=Paper=
{{Paper
|id=Vol-1609/16090322
|storemode=property
|title=Image Annotation and Two Paths to Text Illustration
|pdfUrl=https://ceur-ws.org/Vol-1609/16090322.pdf
|volume=Vol-1609
|authors=Herve Le Borgne,Etienne Gadeski,Ines Chami,Thi Quynh Nhi Tran,Youssef Tamaazousti,Alexandru Lucian Gînscã,Adrian Popescu
|dblpUrl=https://dblp.org/rec/conf/clef/BorgneGCTTG016
}}
==Image Annotation and Two Paths to Text Illustration==
https://ceur-ws.org/Vol-1609/16090322.pdf
Image annotation and two paths to text
illustration
Hervé Le Borgne, Etienne Gadeski, Ines Chami, Thi Quynh Nhi Tran, Youssef
Tamaazousti, Alexandru Lucian Ginsca, and Adrian Popescu
CEA, LIST, Laboratory of Vision and Content Engineering, France
herve.le-borgne@cea.fr, ines.chami@cea.fr, etienne.gadeski@cea.fr,
thiquynhnhi.tran@cea.fr, youssef.tamaazousti@cea.fr,
alexandru.ginsca@cea.fr, adrian.popescu@cea.fr
Abstract. This paper describes our participation to the ImageCLEF 2016
scalable concept image annotation main task and Text Illustration teaser.
Regarding image annotation, we focused on better localizing the detected
features. For this, we identified the saliency of the image to collect a list
of potential interesting places into the image. We also added a specific
human attribute detector that boosted the results of the best performing
team in 2015. For the text illustration, we proposed two complementary
approaches. The first one relies on semantic signatures that give a tex-
tual description of an image. This description is further matched to the
textual query. The second approach learns a common latent space, in
which visual and textual features are directly comparable. We propose a
robust description, as well as the use of an auxiliary dataset to improve
retrieval. While the first approach only uses external data, the second
one was mainly learned from the provided training dataset.
1 Introduction
This paper describes our participation to the ImageCLEF 2016 [27] scalable
concept image annotation main task (IAL: image annotation and localization)
and Text Illustration teaser that are described in detail in [5].
Regarding image annotation, we improved our 2015 system [4] and focused
on better localizing the detected features. In 2015, we proposed a concept local-
ization pipeline which uses the spatial information that CNNs offer. To improve
this, we identified the saliency of the image to collect a list of potential interest-
ing places from the image, then detected the concepts found in these boxes. We
also added a specific human attribute detector that boosted the results of the
best performing team in 2015.
For text illustration, we proposed two complementary approaches. The first
one relies on semantic signatures that give a textual description of an image. This
description is further matched to the textual query. The second approach relies
on the learning of a common latent space, in which visual and textual features
are directly comparable, using a robust description and an auxiliary dataset to
improve retrieval. While the first approach only uses external data, the second
one was mainly learned from the provided training dataset.
This manuscript is organized as follows. Section 2 deals with our participation
to the image annotation and localization subtask, while Section 3 is dedicated
to the text illustration teaser. Each time, we discuss some limits of the task
itself that are important to better understand the results. Then we present the
method(s) and finally comment the results of the campaign.
2 Image annotation task
2.1 Dataset limitation
As highlighted last year by the team that obtained the best results [15], the
development set of the image annotation subtask (and it is probably the case
for the test set as well) suffers from severe limitations due to the crowd-sourcing
ground-truth annotation. They explain the annotation are inconsistent, incom-
plete, sometimes incorrect and there are even some cases that are “impossible”
according to the assumptions of the task (e.g. the fact there are at most 100
concepts per image).
It seems these issues have not been addressed in the 2016 development set,
thus the results are still subject to some limitations. However, on the other hand,
the task is thus consistent with last year’s results and we can directly compare
the improvement from one year to another.
2.2 Method
In this section, we detail the training and testing frameworks that we used. Our
method is based upon deep CNNs which have lately shown outstanding perfor-
mances in diverse computer vision tasks such as object classification, localization
and action recognition [20, 7].
