Verifying the authenticity of media has become an increasingly challenging task. Rapid advances in AI-generated content, spanning modalities like text, images, video, audio have significantly blurred the line between genuine and synthetic information. Nowadays, powerful foundation models can easily be leveraged to create, amplify and disseminate information at scale, enabling disinformation campaigns, defamation, or impersonation. This results in the erosion of trust in online information, which poses a great threat to society. The MAD’26 workshop seeks to address this problem by bringing together researchers and practitioners from diverse disciplines, united by the goal of combating disinformation through AI-driven approaches. Now in its fifth edition, the workshop aims to cultivate a collaborative environment that encourages the exchange of ideas, methodologies, and practical experiences. The workshop focuses on key research directions, including the detection of AI-generated and manipulated content, the analysis of disinformation propagation, and the examination of its broader societal impact.

The performance of Recommender Systems (RS) varies significantly across users, yet the underlying reasons for this variance remain poorly understood. This paper introduces a unified framework to analyze and explain this performance gap by quantifying user profile characteristics. We propose two novel, information-theoretic measures: Mean Surprise (𝑆(𝑢)), which captures a user's deviation from popular items and is closely related to popularity bias, and Mean Conditional Surprise (𝐂𝑆(𝑢)), which measures the internal coherence of a user's interactions in a domain-agnostic manner. Through extensive experiments on 7 algorithms and 9 datasets, we demonstrate that these measures are strong predictors of recommendation performance. Our analysis reveals that performance gains from complex models are concentrated on ''coherent'' users, while all algorithms perform poorly on ''incoherent'' users. We show how these measures provide practical utility for the Web community by: (1) enabling robust, stratified evaluation to identify model weaknesses; (2) facilitating a novel analysis of the behavioral alignment of recommendations; and (3) guiding targeted system design, which we validate by training a specialized model on a segment of ''coherent'' users that achieves superior performance for that group with significantly less data. This work provides a new lens for understanding user behavior and offers practical tools for building more robust and efficient large-scale recommender systems.

2148

AI systems must adapt to the evolving visual landscape, especially in domains where object appearance shifts over time. While prior work on time-aware vision models has primarily addressed commonsense-level categories, we introduce Car Models in Time (CaMiT). This fine-grained dataset captures the temporal evolution of this representative subset of technological artifacts. CaMiT includes 787K labeled samples of 190 car models (2007–2023) and 5.1M unlabeled samples (2005–2023), supporting supervised and self-supervised learning. We show that static pretraining on in-domain data achieves competitive performance with large-scale generalist models, offering a more resource-efficient solution. However, accuracy degrades when testing a year's models backward and forward in time. To address this, we evaluate CaMiT in a time-incremental classification setting, a realistic continual learning scenario with emerging, evolving, and disappearing classes. We investigate two mitigation strategies: time-incremental pretraining, which updates the backbone model, and time-incremental classifier learning, which updates the final classification layer, with positive results in both cases. Finally, we introduce time-aware image generation by consistently using temporal metadata during training. Results indicate improved realism compared to standard generation. CaMiT provides a rich resource for exploring temporal adaptation in a fine-grained visual context for discriminative and generative AI systems.

Exemplar-free class-incremental learning (EFCIL) aims to classify streaming data without storing examples from the past. Recent EFCIL works suggest that (i) models pre-trained with large amounts of data should be used to initialize learning, (ii) self-supervised learned transformers generalize better than supervised convolutional models, (iii) adding generated data to the pre-training dataset can improve incremental accuracy. In this article, we question the above assertions by comprehensively evaluating various initial training strategies combined with four EFCIL algorithms using four large-scale datasets. Our results indicate that: (i) Pre-trained models are preferable when the domain of the incremental classification task is well represented in the pre-training datasets, but training with initial data remains useful when the domain shift is significant, (ii) supervised convolutional networks remain competitive, particularly when improving representation transferability using data augmentation or a projector, (iii) adding classes from an external dataset to train the initial model boosts performance when the initial set of classes is small but has a limited effect otherwise, (iv) additional classes generated with a diffusion model are not necessarily more useful than a well-chosen set of ImageNet classes to improve model transferability. We provide a nuanced analysis of these results and formulate recommendations to facilitate the practical adoption of EFCIL algorithms.

Face recognition and verification are two computer vision tasks whose performances have advanced with the introduction of deep representations. However, ethical, legal, and technical challenges due to the sensitive nature of face data and biases in real-world training datasets hinder their development. Generative AI addresses privacy by creating fictitious identities, but fairness problems remain. Using the existing DCFace SOTA framework, we introduce a new controlled generation pipeline that improves fairness. Through classical fairness metrics and a proposed indepth statistical analysis based on logit models and ANOVA, we show that our generation pipeline improves fairness more than other bias mitigation approaches while slightly improving raw performance.

