Abstract: Following the increasingly popular trend of social interaction analysis in egocentric vision, this article presents a comprehensive pipeline for automatic social pattern characterization of a wearable photo-camera user. The proposed framework relies merely on the visual analysis of egocentric photo-streams and consists of three major steps. The first step is to detect social interactions of the user where the impact of several social signals on the task is explored. The detected social events are inspected in the second step for categorization into different social meetings. These two steps act at event-level where each potential social event is modeled as a multi-dimensional time-series, whose dimensions correspond to a set of relevant features for each task; finally, LSTM is employed to classify the time-series. The last step of the framework is to characterize social patterns of the user. Our goal is to quantify the duration, the diversity and the frequency of the user social relations in various social situations. This goal is achieved by the discovery of recurrences of the same people across the whole set of social events related to the user. Experimental evaluation over EgoSocialStyle – the proposed dataset in this work, and EGO-GROUP demonstrates promising results on the task of social pattern characterization from egocentric photo-streams.

Abstract: We live in a society where the large majority of the population has a camera-equipped smartphone. In addition, hard drives and cloud storage are getting cheaper and cheaper, leading to a tremendous growth in stored personal photos. Unlike photo collections captured by a digital camera, which typically are pre-processed by the user who organizes them into event-related folders, smartphone pictures are automatically stored in the cloud. As a consequence, photo collections captured by a smartphone are highly unstructured and because smartphones are ubiquitous, they present a larger variability compared to pictures captured by a digital camera. To solve the need of organizing large smartphone photo collections automatically, we propose here a new methodology for hierarchical photo organization into topics and topic-related categories. Our approach successfully estimates latent topics in the pictures by applying probabilistic Latent Semantic Analysis, and automatically assigns a name to each topic by relying on a lexical database. Topic-related categories are then estimated by using a set of topic-specific Convolutional Neuronal Networks. To validate our approach, we ensemble and make public a large dataset of more than 8,000 smartphone pictures from 40 persons. Experimental results demonstrate major user satisfaction with respect to state of the art solutions in terms of organization.

Abstract: Recent progress in the field of human action recognition points towards the use of Spatio-TemporalInterestPoints (STIPs) for local descriptor-based recognition strategies. In this paper, we present a novel approach for robust and selective STIP detection, by applying surround suppression combined with local and temporal constraints. This new method is significantly different from existing STIP detection techniques and improves the performance by detecting more repeatable, stable and distinctive STIPs for human actors, while suppressing unwanted background STIPs. For action representation we use a bag-of-video words (BoV) model of local N-jet features to build a vocabulary of visual-words. To this end, we introduce a novel vocabulary building strategy by combining spatial pyramid and vocabulary compression techniques, resulting in improved performance and efficiency. Action class specific Support Vector Machine (SVM) classifiers are trained for categorization of human actions. A comprehensive set of experiments on popular benchmark datasets (KTH and Weizmann), more challenging datasets of complex scenes with background clutter and camera motion (CVC and CMU), movie and YouTube video clips (Hollywood 2 and YouTube), and complex scenes with multiple actors (MSR I and Multi-KTH), validates our approach and show state-of-the-art performance. Due to the unavailability of ground truth action annotation data for the Multi-KTH dataset, we introduce an actor specific spatio-temporal clustering of STIPs to address the problem of automatic action annotation of multiple simultaneous actors. Additionally, we perform cross-data action recognition by training on source datasets (KTH and Weizmann) and testing on completely different and more challenging target datasets (CVC, CMU, MSR I and Multi-KTH). This documents the robustness of our proposed approach in the realistic scenario, using separate training and test datasets, which in general has been a shortcoming in the performance evaluation of human action recognition techniques.