Training
Data. We collected a set of roughly 251, 000 images (1, 000 images per concept)
from the Bing Images search engine. For each concept we used its name and its
synonyms (if present) to query the search engine. This dataset is of course noisy
but some works showed it is not a big issue to train a deep CNN [6, 29]. We used
this additional data to train a 16-layer CNN [21] and the 50-layer ResNet [10].
We used 90% of the dataset for training and 10% for validation.
Network Settings. The networks were initialized with ImageNet weights. The
initial learning rate is set to 0.001 and the batch size is set to 256. The last layer
(the classifier) is trained from scratch, i.e. it is initialized with random weights
sampled from a Gaussian distribution (σ = 0.01 and µ = 0) and its learning
rate is 10 times larger than for other layers. During training, the dataset is
enhanced with random transformations: RGB jittering, scale jittering, contrast
adjustment, JPEG compression and flips. It is known that data augmentation
leads to better models [30] and reduces overfitting. Finally, the networks take
a 224 × 224 RGB image as input and produce 251 outputs, i.e. the number
of concepts. The models were trained on a single Nvidia Titan Black with our
modified version of the Caffe framework [14].
Localizing concepts. We provide two approaches to detect the concepts and
localize them.
The first method, named FCN, is the same as described in [4]. It is a simple
and efficient framework where the concept detection and localization are done
simultaneously with a unique forward pass of the image to process. More in-
depth information about this framework can be found in [4].
The second method is based upon the generic object detector EdgeBoxes [31],
which takes an image as input and produces R regions where visual objects are
likely to appear (objectness detection). In our experiments, we extracted a max-
imum of 100 regions per image then fed each one to the CNN models. We fi-
nally kept the concept that had the highest probability among the 251 concepts.
Therefore this framework outputs R predictions per image.
In addition to these methods, we also used a face detector [26] to categorize
more precisely the faces, eyes, noses and mouths. We extracted those features
on all images and aggregated the results to the boxes detected by the CNN
frameworks. It was our belief that it would slightly boost our accuracy since
these kind of ”objects“ are quite hard to capture even with a good generic
object detector.
Combination of runs We also combined some runs by simply concatenating
detected the boxes. When the number of boxes was above the allowed limit (100)
we randomly removed some of them (this case was very rare).
2.3 Results
We submitted ten runs to the campaign, with settings that allow to measure the
benefit of different choices. We studied the influence of three parameters:
– our last year’s method used to localize the concepts [4] compared to the use
of EdgeBoxes [31]
– the CNN architecture, by comparing VGG [21] and ResNet [10] that obtained
good results at the ILSVRC campaign in 2014 and 2015.
– the use of a face part detector [26]
Results of individual runs are reported in Table 1.
On the ILSVRC challenge, VGG had 7.3% classification error and ResNet
obtained 5.7%. On the 2016 ImageCLEF dataset, we obtain similar results with
VGG and ResNet when we use FCN to localize the concepts and VGG is better
Table 1: Results of individual runs submitted in terms of mAP (×100) with 0.5
overlap. We report results including the facepart detection (face) or not (raw ).
FCN EdgeBoxes
raw - 32.16
VGG
face 37.66 31.75
raw 37.35 27.18
ResNet
face 37.84 24.14
Table 2: Results (mAP ×100) for two individual runs and four concepts con-
cerned by the face part detector.
VGG + EdgeBoxes ResNet + FCN
raw face raw face
mouth 100 54 2 55
eye 100 36 33 37
nose 25 28 7 42
face 67 75 50 76
with EdgeBoxes. Regarding the VGG-based scores, we noticed that our results
are about 8 points better than last year, showing the benefit of the new learning
process.
The use of a face part detector does not significantly improve our results and
they are even lower when we use the EdgeBoxes-based localization. This is quite
surprising since a similar process boosted the results of last year’s best perform-
ing team from 30.39 to 65.95 [15]. However, regarding these results, we should
notice this “boost” was observed on the test dataset only (on the development
set, the performances were more or less the same with and without the body
part detection). It is also hard to explain how the results can increase by 35
points while the body-part detector deals with less than 10 classes among 250.