Visual datasets are generally constructed from the samples available at the time of their collection and are not further updated. However, these static datasets do not reflect the distribution changes that occur in real data. We analyze how different collection times lead to a shift in class distribution by collecting a set of Flickr images published over 14 years. The proposed “Visual Classes through Time” (VCT-107) dataset contains images tagged by their publication date and includes 107 classes covering various topics (human-made objects, animals, plants, food, etc.). Images from each class are divided into five collection periods to study the impact of time on classification accuracy. When training different classification models using linear probing, we observe an accuracy loss when training on data from one period and testing on other periods. This happens even in the case of a strongly pre-trained model like DinoV2 ViT-B/14. Intuitively, the performance loss is generally more significant when the collection periods between the training and test data are further apart. Our analysis reveals that the temporal shift varies between classes, with the largest shifts observed for human-made objects and the smallest for natural concepts such as animal species. Our results stress the importance of regularly updating models to adapt to timeinduced changes in the distribution of visual classes, even when using a strongly pre-trained model. We release the VCT-107 dataset to facilitate research on temporal shifts.

1339

Class-Incremental Learning (CIL) aims to build classification models from data streams. At each step of the CIL process, new classes must be integrated into the model. Due to catastrophic forgetting, CIL is particularly challenging when examples from past classes cannot be stored, the case on which we focus here. To date, most approaches are based exclusively on the target dataset of the CIL process. However, the use of models pre-trained in a self-supervised way on large amounts of data has recently gained momentum. The initial model of the CIL process may only use the first batch of the target dataset, or also use pre-trained weights obtained on an auxiliary dataset. The choice between these two initial learning strategies can significantly influence the performance of the incremental learning model, but has not yet been studied in depth. Performance is also influenced by the choice of the CIL algorithm, the neural architecture, the nature of the target task, the distribution of classes in the stream and the number of examples available for learning. We conduct a comprehensive experimental study to assess the roles of these factors. We present a statistical analysis framework that quantifies the relative contribution of each factor to incremental performance. Our main finding is that the initial training strategy is the dominant factor influencing the average incremental accuracy, but that the choice of CIL algorithm is more important in preventing forgetting. Based on this analysis, we propose practical recommendations for choosing the right initial training strategy for a given incremental learning use case. These recommendations are intended to facilitate the practical deployment of incremental learning.

Target-dependent sentiment classification (TSC) enables a fine-grained automatic analysis of sentiments expressed in texts. Sentiment expression varies depending on the domain, and it is necessary to create domain-specific datasets.While socially important, TSC in the news domain remains relatively understudied.We introduce MAD-TSC, the first multilingual aligned dataset designed for TSC in news.MAD-TSC differs substantially from existing resources.First, it includes aligned examples in eight languages to facilitate a comparison of performance for individual languages, and a direct comparison of human and machine translation.Second, the dataset is sampled from a diversified parallel news corpus, and is diversified in terms of news sources and geographic spread of entities.Finally, MAD-TSC is more challenging than existing datasets because its samples are more complex.We exemplify the use of MAD-TSC with comprehensive monolingual and multilingual experiments.The latter shows that machine translations can successfully replace manual ones, and that performance for all included languages can match that of English by automatically translating test examples.

With recent advancements in synthetic media manipulation and generation, verifying multimedia content posted online has become increasingly difficult. Additionally, the malicious exploitation of AI technologies by actors to disseminate disinformation on social media, and more generally the Web, at an alarming pace poses significant threats to society and democracy. Therefore, the development of AI-powered tools that facilitate media verification is urgently needed. The MAD ’23 workshop aims to bring together individuals working on the wider topic of detecting disinformation in multimedia to exchange their experiences and discuss innovative ideas, attracting people with varying backgrounds and expertise. The research areas of interest include identifying manipulated and synthetic content in multimedia, as well as examining the dissemination of disinformation and its impact on society. The multimedia aspect is very important since content most often contains a mix of modalities and their joint analysis can boost the performance of verification methods.

Online social networks use AI techniques to automatically infer profiles from users’ shared data. However, these inferences and their effects remain, to a large extent, opaque to the users themselves. We propose a method which raises user awareness about the potential use of their profiles in impactful situations, such as searching for a job or an accommodation. These situations illustrate usage contexts that users might not have anticipated when deciding to share their data. User photographic profiles are described by automatic object detections in profile photos, and associated object ratings in situations. Human ratings of the profiles per situation are also available for training. These data are represented as graph structures which are fed into graph neural networks in order to learn how to automatically rate them. An adaptation of the learning procedure per situation is proposed since the same profile is likely to be interpreted differently, depending on the context. Automatic profile ratings are compared to one another in order to inform individual users of their standing with respect to others. Our method is evaluated on a public dataset, and consistently outperforms competitive baselines. An ablation study gives insights about the role of its main components.

