and video answers. The early work was carried out in the domain of news video (2003), which has motivated several follow-on works in image QA. His recent works tackled the more complicated “how-to” type QA in product domains (2010-13). His recent works (2013-14) exploited SemanticNet to perform attribute-based image retrieval and use of various types of domain knowledge. His current work aims to build a live, continuous-learning system to support the dynamic annotation and retrieval of images and micro videos in social media streams. In information retrieval and social media research, Dr. Chua focused on the key problems of organizing large-scale unstructured text contents to support question-answering (QA). His works point towards the use of linguistics and domain knowledge for effective large-scale information analysis, organization and retrieval. Given his strong interest in both multimedia and text processing, it is natural for him to venture into social media research that involves the analysis of text, multimedia, and social network contents. His group developed a live social observatory system to carry out research in building descriptive, predictive and prescriptive analytics of multiple live social media streams. The system has been well recognized by peers. His recent work on “multi-screen social TV” won the 2015 Best IEEEE Multimedia Best paper Award. Dr. Chua has been involved in most key conferences in these areas by serving as general chair, technical program chair, or invited keynote speaker as well as by leading innovative research and winning many best paper or best student paper awards in recent years. He is the Steering Committee Chair of two international multimedia conference series: ACM ICMR (International Conference on Multimedia Retrieval) and MMM (MultiMedia Modeling). In summary, he is an extraordinarily accomplished and outstanding researcher in multimedia, text and social media processing, truly exemplifying the characteristics of the ACM SIGMM Award for Outstanding Technical Contributions.

ACM SIGMM/TOMM 2015 Award Announcements

The ACM Special Interest Group in Multimedia (SIGMM) and ACM Transactions on Multimedia Computing, Communications and Applications (TOMM) are pleased to announce the following awards for 2015 recognizing outstanding achievements and services made in the multimedia community.
SIGMM Technical Achievement Award:
Dr. Tat-Seng Chua, National University of Singapore

SIGMM Rising Star Award:
Dr. Yu-Gang Jiang, Fudan University
SIGMM Best Ph.D. Thesis Award:
Dr. Ting Yao, City University of Hong Kong (currently Microsoft Research)

TOMM Nicolas D. Georganas Best Paper Award:
“A Quality of Experience Model for Haptic Virtual Environments” by Abdelwahab Hamam, Abdulmotaleb El Saddik, and Jihad Alja’am, published in TOMM, vol. 10, Issue 3, 2014.
TOMM Best Associate Editor Award:
Dr. Pradeep K. Atrey, State University of New York, Albany
Additional information of each award and recipient is available on the SIGMM web site.
Awards will be presented in the annual SIGMM event, ACM Multimedia Conference, held in Brisbane, Australia during October 26-30, 2015.
ACM is the professional society of computer scientists, and SIGMM is the special interest group on multimedia. TOMCCAP is the flagship journal publication of SIGMM.

MPEG Column: 112th MPEG Meeting

This blog post is also available at at bitmovin tech blog and

The 112th MPEG meeting in Warsaw, Poland was a special meeting for me. It was my 50th MPEG meeting which roughly accumulates to one year of MPEG meetings (i.e., one year of my life I’ve spend in MPEG meetings incl. traveling – scary, isn’t it? … more on this in another blog post). But what happened at this 112th MPEG meeting (my 50th meeting)…

  • Requirements: CDVA, Future of Video Coding Standardization (no acronym yet), Genome compression
  • Systems: M2TS (ISO/IEC 13818-1:2015), DASH 3rd edition, Media Orchestration (no acronym yet), TRUFFLE
  • Video/JCT-VC/JCT-3D: MPEG-4 AVC, Future Video Coding, HDR, SCC
  • Audio: 3D audio
  • 3DG: PCC, MIoT, Wearable

MPEG Friday Plenary. Photo (c) Christian Timmerer.

As usual, the official press release and other publicly available documents can be found here. Let’s dig into the different subgroups:


In requirements experts were working on the Call for Proposals (CfP) for Compact Descriptors for Video Analysis (CDVA) including an evaluation framework. The evaluation framework includes 800-1000 objects (large objects like building facades, landmarks, etc.; small(er) objects like paintings, books, statues, etc.; scenes like interior scenes, natural scenes, multi-camera shots) and the evaluation of the responses should be conducted for the 114th meeting in San Diego.

The future of video coding standardization is currently happening in MPEG and shaping the way for the successor of of the HEVC standard. The current goal is providing (native) support for scalability (more than two spatial resolutions) and 30% compression gain for some applications (requiring a limited increase in decoder complexity) but actually preferred is 50% compression gain (at a significant increase of the encoder complexity). MPEG will hold a workshop at the next meeting in Geneva discussing specific compression techniques, objective (HDR) video quality metrics, and compression technologies for specific applications (e.g., multiple-stream representations, energy-saving encoders/decoders, games, drones). The current goal is having the International Standard for this new video coding standard around 2020.

MPEG has recently started a new project referred to as Genome Compression which is about of course about the compression of genome information. A big dataset has been collected and experts working on the Call for Evidence (CfE). The plan is holding a workshop at the next MPEG meeting in Geneva regarding prospect of Genome Compression and Storage Standardization targeting users, manufactures, service providers, technologists, etc.

Summer in Warsaw. Photo (c) Christian Timmerer.