Following a discussion with the organizers of the campaign, it seems that there
was a bug in the evaluation script (fixed since then, and probably reported on
this year’s overview [5]) and that detecting body parts is finally not very interest-
ing, making our results in line with other participants’ ascertainment. However,
there are still some unexpected results with regards to the concepts that are
directly concerned with face part detection. In Table 2, we report the results for
four of these concepts and two different settings (results of ResNet+EdgeBoxes
are similar to VGG+EdgeBoxes). With FCN, the behavior is in line with ex-
pectation. On the contrary with EdgeBoxes, the concepts mouth and eye are
perfectly detected, that is quite unlikely. Although there are obvious issues with
EdgeBoxes as explained below, this strange result may be due to a remaining
bug in the evaluation script.
The most disappointing result is that the use of the EdgeBoxes-based local-
ization gives globally lower results than the FCN one. A possible reason is that
EdgeBoxes generates much more boxes than FCN and that a significant part
leads to wrong concept estimation, hence penalizing the global score.
Last, we evaluated the combination of runs, as reported in Table 3. Once
again, the results are quite disappointing since the more we combine, the lower
the results are. Since the combination of runs is a simple concatenation of the
boxes found by each run, it is not clear to us how the results can decrease (the
mAP should be at least better than the weaker run). It probably results from
the way the results are evaluated but unfortunately the exact method used is
not available.
Table 3: Results for three combinations of runs.
Methods combined mAP (0.5 overlap) ×100
All VGG-based 32.76
All ResNet-based 27.11
All (both above) 21.7
3 Text illustration
3.1 Task realism
The task consists of matching textual queries to images without using the textual
features derived from the web page the images belong to, although these last
ones are available since they are part of the 500k noisy dataset. In practice, the
queries were mainly obtained by removing the HTML tags from these web pages,
and retaining all the remaining text. It thus raises an issue with regards to the
realism of the task.
Indeed, when one wants to illustrate a text in practice, she would submit the
interesting part of the text as query to the system. It does not make sense to
add some noisy data in the query such as that coming from the generic task bar
as in the query --/--diUdSrlGyv7zF4 that starts with:
Taakbalk Navigation Subnavigation Content home Who is who organ-
isational chart contact intranet nederlands zoekterm: Navigatie About
K.U.Leuven Education Research Admissions Living in Leuven Alumni
Libraries Faculties, Departments & Schools International cooperation
Virology - home Current Labmembers Former Labmembers Research
Projects Publications Contact Us Where To Find Us Courses (...)
Of course, it is hard for the organizers to extract this “relevant text” at
a large scale, since there are 180, 000 queries. However, this could be part of
the task: if the query was the actual HTML page, the system could include an
automatic search of the relevant text by using the DOM structure of the query.
Fig. 1: The semantic signature principle: an image is described in terms of like-
lihood to be similar to some concepts.
3.2 Semantic signature approach
Specific object detectors have been developed for a long time, to be able to rec-
ognize e.g. faces [28], pedestrians [3] or buildings [18]. More recently, it has been
proposed to use a set of object or concept detectors as image descriptors [24, 17],
introducing the “semantic features”. With this approach, images are described
into a fixed size vector space as it is the case with Bag-of-visual words, Ficher
Kernels [13] or even when one uses the last fully connected layer of a CNN as
a feature [20]. However, contrary to these approaches, each dimension of a se-
mantic feature is associated to a precise concepts that makes sense for a human
(Fig. 1). The “semantic signature approach” to text illustration thus consists in:
– (i) extracting relevant concept from images of reference
– (ii) expressing the corresponding concepts with words and index them
– (iii) matching the query to the index textually
During the campaign, we tested several alternatives for each of these steps.