Recent class-incremental learning methods combine deep neural architectures and learning algorithms to handle streaming data under memory and computational constraints. The performance of existing methods varies depending on the characteristics of the incremental process. To date, there is no other approach than to test all pairs of learning algorithms and neural architectures on the training data available at the start of the learning process to select a suited algorithm-architecture combination. To tackle this problem, in this article, we introduce AdvisIL, a method which takes as input the main characteristics of the incremental process (memory budget for the deep model, initial number of classes, size of incremental steps) and recommends an adapted pair of learning algorithm and neural architecture. The recommendation is based on a similarity between the user-provided settings and a large set of pre-computed experiments. AdvisIL makes class-incremental learning easier, since users do not need to run cumbersome experiments to design their system. We evaluate our method on four datasets under six incremental settings and three deep model sizes. We compare six algorithms and three deep neural architectures. Results show that AdvisIL has better overall performance than any of the individual combinations of a learning algorithm and a neural architecture. AdvisIL’s code is available at https://github.com/EvaJF/AdvisIL.

Exemplar-free class-incremental learning is very challenging due to the negative effect of catastrophic forgetting. A balance between stability and plasticity of the incremental process is needed in order to obtain good accuracy for past as well as new classes. Existing exemplar-free class-incremental methods focus either on successive fine tuning of the model, thus favoring plasticity, or on using a feature extractor fixed after the initial incremental state, thus favoring stability. We introduce a method which combines a fixed feature extractor and a pseudo-features generator to improve the stability-plasticity balance. The generator uses a simple yet effective geometric translation of new class features to create representations of past classes, made of pseudo-features. The translation of features only requires the storage of the centroid representations of past classes to produce their pseudo-features. Actual features of new classes and pseudo-features of past classes are fed into a linear classifier which is trained incrementally to discriminate between all classes. The incremental process is much faster with the proposed method compared to mainstream ones which update the entire deep model. Experiments are performed with three challenging datasets, and different incremental settings. A comparison with ten existing methods shows that our method outperforms the others in most cases. FeTrIL code is available at https://github.com/GregoirePetit/FeTrIL.

Most recommendation datasets for tourism are restricted to one world region and rely on explicit data such as checkins. However, in reality, tourists visit various places world-wide and document their trips primarily through photos. These images contain a wealth of raw information that can be used to capture users’ preferences and recommend personalized content. Visual content was already used in past works, but no large-scale publicly-available dataset that gives access to users’ personal images exists for recommender systems. As such a resource would open-up possibilities for new image-based recommendation algorithms, we introduce Vis2Rec, a new dataset based on visit data extracted from users’ Flickr photographic streams, which includes over 7 million photos, 36k recognizable points of interest, and 14k user profiles. Google Landmarks v2 is used as an auxiliary dataset to identify points of interest in users’ photos, using a state-of-the-art image-matching deep architecture. Image-based user profiles are then constituted by aggregating the points of interest detected for each user. In addition, ground truth visits were determined for the test subset in order to enable accurate evaluation. Finally, we benchmark Vis2Rec using various existing recommender systems, and discuss the possibilities opened up by the availability of user images, as well as the societal issues that come with them. Following good practice in dataset sharing, Vis2Rec is created using only freely distributable content, and additional anonymization is performed to ensure the privacy of users. The raw dataset and the preprocessed user profiles will be publicly available at https://github.com/MSoumm/Vis2Rec.

The verification of multimedia content posted online becomes increasingly challenging due to recent advancements in synthetic media manipulation and generation. Moreover, malicious actors can easily exploit AI technologies to spread disinformation across social media at a rapid pace, which poses very high risks for society and democracy. There is, therefore, an urgent need for AI-powered tools that facilitate the media verification process. The objective of the MAD '22 workshop is to bring together those who work on the broader topic of disinformation detection in multimedia in order to share their experiences and discuss their novel ideas, reaching out to people with different backgrounds and expertise. The research domains of interest vary from the detection of manipulated and synthetic content in multimedia to the analysis of the spread of disinformation and its impact on society. The MAD '22 workshop proceedings are available at: https://dl.acm.org/citation.cfm?id=3512732.

