Publications by Lorenzo Baraldi

Data-driven saliency has recently gained a lot of attention thanks to the use of Convolutional Neural Networks for predicting gaze fixations. In this paper we go beyond standard approaches to saliency prediction, in which gaze maps are computed with a feed-forward network, and present a novel model which can predict accurate saliency maps by incorporating neural attentive mechanisms. The core of our solution is a Convolutional LSTM that focuses on the most salient regions of the input image to iteratively refine the predicted saliency map. Additionally, to tackle the center bias typical of human eye fixations, our model can learn a set of prior maps generated with Gaussian functions. We show, through an extensive evaluation, that the proposed architecture outperforms the current state of the art on public saliency prediction datasets. We further study the contribution of each key component to demonstrate their robustness on different scenarios.

The prediction of human eye fixations has been recently gaining a lot of attention thanks to the improvements shown by deep architectures. In our work, we go beyond classical feed-forward networks to predict saliency maps and propose a Saliency Attentive Model which incorporates neural attention mechanisms to iteratively refine predictions. Experiments demonstrate that the proposed strategy overcomes by a considerable margin the state of the art on the largest dataset available for saliency prediction. Here, we provide experimental results on other popular saliency datasets to confirm the effectiveness and the generalization capabilities of our model, which enable us to reach the state of the art on all considered datasets.

We present a novel video browsing and retrieval system for edited videos, in which videos are automatically decomposed into meaningful and storytelling parts (i.e. scenes) and tagged according to their transcript. The system relies on a Triplet Deep Neural Network which exploits multimodal features, and has been implemented as a set of extensions to the eXo Platform Enterprise Content Management System (ECMS). This set of extensions enable the interactive visualization of a video, its automatic and semi-automatic annotation, as well as a keyword-based search inside the video collection. The platform also allows a natural integration with third-party add-ons, so that automatic annotations can be exploited outside the proposed platform.

Estimating the focus of attention of a person looking at an image or a video is a crucial step which can enhance many vision-based inference mechanisms: image segmentation and annotation, video captioning, autonomous driving are some examples. The early stages of the attentive behavior are typically bottom-up; reproducing the same mechanism means to find the saliency embodied in the images, i.e. which parts of an image pop out of a visual scene. This process has been studied for decades in neuroscience and in terms of computational models for reproducing the human cortical process. In the last few years, early models have been replaced by deep learning architectures, that outperform any early approach compared against public datasets. In this paper, we propose a discussion on why convolutional neural networks (CNNs) are so accurate in saliency prediction. We present our DL architectures which combine both bottom-up cues and higher-level semantics, and incorporate the concept of time in the attentional process through LSTM recurrent architectures. Eventually, we present a video-specific architecture based on the C3D network, which can extracts spatio-temporal features by means of 3D convolutions to model task-driven attentive behaviors. The merit of this work is to show how these deep networks are not mere brute-force methods tuned on massive amount of data, but represent well-defined architectures which recall very closely the early saliency models, although improved with the semantics learned by human ground-thuth.

The use of Recurrent Neural Networks for video captioning has recently gained a lot of attention, since they can be used both to encode the input video and to generate the corresponding description. In this paper, we present a recurrent video encoding scheme which can discover and leverage the hierarchical structure of the video. Unlike the classical encoder-decoder approach, in which a video is encoded continuously by a recurrent layer, we propose a novel LSTM cell, which can identify discontinuity points between frames or segments and modify the temporal connections of the encoding layer accordingly. We evaluate our approach on three large-scale datasets: the Montreal Video Annotation dataset, the MPII Movie Description dataset and the Microsoft Video Description Corpus. Experiments show that our approach can discover appropriate hierarchical representations of input videos and improve the state of the art results on movie description datasets.

Automatic layout analysis has proven to be extremely important in the process of digitization of large amounts of documents. In this paper we present a mixed approach to layout analysis, introducing a SVM-aided layout segmentation process and a classification process based on local and geometrical features. The final output of the automatic analysis algorithm is a complete and structured annotation in JSON format, containing the digitalized text as well as all the references to the illustrations of the input page, and which can be used by visualization interfaces as well as annotation interfaces. We evaluate our algorithm on a large dataset built upon the first volume of the “Enciclopedia Treccani”.

In this paper, we propose a multimodal deep learning architecture for emotion recognition in video regarding our participation to the audio-video based sub-challenge of the Emotion Recognition in the Wild 2017 challenge. Our model combines cues from multiple video modalities, including static facial features, motion patterns related to the evolution of the human expression over time, and audio information. Specifically, it is composed of three sub-networks trained separately: the first and second ones extract static visual features and dynamic patterns through 2D and 3D Convolutional Neural Networks (CNN), while the third one consists in a pretrained audio network which is used to extract useful deep acoustic signals from video. In the audio branch, we also apply Long Short Term Memory (LSTM) networks in order to capture the temporal evolution of the audio features. To identify and exploit possible relationships among different modalities, we propose a fusion network that merges cues from the different modalities in one representation. The proposed architecture outperforms the challenge baselines (38.81% and 40.47%): we achieve an accuracy of 50.39% and 49.92% respectively on the validation and the testing data.

In the last years video has been swamping the Internet: websites, social networks, and business multimedia systems are adopting video as the most important form of communication and information. Video are normally accessed as a whole and are not indexed in the visual content. Thus, they are often uploaded as short, manually cut clips with user-provided annotations, keywords and tags for retrieval. In this paper, we propose a prototype multimedia system which addresses these two limitations: it overcomes the need of human intervention in the video setting, thanks to fully deep learning-based solutions, and decomposes the storytelling structure of the video into coherent parts. These parts can be shots, key-frames, scenes and semantically related stories, and are exploited to provide an automatic annotation of the visual content, so that parts of video can be easily retrieved. This also allows a principled re-use of the video itself: users of the platform can indeed produce new storytelling by means of multi-modal presentations, add text and other media, and propose a different visual organization of the content. We present the overall solution, and some experiments on the re-use capability of our platform in edutainment by conducting an extensive user valuation %with students from primary schools.

In this paper, we propose a novel scene detection algorithm which employs semantic, visual, textual and audio cues. We also show how the hierarchical decomposition of the storytelling video structure can improve retrieval results presentation with semantically and aesthetically effective thumbnails. Our method is built upon two advancements of the state of the art: 1) semantic feature extraction which builds video specific concept detectors; 2) multimodal feature embedding learning, that maps the feature vector of a shot to a space in which the Euclidean distance has task specific semantic properties. The proposed method is able to decompose the video in annotated temporal segments which allow for a query specific thumbnail extraction. Extensive experiments are performed on different data sets to demonstrate the effectiveness of our algorithm. An in-depth discussion on how to deal with the subjectivity of the task is conducted and a strategy to overcome the problem is suggested.