Regarding step (i), our system is based on recently published work [23, 22],
that is itself an extension of the Semfeat descriptor [6]. Relying on powerful mid-
level-features such as [21], this semantic feature is a large set of linear classifiers
that are built automatically. The authors showed that keeping a small part of
the K most active concepts and setting the others to zero (sparsification) led
to a more efficient descriptor for an image retrieval task. However, there are
two limitation to this approach: first, the value of K has to be fixed in advance;
secondly, sparsification is not efficient in a classification context, in the sense that
the performance obtained is below that of the mid-level feature it is built on. For
these reasons, [23] proposed to compute K for each image independently, with
regard to the actual content of the image. The principle is to keep the “dominant
concepts” only, when we are confident on their detection.
For step (ii), we considered two sets of concepts. The first one, based on
WordNet, contains 17, 467 concepts, each being described by its main synset
(one word). The second set is that collected automatically in [6]. It contains
around 30, 000 concepts derived from Flickr groups, each described by three
words.
For the textual matching step (iii), we considered classical inverse indexing,
computing the query-document similarity from the “weight/score” associated to
each indexed document.
3.3 Text-image common space approach
The design of common latent spaces has been proposed for a while [16, 19], in
particular in the case of textual and visual modality [12, 32, 11, 2, 8]. Given two
modalities, let say a visual and a textual modality described by their respective
features, the general idea is to learn a latent common sub-space of both feature
spaces, such that visual points are directly comparable to textual ones. One of
the recent popular approach is to use Canonical Correlated Analysis (CCA), in
particular in its kernelised version (KCCA) [9].
Let us consider N data samples {(xTi , xIi )}N
i=1 ⊂ R
dT
× RdI , simultaneously
represented in two different vector spaces. The purpose of CCA is to find max-
imally correlated linear subspaces of these two vector spaces. More precisely, if
one notes X T ∈ RdT and X I ∈ RdI two random variables, CCA simultaneously
seeks directions wT ∈ RdT and wI ∈ RdI that maximize the correlation between
the projections of xT onto wT and of xI onto wI ,
0
wT CT I wI
wT∗ , wI∗ = arg maxp (1)
wT ,wI wT 0 CT T wT wI 0 CII wI
where CT T , CII denote the autocovariance matrices of X T and X I respectively,
while CT I is the cross-covariance matrix. The solutions wT∗ and wI∗ are found
solution of an eigenvalue problem. The d eigenvectors associated to the d largest
eigenvalues define maximally correlated d-dimensional subspaces in RdT and
respectively RdI . Even though these are linear subspaces of two different spaces,
they are often referred to as “common” representation space.
KCCA aims to remove the linearity constraint by using the “kernel trick”
to first map the data from each initial space to the reproducing kernel Hilbert
space associated to a selected kernel and then looking for correlated subspaces
in these RKHS.
In this space, textual and visual documents are directly comparable, thus it
is possible to perform cross-modal retrieval [12, 11, 2, 8]. However, it has been
recently found that the learned common space may not represent adequately all
data [25]. It has thus be proposed a more robust representation data within the
common space consisting in coding the original visual and textual point with
respect to a codebook (Figure 2a). This method, named Multimedia Aggregated
Correlated components (MACC) is detailed in [25]. Another contribution that
“‘compensates” the defaults of the representation space is to project a bi-modal
auxiliary dataset into the common space and use the known text-image connec-
tions of this dataset as a “pivot” to link e.g. a textual query to an appropriate
image of the reference database (Figure 2b).
(a) robust description (b) pivot principle
Fig. 2: Illustration of (a) the approach of [25] to describe robustly the bi-modal
documents into a common description space (b) the pivot principle using an
auxiliary database
MACC considers the use of two datasets. A first training dataset T is used
to learn the common space with a KCCA. Due to computational issues, the
number of documents that can be used to learn this space is limited to few ten
thousands. Hence, at a quite large scale such as that of the text illustration
subtask, it is important to use a second auxiliary dataset A to “compensate”
the possible limitation of the initial learning.
Once the settings of the common latent space are chosen, the principle of the
text illustration is quite straightforward, as illustrated in Figure 3. Regarding
the images of the reference database, we extract the same feature as that used
during learning, namely the FC7 fully connected layer of VGG [21]. This vector
is projected on the latent space and the MACC signature is computed then
stored into the reference database.