Face verification aims to distinguish between genuine and imposter pairs of faces, which include the same or different identities, respectively. The performance reported in recent years gives the impression that the task is practically solved. Here, we revisit the problem and argue that existing evaluation datasets were built using two oversimplifying design choices. First, the usual identity selection to form imposter pairs is not challenging enough because, in practice, verification is needed to detect challenging imposters. Second, the underlying demographics of existing datasets are often insufficient to account for the wide diversity of facial characteristics of people from across the world. To mitigate these limitations, we introduce the F aV CI2D dataset. Imposter pairs are challenging because they include visually similar faces selected from a large pool of demographically diversified identities. The dataset also includes metadata related to gender, country and age to facilitate fine-grained analysis of results. F aV CI2D is generated from freely distributable resources. Experiments with state-of-the-art deep models that provide nearly 100% performance on existing datasets show a significant performance drop for F aV CI2D, confirming our starting hypothesis. Equally important, we analyze legal and ethical challenges which appeared in recent years and hindered the development of face analysis research. We introduce a series of design choices which address these challenges and make the dataset constitution and usage more sustainable and fairer. F aV CI2D is available at https://github.com/AIMultimediaLab/FaVCI2D-Face-Verification-with-Challenging-Imposters-and-Diversified-Demographics.

Social networks give free access to their services in exchange for the right to exploit their users’ data. Data sharing is done in an initial context which is chosen by the users. However, data are used by social networks and third parties in different contexts which are often not transparent. In order to unveil such usages, we propose an approach which focuses on the effects of data sharing in impactful real-life situations. Focus is put on visual content because of its strong influence in shaping online user profiles. The approach relies on three components: (1) a set of visual objects with associated situation impact ratings obtained by crowdsourcing, (2) a corresponding set of object detectors for mining users’ photos and (3) a ground truth dataset made of 500 visual user profiles which are manually rated per situation. These components are combined in LERV UP, a method which learns to rate visual user profiles in each situation. LERV UP exploits a new image descriptor which aggregates object ratings and object detections at user level and an attention mechanism which boosts highly-rated objects to prevent them from being overwhelmed by low-rated ones. Performance is evaluated per situation by measuring the correlation between the automatic ranking of profile ratings and a manual ground truth. Results indicate that LERV UP is effective since a strong correlation of the two rankings is obtained. A practical implementation of the approach in a mobile app which raises user awareness about shared data usage is also discussed.

Incremental learning enables artificial agents to learn from sequential data. While important progress was made by exploiting deep neural networks, incremental learning remains very challenging. This is particularly the case when no memory of past data is allowed and catastrophic forgetting has a strong negative effect. We tackle class-incremental learning without memory by adapting prediction bias correction, a method which makes predictions of past and new classes more comparable. It was proposed when a memory is allowed and cannot be directly used without memory, since samples of past classes are required. We introduce a two-step learning process which allows the transfer of bias correction parameters between reference and target datasets. Bias correction is first optimized offline on reference datasets which have an associated validation memory. The obtained correction parameters are then transferred to target datasets, for which no memory is available. The second contribution is to introduce a finer modeling of bias correction by learning its parameters per incremental state instead of the usual past vs. new class modeling. The proposed dataset knowledge transfer is applicable to any incremental method which works without memory. We test its effectiveness by applying it to four existing methods. Evaluation with four target datasets and different configurations shows consistent improvement, with practically no computational and memory overhead.

Active learning aims to optimize the dataset annotation process when resources are constrained. Most existing methods are designed for balanced datasets. Their practical applicability is limited by the fact that a majority of real-life datasets are actually imbalanced. Here, we introduce a new active learning method which is designed for imbalanced datasets. It favors samples likely to be in minority classes so as to reduce the imbalance of the labeled subset and create a better representation for these classes. We also compare two training schemes for active learning: (1) the one commonly deployed in deep active learning using model fine tuning for each iteration and (2) a scheme which is inspired by transfer learning and exploits generic pre-trained models and train shallow classifiers for each iteration. Evaluation is run with three imbalanced image datasets. Results show that the proposed active learning method outperforms competitive baselines. Equally interesting, they also indicate that the transfer learning training scheme outperforms model fine tuning if features are transferable from the generic dataset to the unlabeled one. This last result is surprising and should encourage the community to explore the design of deep active learning methods.

Active learning increases the effectiveness of labeling when only subsets of unlabeled datasets can be processed manually. To our knowledge, existing algorithms are designed under the assumption that datasets are balanced. However, many real-life datasets are actually imbalanced and we propose two adaptations of active learning to tackle imbalance. First, we modify acquisition functions to select samples by taking advantage of a deep model pretrained on a source domain. Second, we introduce a balancing step in the acquisition process to reduce the imbalance of the labeled subset. Evaluation is done with four imbalanced datasets using existing active learning methods and their modifications introduced here. Results show that our adaptations are useful as long as knowledge from the source domain is transferable to target domains.

