22 results
Search Results
Now showing 1 - 10 of 22
Item Evaluating Deep Learning Methods for Low Resolution Point Cloud Registration in Outdoor Scenarios(The Eurographics Association, 2021) Siddique, Arslan; Corsini, Massimiliano; Ganovelli, Fabio; Cignoni, Paolo; Frosini, Patrizio and Giorgi, Daniela and Melzi, Simone and Rodolà , EmanuelePoint cloud registration is a fundamental task in 3D reconstruction and environment perception. We explore the performance of modern Deep Learning-based registration techniques, in particular Deep Global Registration (DGR) and Learning Multiview Registration (LMVR), on an outdoor real world data consisting of thousands of range maps of a building acquired by a Velodyne LIDAR mounted on a drone. We used these pairwise registration methods in a sequential pipeline to obtain an initial rough registration. The output of this pipeline can be further globally refined. This simple registration pipeline allow us to assess if these modern methods are able to deal with this low quality data. Our experiments demonstrated that, despite some design choices adopted to take into account the peculiarities of the data, more work is required to improve the results of the registration.Item IMGD: Image-based Multiscale Global Descriptors of Airborne LiDAR Point Clouds Used for Comparative Analysis(The Eurographics Association, 2021) Sreevalsan-Nair, Jaya; Mohapatra, Pragyan; Singh, Satendra; Frosini, Patrizio and Giorgi, Daniela and Melzi, Simone and Rodolà , EmanueleBoth geometric and semantic information are required for a complete understanding of regions acquired as three-dimensional (3D) point clouds using the Light Detection and Ranging (LiDAR) technology. However, the global descriptors of such datasets that integrate both the information types are rare. With a focus on airborne LiDAR point clouds, we propose a novel global descriptor that transforms the point cloud from Cartesian to barycentric coordinate spaces. We use both the probabilistic geometric classification, aggregated from multiple scales, and the semantic classification to construct our descriptor using point rendering. Thus, we get an image-based multiscale global descriptor, IMGD. To demonstrate its usability, we propose the use of distribution distance measures between the descriptors for comparing the point clouds. Our experimental results demonstrate the effectiveness of our descriptor, when constructed of publicly available datasets, and on applying our selected distance measures.Item A Geometric Approach for Computing the Kernel of a Polyhedron(The Eurographics Association, 2021) Sorgente, Tommaso; Biasotti, Silvia; Spagnuolo, Michela; Frosini, Patrizio and Giorgi, Daniela and Melzi, Simone and Rodolà , EmanueleWe present a geometric algorithm to compute the geometric kernel of a generic polyhedron. The geometric kernel (or simply kernel) is defined as the set of points from which the whole polyhedron is visible. Whilst the computation of the kernel for a polygon has already been largely addressed in the literature, less has been done for polyhedra. Currently, the principal implementation of the kernel estimation is based on the solution of a linear programming problem. We compare against it on several examples, showing that our method is more efficient in analysing the elements of a generic tessellation. Details on the technical implementation and discussions on pros and cons of the method are also provided.Item Visual Analysis of Popping in Progressive Visualization(The Eurographics Association, 2021) Waterink, Ethan; Kosinka, Jiri; Frey, Steffen; Frosini, Patrizio and Giorgi, Daniela and Melzi, Simone and Rodolà , EmanueleProgressive visualization allows users to examine intermediate results while they are further refined in the background. This makes them increasingly popular when dealing with large data and computationally expensive tasks. The characteristics of how preliminary visualizations evolve over time are crucial for efficient analysis; in particular unexpected disruptive changes between iterations can significantly hamper the user experience. This paper proposes a visualization framework to analyze the refinement behavior of progressive visualization. We particularly focus on sudden significant changes between the iterations, which we denote as popping artifacts, in reference to undesirable visual effects in the context of level of detail representations in computer graphics. Our visualization approach conveys where in image space and when during the refinement popping artifacts occur. It allows to compare across different runs of stochastic processes, and supports parameter studies for gaining further insights and tuning the algorithms under consideration. We demonstrate the application of our framework and its effectiveness via two diverse use cases with underlying stochastic processes: adaptive image space sampling, and the generation of grid layouts.Item STRONGER: Simple TRajectory-based ONline GEsture Recognizer(The Eurographics Association, 2021) Emporio, Marco; Caputo, Ariel; Giachetti, Andrea; Frosini, Patrizio and Giorgi, Daniela and Melzi, Simone and Rodolà , EmanueleIn this paper, we present STRONGER, a client-server solution for the online gesture recognition from captured hands' joints sequences. The system leverages a CNN-based recognizer improving current state-of-the-art solutions for segmented gestures classification, trained and tested for the online gesture recognition task on a recent benchmark including heterogeneous gestures. The recognizer provides good classification accuracy and a limited number of false positives on most of the gesture classes of the benchmark used and has been used to create a demo application in a Mixed Reality scenario using an Hololens 2 optical see through Head-Mounted Display with hand tracking capability.Item A High Quality 3D Controller for Mobile and Desktop Web Applications(The Eurographics Association, 2021) Fornari, Daniele; Malomo, Luigi; Cignoni, Paolo; Frosini, Patrizio and Giorgi, Daniela and Melzi, Simone and Rodolà , EmanueleThe interaction between a 2D input device (like a mouse or a touchscreen) and a 3D object on the screen with the purpose of examining it in detail is a well-studied interaction problem. The inherent difference in degrees of freedom between input devices and possible 3D transformations makes it difficult to intuitively map inputs to operations to be performed on 3D objects. Although, over the years, studies led to a wide variety of