Abstract: Wearable cameras offer a hands-free way to record egocentric images of daily experiences, where social events are of special interest. The first step towards detection of social events is to track the appearance of multiple persons involved in them. In this paper, we propose a novel method to find correspondences of multiple faces in low temporal resolution egocentric videos acquired through a wearable camera. This kind of photo-stream imposes additional challenges to the multi-tracking problem with respect to conventional videos. Due to the free motion of the camera and to its low temporal resolution, abrupt changes in the field of view, in illumination condition and in the target location are highly frequent. To overcome such difficulties, we propose a multi-face tracking method that generates a set of tracklets through finding correspondences along the whole sequence for each detected face and takes advantage of the tracklets redundancy to deal with unreliable ones. Similar tracklets are grouped into the so called extended bag-of-tracklets (eBoT), which is aimed to correspond to a specific person. Finally, a prototype tracklet is extracted for each eBoT, where the occurred occlusions are estimated by relying on a new measure of confidence. We validated our approach over an extensive dataset of egocentric photo-streams and compared it to state of the art methods, demonstrating its effectiveness and robustness.

Abstract: Bag of visual words (BoVW) models are widely utilized in image/ video representation and recognition. The cornerstone of these models is the encoding stage, in which local features are decomposed over a codebook in order to obtain a representation of features. In this paper, we propose a new encoding algorithm by jointly encoding the set of local descriptors of each sample and considering the locality structure of descriptors. The proposed method takes advantages of locality coding such as its stability and robustness to noise in descriptors, as well as the strengths of the group coding strategy by taking into account the potential relation among descriptors of a sample. To efficiently implement our proposed method, we consider the Alternating Direction Method of Multipliers (ADMM) framework, which results in quadratic complexity in the problem size. The method is employed for a challenging classification problem: action recognition by depth cameras. Experimental results demonstrate the outperformance of our methodology compared to the state-of-the-art on the considered datasets.

Abstract: Deformable object models capture variations in an object’s appearance that can be represented as image deformations. Other effects such as out-of-plane rotations, three-dimensional articulations, and self-occlusions are often captured by considering mixture of deformable models, one per object aspect. A more scalable approach is representing instead the variations at the level of the object parts, applying the concept of a mixture locally. Combining a few part variations can in fact cheaply generate a large number of global appearances.
A limited version of this idea was proposed by Yang and Ramanan [1], for human pose dectection. In this paper we apply it to the task of generic object category detection and extend it in several ways. First, we propose a model for the relationship between part appearances more general than the tree of Yang and Ramanan [1], which is more suitable for generic categories. Second, we treat part locations as well as their appearance as latent variables so that training does not need part annotations but only the object bounding boxes. Third, we modify the weakly-supervised learning of Felzenszwalb et al. and Girshick et al. [2], [3] to handle a significantly more complex latent structure.
Our model is evaluated on standard object detection benchmarks and is found to improve over existing approaches, yielding state-of-the-art results for several object categories.

Abstract: JCR 2019 Q2, IF=3.121
The task of unpaired image-to-image translation is highly challenging due to the lack of explicit cross-domain pairs of instances. We consider here diverse image translation (DIT), an even more challenging setting in which an image can have multiple plausible translations. This is normally achieved by explicitly disentangling content and style in the latent representation and sampling different styles codes while maintaining the image content. Despite the success of current DIT models, they are prone to suffer from bias. In this paper, we study the problem of bias in image-to-image translation. Biased datasets may add undesired changes (e.g. change gender or race in face images) to the output translations as a consequence of the particular underlying visual distribution in the target domain. In order to alleviate the effects of this problem we propose the use of semantic constraints that enforce the preservation of desired image properties. Our proposed model is a step towards unbiased diverse image-to-image translation (UDIT), and results in less unwanted changes in the translated images while still performing the wanted transformation. Experiments on several heavily biased datasets show the effectiveness of the proposed techniques in different domains such as faces, objects, and scenes.

Abstract: We address the source-free domain adaptation (SFDA) problem, where only the source model is available during adaptation to the target domain. We consider two settings: the offline setting where all target data can be visited multiple times (epochs) to arrive at a prediction for each target sample, and the online setting where the target data needs to be directly classified upon arrival. Inspired by diverse classifier based domain adaptation methods, in this paper we introduce a second classifier, but with another classifier head fixed. When adapting to the target domain, the additional classifier initialized from source classifier is expected to find misclassified features. Next, when updating the feature extractor, those features will be pushed towards the right side of the source decision boundary, thus achieving source-free domain adaptation. Experimental results show that the proposed method achieves competitive results for offline SFDA on several benchmark datasets compared with existing DA and SFDA methods, and our method surpasses by a large margin other SFDA methods under online source-free domain adaptation setting.