Incremental learning is useful if an AI agent needs to integrate data from a stream. The problem is non trivial if the agent runs on a limited computational budget and has a bounded memory of past data. In a deep learning approach, the constant computational budget requires the use of a fixed architecture for all incremental states. The bounded memory generates imbalance in favor of new classes and a prediction bias toward them appears. This bias is commonly countered by introducing a data balancing step in addition to the basic network training. We depart from this approach and propose simple but efficient scaling of past classifiers' weights to make them more comparable to those of new classes. Scaling exploits incremental state statistics and is applied to the classifiers learned in the initial state of classes to profit from all their available data. We also question the utility of the widely used distillation loss component of incremental learning algorithms by comparing it to vanilla fine tuning in presence of a bounded memory. Evaluation is done against competitive baselines using four public datasets. Results show that the classifier weights scaling and the removal of the distillation are both beneficial.

This paper presents a class incremental learning (IL) method which exploits fine tuning and a dual memory to reduce the negative effect of catastrophic forgetting in image recognition. First, we simplify the current fine tuning based approaches which use a combination of classification and distillation losses to compensate for the limited availability of past data. We find that the distillation term actually hurts performance when a memory is allowed. Then, we modify the usual class IL memory component. Similar to existing works, a first memory stores exemplar images of past classes. A second memory is introduced here to store past class statistics obtained when they were initially learned. The intuition here is that classes are best modeled when all their data are available and that their initial statistics are useful across different incremental states. A prediction bias towards newly learned classes appears during inference because the dataset is imbalanced in their favor. The challenge is to make predictions of new and past classes more comparable. To do this, scores of past classes are rectified by leveraging contents from both memories. The method has negligible added cost, both in terms of memory and of inference complexity. Experiments with three large public datasets show that the proposed approach is more effective than a range of competitive state-of-the-art methods.

Semantic features represent images by the outputs of a set of visual concept classifiers and have shown interesting performances in image classification and retrieval. All classifier outputs are usually exploited but it was recently shown that feature sparsification improves both performance and scalability. However, existing approaches consider a fixed sparsity level which disregards the actual content of individual images. In this paper, we propose a method to determine automatically a level of sparsity for the semantic features that is adapted to each image content. This method takes into account the amount of information contained by the image through a modeling of the semantic feature entropy and the confidence of individual dimensions of the feature. We also investigate the use of local regions of the image to further improve the quality of semantic features. Experimental validation is conducted on three benchmarks (Pascal VOC 2007, VOC 2012 and MIT Indoor) for image classification and two of them for image retrieval. Our method obtains competitive results on image classification and achieves state-of-the-art performances on image retrieval.

Information sharing in online social networks is a daily practice for billions of users. The sharing process facilitates the maintenance of users' social ties but also entails privacy disclosure in relation to other users and third parties. Depending on the intentions of the latter, this disclosure can become a risk. It is thus important to propose tools that empower the users in their relations to social networks and third parties connected to them. As part of USEMP, a coordinated research effort aimed at user empowerment, we introduce a system that performs privacy-aware classification of images. We show that generic privacy models perform badly with real-life datasets in which images are contributed by individuals because they ignore the subjective nature of privacy. Motivated by this, we develop personalized privacy classification models that, utilizing small amounts of user feedback, provide significantly better performance than generic models. The proposed semi-personalized models lead to performance improvements for the best generic model ranging from 4%, when 5 user-specific examples are provided, to 18% with 35 examples. Furthermore, by using a semantic representation space for these models we manage to provide intuitive explanations of their decisions and to gain novel insights with respect to individuals' privacy concerns stemming from image sharing. We hope that the results reported here will motivate other researchers and practitioners to propose new methods of exploiting user feedback and of explaining privacy classifications to users.

In this paper we introduce a new dataset, Div150Multi, that was designed to support shared evaluation of diversification techniques in different areas of social media photo retrieval and related areas. The dataset comes with associated relevance and diversity assessments performed by trusted annotators. The data consists of around 300 complex queries represented via 86,769 Flickr photos, around 27M photo links for around 6,000 users, metadata, Wikipedia pages and content descriptors for text and visual modalities, including state of the art deep features. To facilitate distribution, only Creative Commons content allowing redistribution was included in the dataset. The proposed dataset was validated during the 2015 Retrieving Diverse Social Images Task at the MediaEval Benchmarking.