The 5th edition of the MPEG-2 Systems standard has been published as ISO/IEC 13818-1:2015 on the 1st of July 2015 and is a consolidation of the 4th edition + Amendments 1-5.

In terms of MPEG-DASH, the draft text of ISO/IEC 23009-1 3rd edition comprising 2nd edition + COR 1 + AMD 1 + AMD 2 + AMD 3 + COR 2 is available for committee internal review. The expected publication date is scheduled for, most likely, 2016. Currently, MPEG-DASH includes a lot of activity in the following areas: spatial relationship description, generalized URL parameters, authentication, access control, multiple MPDs, full duplex protocols (aka HTTP/2 etc.), advanced and generalized HTTP feedback information, and various core experiments:

  • SAND (Sever and Network Assisted DASH)
  • FDH (Full Duplex DASH)
  • SAP-Independent Segment Signaling (SISSI)
  • URI Signing for DASH
  • Content Aggregation and Playback COntrol (CAPCO)

In particular, the core experiment process is very open as most work is conducted during the Ad hoc Group (AhG) period which is discussed on the publicly available MPEG-DASH reflector.

MPEG systems recently started an activity that is related to media orchestration which applies to capture as well as consumption and concerns scenarios with multiple sensors as well as multiple rendering devices, including one-to-many and many-to-one scenarios resulting in a worthwhile, customized experience.

Finally, the systems subgroup started an exploration activity regarding real-time streaming of file (a.k.a TRUFFLE) which should perform an gap analysis leading to extensions of the MPEG Media Transport (MMT) standard. However, some experts within MPEG concluded that most/all use cases identified within this activity could be actually solved with existing technology such as DASH. Thus, this activity may still need some discussions…


The MPEG video subgroup is working towards a new amendment for the MPEG-4 AVC standard covering resolutions up to 8K and higher frame rates for lower resolution. Interestingly, although MPEG most of the time is ahead of industry, 8K and high frame rate is already supported in browser environments (e.g., using bitdash 8K, HFR) and modern encoding platforms like bitcodin. However, it’s good that we finally have means for an interoperable signaling of this profile.

In terms of future video coding standardization, the video subgroup released a call for test material. Two sets of test sequences are already available and will be investigated regarding compression until next meeting.

After a successful call for evidence for High Dynamic Range (HDR), the technical work starts in the video subgroup with the goal to develop an architecture (“H2M”) as well as three core experiments (optimization without HEVC specification change, alternative reconstruction approaches, objective metrics).

The main topic of the JCT-VC was screen content coding (SCC) which came up with new coding tools that are better compressing content that is (fully or partially) computer generated leading to a significant improvement of compression, approx. or larger than 50% rate reduction for specific screen content.


The audio subgroup is mainly concentrating on 3D audio where they identified the need for intermediate bitrates between 3D audio phase 1 and 2. Currently, phase 1 identified 256, 512, 1200 kb/s whereas phase 2 focuses on 128, 96, 64, 48 kb/s. The broadcasting industry needs intermediate bitrates and, thus, phase 2 is extended to bitrates between 128 and 256 kb/s.


MPEG 3DG is working on point cloud compression (PCC) for which open source software has been identified. Additionally, there’re new activity in the area of Media Internet of Things (MIoT) and wearable computing (like glasses and watches) that could lead to new standards developed within MPEG. Therefore, stay tuned on these topics as they may shape your future.

The week after the MPEG meeting I met the MPEG convenor and the JPEG convenor again during ICME2015 in Torino but that’s another story…

L. Chiariglione, H. Hellwagner, T. Ebrahimi, C. Timmerer (from left to right) during ICME2015. Photo (c) T. Ebrahimi.

Call for Nominations: Editor-In-Chief of ACM TOMM

ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM) The term of the current Editor-in-Chief (EiC) of the ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM) ( is coming to an end, and the ACM Publications Board has set up a nominating committee to assist the Board in selecting the next EiC. Nominations, including self-nominations, are invited for a three-year term as TOMM EiC, beginning on 1 January 2016. The EiC appointment may be renewed at most one time. This is an entirely voluntary position, but ACM will provide appropriate administrative support. The EiC is responsible for maintaining the highest editorial quality, for setting technical direction of the papers published in TOMM, and for maintaining a reasonable pipeline of articles for publication. He/she has final say on acceptance of papers, size of the Editorial Board, and appointment of Associate Editors. The EiC is expected to adhere to the commitments expressed in the policy on Rights and Responsibilities in ACM Publishing ( For more information about the role of the EiC, see ACM’s Evaluation Criteria for Editors-in-Chief ( Nominations should include a vita along with a brief statement of why the nominee should be considered. Self-nominations are encouraged, and should include a statement of the candidate’s vision for the future development of TOMM. The deadline for submitting nominations is 24 July 2015, although nominations will continue to be accepted until the position is filled. Please send all nominations to the nominating committee chair, Nicu Sebe ( The search committee members are:

  • Nicu Sebe (University of Trento), Chair
  • Rainer Lienhart (University of Augsburg)
  • Alejandro Jaimes (Yahoo!)
  • John R. Smith (IBM)
  • Lynn Wilcox (FXPAL)
  • Wei Tsang Ooi (NUS)
  • Mary Lou Soffa (University of Virginia), ACM Publications Board Liaison