Regarding the textual query, we process the raw text in order to fix the
defaults identified in Section 3.1. We first remove the stopwords using the Stan-
ford NLTK package [1] that contains lists of stopwords in several languages. As
months, days and numbers are not included in the stopwords list from NLTK,
we also filtered them out as they are hard to illustrate and might add noise
to our model. Additionally, we also removed words containing special charac-
ters such as ””’ that are often found in noisy words. Furthermore, we combined
the stopwords list with a part of speech tagger developed in the NLTK library.
NLTK Pos Tag categorizes our set of words and labels each word according to its
grammatical properties. In order to take the more descriptive words, we choose
to keep only nouns (proper and common) and adjectives.
Fig. 3: General principle to illustrate a textual query from a purely visual
database, using a common space and MACC representation.
For each word i of the resulting vocabulary, we extract a word2vec vectorial
representation ti . Then, we compute a weight wi equal to its tfidf value. For
a document d, we select the k terms in the textual description that have the
largest weights. We then compute a unique vector vd , representing d, from the k
selected words describing a document, weighting each ti with its corresponding
Pn
wi ti
weight wi , resulting into the Weighted Arithmetic Mean (WAM) vd = Pi=1 n .
i=1 wi
As said above, the classical KCCA algorithm does only support some dozen
thousand document to learn the latent common space. To get around this lim-
itation, we first proceed to a selection of the training data, in order to build a
corpus with a diversified vocabulary. To do so, we divide our train set of 300k
documents into 10 groups of 30k documents each. We then clustered every group
of textual features with K-Means and compute 100 clusters per group. To build
a 20k documents corpus for instance, we select for each group 20 random docu-
ments per cluster (20K = ngroups nclusters ndoc = 10x100x20). Similarly, we build
diversified sets for the pivotal basis by selecting a certain number of random
documents per cluster.
3.4 Submission and Results
We submitted four individual runs and three runs that merge them differently.
Some synthetic results are presented in Table 4. Globally, the recall at 100 is
low, that is explained by the difficulty of the task as well as the noise in the
queries.
Table 4: Results of the seven run submitted.
Method Recall @ 100
CBS + wordnet 1.44
CBS + flickrgroup 1.74
Wam5 2.47
Wam7 4.33
MergeA 4.47
MergeB 4.51
MergeC 4.50
We run the method described in Section 3.2 with a semantic signature com-
puted with CBS and both the WordNet and FlickRgroups vocabularies. We
obtained better results with the smaller vocabulary of WordNet. The basic clas-
sifiers of the Wornet-based semantic features are learned with “cleaner” anno-
tated images then those based on the FlickRGroups. However, since the original
Semfeat paper [6] showed better or similar results with both types of classifiers,
we suggest that in the current task the (small) difference of performance may
be due to a better coverage of the vocabulary with respect to the queries.
The approach based on the common space and the MACC representation
lead to significantly better results. A first run Wam5 used 22k images for T , 64k
for A and 5 best words were retained to build the training and testing textual
features. In the second run Wam7 we used the same training dataset to learn the
common space but A was extended to 164k documents while we retained up to
30k words to build the textual training features. For the textual testing features,
we kept only 10 words to build the WAM, because of the noisy aspect of the
query data.
The experiments we run on a development dataset (not reported) extracted
from the 300k development images suggest that a large part of the improvement
between Wam5 and Wam7 is due to the growth of the auxiliary dataset A.
By merging several runs, the results are marginally improved. The run mergeA
concatenates 10 best results of CBS+WordNet and Wam5 with 80 best of Wam7. We
then also consider run wam8 similar to Wam7 but its testing textual queries that
were built with five best words (instead of 10 for Wam7). The run mergeB merges
the 10 best results of CBS+WordNet, Wam5 and W am8 with the 70 best of Wam7.