Handwriting recognition systems usually rely on static dictionaries and language models. Full coverage of these dictionaries is generally not achieved when dealing with unrestricted document corpora due to the presence of Out-Of-Vocabulary words. In a previous work, dynamic dictionaries were built from Web resources and successfully applied to isolated word recognition. In the present work we extend this approach to text-line recognition. Line segmentation into words is needed to exploit dynamic dictionaries and it is performed using BLSTM classifiers to align filler models and word sequence outputs. Words are then classified based on the confidence score into anchor and non-anchor words (AWs and NAWs). AWs are equated to the BLSTM outputs and used as such. Dynamic dictionaries are built for NAWs by exploiting Web resources for their character sequence and for neighboring AWs. Text-lines are decoded again using dynamic dictionaries and re-estimated language model. We conduct experiments on the publicly available RIMES database and show that the introduction of the dynamic dictionary is beneficial. Equally important, we show that the gain increases as the proportion of OOVs increases.

Results returned by commercial image search engines should include relevant and diversified depictions of queries in order to ensure good coverage of users' information needs. While relevance has drastically improved in recent years, diversity is still an open problem. In this paper we propose a reranking method that could be implemented on top of such engines in order to provide a better balance between relevance and diversity. Our method formulates the reranking problem as an optimization of a utility function that jointly considers relevance and diversity. Our main contribution is the replacement of the unsupervised definition of relevance that is commonly used in this formulation with a supervised classification model that strives to capture a query and application-specific notion of relevance. This model provides more accurate relevance scores that lead to significantly improved diversification performance. Furthermore, we propose a stacking-type ensemble learning approach that allows combining multiple features in a principled way when computing the relevance of an image. An empirical evaluation carried out on the datasets of the MediaEval 2013 and 2014 "Retrieving Diverse Social Images" (RDSI) benchmarks confirms the superior performance of the proposed method compared to other participating systems as well as a state-of-the-art, unsupervised reranking method.

In this paper we introduce a new dataset and its evaluation tools, Div150Cred, that was designed to support shared evaluation of diversification techniques in different areas of social media photo retrieval and related areas. The dataset comes with associated relevance and diversity assessments performed by human annotators. The data consists of 300 landmark locations represented via 45,375 Flickr photos, 16M photo links for around 3,000 users, metadata, Wikipedia pages and content descriptors for text and visual modalities. To facilitate distribution, only Creative Commons content was included in the dataset. The proposed dataset was validated during the 2014 Retrieving Diverse Social Images Task at the MediaEval Benchmarking Initiative.

This article addresses the issue of retrieval result diversification in the context of social image retrieval and discusses the results achieved during the MediaEval 2013 benchmarking. 38 runs and their results are described and analyzed in this text. A comparison of the use of expert vs. crowdsourcing annotations shows that crowdsourcing results are slightly different and have higher inter observer differences but results are comparable at lower cost. Multimodal approaches have best results in terms of cluster recall. Manual approaches can lead to high precision but often lower diversity. With this detailed results analysis we give future insights on this matter.

Existing image retrieval systems exploit textual or/and visual information to return results. The retrieval process is mostly focused on data themselves and disregards the data sources. In Web 2.0 platforms, the quality of annotations provided by different users can vary strongly. To account for this variability, we complement existing methods by introducing user tagging credibility in the retrieval process. Tagging credibility is automatically estimated by leveraging a large set of visual concept classifiers learned with Overfeat, a convolutional neural network (CNN) based feature. A good image retrieval system should return results that are both relevant and diversified and tackle both challenges. We diversify results by using a k-Means algorithm with user based cluster ranking. We increase relevance by favoring images uploaded from users with good credibility estimates. Evaluation is performed on DIV400, a publicly available social image retrieval dataset. Experiments show that our method provides interesting performances compared to existing approaches.

In this paper we propose a new dataset, Div400, that was designed to support shared evaluation in different areas of social media photo retrieval, e.g., machine analysis (re-ranking, machine learning), human-based computation (crowdsourcing) or hybrid approaches (relevance feedback, machine-crowd integration). Div400 comes with associated relevance and diversity assessments performed by human annotators. 396 landmark locations are represented via 43,418 Flickr photos and metadata, Wikipedia pages and content descriptors for text and visual modalities. To facilitate distribution, only Creative Commons content was included in the dataset. The proposed dataset was validated during the 2013 Retrieving Diverse Social Images Task at the MediaEval Benchmarking Initiative for Multimedia Evaluation.