solutions to overcome this problem, most of them are not actually available in real-world applications. In particular, for 3D web applications, only basic solutions are often implemented, and even the most used web framework for 3D still lacks state of the art implementations. We will face the problem of 3D interaction through touch and mouse input, and we propose our implementation of a 3D view manipulator for web applications, which offers a natural control, advanced functionalities, and provides an easy-to-use interface for both desktop and mobile environments.Item Exploring Upper Limb Segmentation with Deep Learning for Augmented Virtuality(The Eurographics Association, 2021) Gruosso, Monica; Capece, Nicola; Erra, Ugo; Frosini, Patrizio and Giorgi, Daniela and Melzi, Simone and Rodolà , EmanueleSense of presence, immersion, and body ownership are among the main challenges concerning Virtual Reality (VR) and freehand-based interaction methods. Through specific hand tracking devices, freehand-based methods can allow users to use their hands for VE interaction. To visualize and make easy the freehand methods, recent approaches take advantage of 3D meshes to represent the user's hands in VE. However, this can reduce user immersion due to their unnatural correspondence with the real hands. We propose an augmented virtuality (AV) pipeline allows users to visualize their limbs in VE to overcome this limit. In particular, they were captured by a single monocular RGB camera placed in an egocentric perspective, segmented using a deep convolutional neural network (CNN), and streamed in the VE. In addition, hands were tracked through a Leap Motion controller to allow user interaction. We introduced two case studies as a preliminary investigation for this approach. Finally, both quantitative and qualitative evaluations of the CNN results were provided and highlighted the effectiveness of the proposed CNN achieving remarkable results in several real-life unconstrained scenarios.Item Mesh Colours for Gradient Meshes(The Eurographics Association, 2021) Baksteen, Sarah D.; Hettinga, Gerben J.; Echevarria, Jose; Kosinka, Jiri; Frosini, Patrizio and Giorgi, Daniela and Melzi, Simone and Rodolà , EmanueleWe present an extension of the popular gradient mesh vector graphics primitive with the addition of mesh colours, aiming to reduce the mesh complexity needed to describe intricate colour gradients and textures. We present interesting applications to user-guided authoring of detailed vector graphics and image vectorisation.Item SlowDeepFood: a Food Computing Framework for Regional Gastronomy(The Eurographics Association, 2021) Gilal, Nauman Ullah; Al-Thelaya, Khaled; Schneider, Jens; She, James; Agus, Marco; Frosini, Patrizio and Giorgi, Daniela and Melzi, Simone and Rodolà , EmanueleFood computing recently emerged as a stand-alone research field, in which artificial intelligence, deep learning, and data science methodologies are applied to the various stages of food production pipelines. Food computing may help end-users in maintaining healthy and nutritious diets by alerting of high caloric dishes and/or dishes containing allergens. A backbone for such applications, and a major challenge, is the automated recognition of food by means of computer vision. It is therefore no surprise that researchers have compiled various food data sets and paired them with well-performing deep learning architecture to perform said automatic classification. However, local cuisines are tied to specific geographic origins and are woefully underrepresented in most existing data sets. This leads to a clear gap when it comes to food computing on regional and traditional dishes. While one might argue that standardized data sets of world cuisine cover the majority of applications, such a stance would neglect systematic biases in data collection. It would also be at odds with recent initiatives such as SlowFood, seeking to support local food traditions and to preserve local contributions to the global variation of food items. To help preserve such local influences, we thus present a full end-to-end food computing network that is able to: (i) create custom image data sets semi-automatically that represent traditional dishes; (ii) train custom classification models based on the EfficientNet family using transfer learning; (iii) deploy the resulting models in mobile applications for real-time inference of food images acquired through smart phone cameras. We not only assess the performance of the proposed deep learning architecture on standard food data sets (e.g., our model achieves 91:91% accuracy on ETH’'s Food-101), but also demonstrate the performance of our models on our own, custom data sets comprising local cuisine, such as the Pizza-Styles data set and GCC-30. The former comprises 14 categories of pizza styles, whereas the latter contains 30 Middle Eastern dishes from the Gulf Cooperation Council members.Item An Information-theoretic Visual Analysis Framework for Convolutional Neural Networks(The Eurographics Association, 2021) Shen, Jingyi; Shen, Han-Wei; Frosini, Patrizio and Giorgi, Daniela and Melzi, Simone and Rodolà , EmanueleDespite the great success of Convolutional Neural Networks (CNNs) in Computer Vision and Natural Language Processing, the working mechanism behind CNNs is still under extensive discussion and research. Driven by strong demand for the theoretical explanation of neural networks, some researchers utilize information theory to provide insight into the black-box model. However, to the best of our knowledge, employing information theory to quantitatively analyze and qualitatively visualize neural networks has not been extensively studied in the visualization community. In this paper, we combine information entropies and visualization techniques to shed light on how CNN works. Specifically, we first introduce a data model to organize the data that can be extracted from CNN models. Then we propose two ways to calculate entropy under different circumstances. To provide a fundamental understanding of the basic building blocks of CNNs (e.g., convolutional layers, pooling layers, normalization layers) from an information-theoretic perspective, we develop a visual analysis system, CNNSlicer. CNNSlicer allows users to interactively explore the amount of information changes inside the model. With case studies on the widely used benchmark datasets (MNIST and CIFAR-10), we demonstrate the effectiveness of our system in opening the black-box of CNNs.