Finally, mergeC concatenates the 5 best results of CBS+FlickRGroups, the 10
best results of CBS+WordNet and Wam5, the 15 best of Wam8 and 60 best of Wam7.
While the settings are quite different between the three merging methods, the
results are similar, showing that the results is mainly due to the first answers of
Wam7.
4 Conclusion
We presented the results to the Image Annotation and Localization subtask and
the Text Illustration teaser. The results to IAL are good in comparison to other
participants, but the contribution proposed in 2016 did not lead to significantly
better results than our 2015 system. It is partially due to the fairer evaluation of
the task, but since the exact method of evaluation is not released, it is hard to
fully interpret the results, in particular why the combination of runs decreases
the mAP. Regarding the Text Illustration teaser, we proposed two methods based
on recently published work. The results are globally low, due to the difficulty of
the task in general1 and the very noisy queries in particular.
References
1. Bird, S., Klein, E., Loper, E.: Natural language processing with Python. ” O’Reilly
Media, Inc.” (2009)
2. Costa Pereira, J., Coviello, E., Doyle, G., Rasiwasia, N., Lanckriet, G., Levy, R.,
Vasconcelos, N.: On the role of correlation and abstraction in cross-modal multi-
media retrieval. TPAMI 36(3), 521–535 (2014)
3. Dalal, N., Triggs, B.: Histograms of oriented gradients for human detection. In: In
CVPR. pp. 886–893 (2005)
4. Gadeski, E., Le Borgne, H., Popescu, A.: Cea list’s participation to the scalable
concept image annotation task of imageclef 2015. In: CLEF2015 Working Notes.
CEUR Workshop Proceedings (2015)
5. Gilbert, A., Piras, L., Wang, J., Yan, F., Ramisa, A., Dellandrea, E., Gaizauskas,
R., Villegas, M., Mikolajczyk, K.: Overview of the ImageCLEF 2016 Scalable Con-
cept Image Annotation Task. In: CLEF2016 Working Notes. CEUR Workshop
Proceedings, CEUR-WS.org, Évora, Portugal (September 5-8 2016)
6. Ginsca, A.L., Popescu, A., Le Borgne, H., Ballas, N., Vo, P., Kanellos, I.: Large-
scale image mining with flickr groups. In: 21th International Conference on Multi-
media Modelling (MMM 15) (2015)
7. Girshick, R., Donahue, J., Darrell, T., Malik, J.: Rich feature hierarchies for ac-
curate object detection and semantic segmentation. In: Proceedings of the IEEE
Conference on Computer Vision and Pattern Recognition (CVPR) (2014)
8. Gong, Y., Ke, Q., Isard, M., Lazebnik, S.: A multi-view embedding space for mod-
eling internet images, tags, and their semantics. IJCV 106(2), 210–233 (Jan 2014)
9. Hardoon, D.R., Szedmak, S.R., Shawe-Taylor, J.R.: Canonical correlation analysis:
An overview with application to learning methods. Neural Comput. 16(12), 2639–
2664 (2004)
10. He, K., Zhang, X., Ren, S., Sun, J.: Delving deep into rectifiers: Surpassing human-
level performance on imagenet classification. In: ICCV (2015)
11. Hodosh, M., Young, P., Hockenmaier, J.: Framing image description as a rank-
ing task: Data, models and evaluation metrics. Journal of Artificial Intelligence
Research pp. 853–899 (2013)
12. Hwang, S.J., Grauman, K.: Learning the relative importance of objects from tagged
images for retrieval and cross-modal search. IJCV 100(2), 134–153 (Nov 2012)
1
the other participant to the task obtained outstanding results around 80%. If they
actually used the same data as us, we’ll of course revised the interest of our methods!