We propose an ontology alignment framework with two core features: the use of background knowledge and the ability to handle vagueness in the matching process and the resulting concept alignments. The procedure is based on the use of a generic reference vocabulary, which is used for fuzzifying the ontologies to be matched. The choice of this vocabulary is problem-dependent in general, although Wikipedia represents a general-purpose source of knowledge that can be used in many cases, and even allows cross language matchings. In the first step of our approach, each domain concept is represented as a fuzzy set of reference concepts. In the next step, the fuzzified domain concepts are matched to one another, resulting in fuzzy descriptions of the matches of the original concepts. Based on these concept matches, we propose an algorithm that produces a merged fuzzy ontology that captures what is common to the source ontologies. The paper describes experiments in the domain of multimedia by using ontologies containing tagged images, as well as an evaluation of the approach in an information retrieval setting. The undertaken fuzzy approach has been compared to a classical crisp alignment by the help of a ground truth that was created based on human judgment.

Bilingual lexicons are central components of machine translation and cross-lingual information retrieval systems.Their manual construction requires strong expertise in both languages involved and is a costly process.Several automatic methods were proposed as an alternative but they often rely on resources available in a limited number of languages and their performances are still far behind the quality of manual translations.We introduce a novel approach to the creation of specific domain bilingual lexicon that relies on Wikipedia.This massively multilingual encyclopedia makes it possible to create lexicons for a large number of language pairs.Wikipedia is used to extract domains in each language, to link domains between languages and to create generic translation dictionaries.The approach is tested on four specialized domains and is compared to three state of the art approaches using two language pairs: French-English and Romanian-English.The newly introduced method compares favorably to existing methods in all configurations tested.

Handwriting recognition systems rely on predefined dictionaries obtained from training data. Small and static dictionaries are usually exploited to obtain high in-vocabulary (IV) accuracy at the expense of coverage. Thus the recognition of out-of-vocabulary (OOV) words cannot be handled efficiently. To improve OOV recognition while keeping IV dictionaries small, we introduce a multi-step approach that exploits Web resources. After an initial IV-OOV sequence classification, external resources are used to create OOV sequence-adapted dynamic dictionaries. A final Viterbi-based decoding is performed over the dynamic dictionary to determine the most probable word for the OOV sequence. We validate our approach with experiments conducted on RIMES, a publicly available database. Results show that improvements are obtained compared to standard handwriting recognition, performed with a static dictionary. Both domain adapted and generic dynamic dictionaries are studied and we show that domain adaptation is beneficial.

Points of interest (POIs) are a core component of geographical databases and of location based services. POI acquisition was performed by domain experts but associated costs and access difficulties in many regions of the world reduce the coverage of manually built geographical databases. With the availability of large geotagged multimedia datasets on the Web, a sustained research effort was dedicated to automatic POI discovery and characterization. However, in spite of its practical importance, POI localization was only marginally addressed. To compute POI coordinates an assumption was made that the more data were available, the more precise the localization will be. Here we shift the focus of the process from data quantity to data quality. Given a set of geotagged Flickr photos associated to a POI, close-up classification is used to trigger a spatial clustering process. To evaluate the newly introduced method against different other localization schemes, we create an accurate ground truth. We show that significant localization error reductions are obtained compared to a coordinate averaging approach and to a X-Means clustering scheme.

The automatic attribution of semantic labels to unlabeled or weakly labeled images has received considerable attention but, given the complexity of the problem, remains a hard research topic. Here we propose a unified classification framework which mixes textual and visual information in a seamless manner. Unlike most recent previous works, computer vision techniques are used as inspiration to process textual information. To do so, we consider two types of complementary tag similarities, respectively computed from a conceptual hierarchy and from data collected from a photo sharing platform. Visual content is processed using recent techniques for bag-of visual-words feature generation. A central contribution of our work is to infer the coding step of the general bag-of-word framework with such similarities and to aggregate these tag-codes by max-pooling to obtain a single representative vector (signature). Final image annotations are obtained via late fusion, where the three modalities (two text-based and one visual-based) are merged during the classification step. Experimental results on the Pascal VOC 2007 and MIR Flickr datasets show an improvement over the state-of-the-art methods, while significantly decreasing the computational complexity of the learning system.

A point of interest (POI) is a focused geographic entity such as a landmark, a school, an historical building, or a business. Points of interest are the basis for most of the data supporting location-based applications. In this paper we propose to curate POIs from online sources by bootstrapping training data from Web snippets, seeded by POIs gathered from social media. This large corpus is used to train a sequential tagger to recognize mentions of POIs in text. Using Wikipedia data as the training data, we can identify POIs in free text with an accuracy that is 116% better than the state of the art POI identifier in terms of precision, and 50% better in terms of recall. We show that using Foursquare and Gowalla checkins as seeds to bootstrap training data from Web snippets, we can improve precision between 16% and 52%, and recall between 48% and 187% over the state-of-the-art. The name of a POI is not sufficient, as the POI must also be associated with a set of geographic coordinates. Our method increases the number of POIs that can be localized nearly three-fold, from 134 to 395 in a sample of 400, with a median localization accuracy of less than one kilometer.

People often try to find an image using a short query and images are usually indexed using short annotations. Matching the query vocabulary with the indexing vocabulary is a difficult problem when little text is available. Textual user generated content in Web 2.0 platforms contains a wealth of data that can help solve this problem. Here we describe how to use Wikipedia and Flickr content to improve this match. The initial query is launched in Flickr and we create a query model based on co-occurring terms. We also calculate nearby concepts using Wikipedia and use these to expand the query. The final results are obtained by ranking the results for the expanded query using the similarity between their annotation and the Flickr model. Evaluation of these expansion and ranking techniques, over the Image CLEF 2010 Wikipedia Collection containing 237,434 images and their multilingual textual annotations, shows that a consistent improvement compared to state of the art methods.

Tourist photographs constitute a large part of the images uploaded to photo sharing platforms. But filtering methods are needed before one can extract useful knowledge from noisy user-supplied metadata. Here we show how to extract clean trip related information (what people visit, for how long, panoramic spots) from Flickr metadata. We illustrate our technique on a sample of metadata and images covering 183 cities of different size and from different parts of the world.

Web image search is inspired by text search techniques; it mainly relies on indexing textual data that surround the image file. But retrieval results are often noisy and image processing techniques have been proposed to rerank images. Unfortunately, these techniques usually imply a computational overload that makes the reranking process intractable in real time. We introduce here a lightweight reranking method that compares each result not only to the other query results but also to an external, contrastive class of items. The external class contains diversified images; the intuition supporting our approach is that results that are visually similar to other query results but dissimilar to elements of the contrastive class are likely to be good answers. The success of visual reranking depends on the visual coherence of queries; we measure this coherence in order to evaluate the chances of success. Visual reranking tends to emerge near duplicate images and we complement it with a diversification function which ensures that different aspects of a query are presented to the user. Our method is evaluated against a standard search engine using 210 diversified queries. Significant improvements are reported for both quantitative and qualitative tests.

Geographical gazetteers are necessary in a wide variety of applications. In the past, the construction of such gazetteers has been a tedious, manual process and only recently have the first attempts to automate the gazetteers creation been made. Here we describe our approach for mining accurate but large-scale multilingual geographic information by successively filtering information found in heterogeneous data sources (Flickr, Wikipedia, Panoramio, Web pages indexed by search engines). Statistically cross-checking information found in each site, we are able to identify new geographic objects, and to indicate, for each one, its name, its GPS coordinates, its encompassing regions (city, region, country), the language of the name, its popularity, and the type of the object (church, bridge, etc.). We evaluate our approach by comparing, wherever possible, our multilingual gazetteer to other known attempts at automatically building a geographic database and to Geonames, a manually built gazetteer.

Uploading tourist photos is a popular activity on photo sharing platforms. These photographs and their associated metadata (tags, geo-tags, and temporal information) should be useful for mining information about the sites visited. However, user-supplied metadata are often noisy and efficient filtering methods are needed before extracting useful knowledge. We focus here on exploiting temporal information, associated with tourist sites that appear in Flickr. From automatically filtered sets of geo-tagged photos, we deduce answers to questions like "how long does it take to visit a tourist attraction?" or "what can I visit in one day in this city?" Our method is evaluated and validated by comparing the automatically obtained visit duration times to manual estimations.

In this paper, we describe our midterm PhD work that focuses on role of large scale semantic resources in image retrieval. Current Internet picture search engines are based on a raw processing of the text associated to the images and this leads to an important degree of imprecision in the results sets, to an unstructured presentation of the answers and to poor interactivity options. The introduction of semantic layers in the retrieval frameworks may enhance the quality of results obtained with existing systems. One of the main challenges in the field is to develop architectures that include semantics and fit the requirements of real-life applications, like the Internet search engines, and this challenge constitutes a central point in our approach. The expected outcome of our approach is building picture search applications that better fit users' expectances.

In this paper we describe Olive, a concept based image search engine implemented using the WordNet nouns hierarchy. The system reformulates textual queries and performs an on the fly search for Google images corresponding to leaf nodes that are found under the currently demanded term. The retrieved pictures are rendered in a conceptually structured fashion and semantically related queries are presented to the user. In addition, a content based search in conceptually controlled neighborhoods is proposed.