13. Jegou, H., Perronnin, F., Douze, M., Sanchez, J., Perez, P., Schmid, C.: Aggregat-
ing local image descriptors into compact codes. Transactions on Pattern Analysis
and Machine Intelligence 34(9), 1704–1716 (2012)
14. Jia, Y., Shelhamer, E., Donahue, J., Karayev, S., Long, J., Girshick, R., Guadar-
rama, S., Darrell, T.: Caffe: Convolutional architecture for fast feature embedding.
arXiv preprint arXiv:1408.5093 (2014)
15. Kakar, P., Wang, X., Chia, A.Y.S.: Automatic image annotation using weakly
labelled web data. In: CLEF2015 Working Notes (2015)
16. Li, D.: Multimedia content processing through cross-modal association. In: Proc.
ACM international conference on Multimedia. pp. 604–611. ACM Press (2003)
17. Li, L.j., Su, H., Fei-fei, L., Xing, E.P.: Object bank: A high-level image represen-
tation for scene classification & semantic feature sparsification. In: NIPS (2010)
18. Malobabić, J., Le Borgne, H., Murphy, N., O’Connor, N.E.: Detecting the pres-
ence of large buildings in natural images. In: Content-Based Multimedia Indexing
(CBMI), 2005 International Workshop on (2005)
19. Ngiam, J., Khosla, A., Kim, M., Nam, J., Lee, H., Ng, A.Y.: Multimodal deep
learning. In: ICML. pp. 689–696 (2011)
20. Sharif Razavian, A., Azizpour, H., Sullivan, J., Carlsson, S.: Cnn features off-
the-shelf: An astounding baseline for recognition. In: The IEEE Conference on
Computer Vision and Pattern Recognition (CVPR) Workshops (June 2014)
21. Simonyan, K., Zisserman, A.: Very deep convolutional networks for large-scale
image recognition. CoRR abs/1409.1556 (2014)
22. Tamaazousti, Y., Le Borgne, H., Hudelot, C.: Diverse concept-level features for
multi-object classification. In: International Conference on Multimedia retrieval
(ICMR’16). New York, USA (June 2016)
23. Tamaazousti, Y., Le Borgne, H., Popescu, A.: Constrained local enhancement of
semantic features by content-based sparsity. In: International Conference on Mul-
timedia retrieval (ICMR’16). New York, USA (June 2016)
24. Torresani, L., Szummer, M., Fitzgibbon, A.: Efficient object category recognition
using classemes. In: European Conference on Computer Vision. ECCV (2010)
25. Tran, T.Q.N., Le Borgne, H., Crucianu, M.: Aggregating image and text quantized
correlated components. In: IEEE Conference on Computer Vision and Pattern
Recognition (CVPR). IEEE, Las Vegas, USA (June 2016)
26. Uřičář, M., Franc, V., Thomas, D., Sugimoto, A., Hlaváč, V.: Real-time Multi-view
Facial Landmark Detector Learned by the Structured Output SVM. In: BWILD
’15: Biometrics in the Wild 2015 (IEEE FG 2015 Workshop) (2015)
27. Villegas, M., Müller, H., Garcı́a Seco de Herrera, A., Schaer, R., Bromuri, S.,
Gilbert, A., Piras, L., Wang, J., Yan, F., Ramisa, A., Dellandrea, E., Gaizauskas,
R., Mikolajczyk, K., Puigcerver, J., Toselli, A.H., Snchez, J.A., Vidal, E.: General
Overview of ImageCLEF at the CLEF 2016 Labs. Lecture Notes in Computer
Science, Springer International Publishing (2016)
28. Viola, P., Jones, M.: Robust real-time object detection. In: IJCV (2001)
29. Vo, P.D., Ginsca, A.L., Le Borgne, H., Popescu, A.: Effective training of convolu-
tional networks using noisy web images. In: CBMI (2015), prague, Czech Republic
30. Wu, R., Ya, S., Shan, Y., Dang, Q., Sun, G.: Deep image: Scaling up image recog-
nition. CoRR abs/1501.02876 (2015)
31. Zitnick, C.L., Dollár, P.: Edge boxes: Locating object proposals from edges. In:
ECCV. European Conference on Computer Vision (September 2014)
32. Znaidia, A., Shabou, A., Le Borgne, H., Hudelot, C., Paragios, N.: Bag-of-
multimedia-words for image classification. In: ICPR. pp. 1509–1512. IEEE (2012)