A5016675020- 3D Shape Modeling and Analysis
- Remote Sensing and LiDAR Applications
- Machine Learning and Algorithms
- Computer Graphics and Visualization Techniques
- Advanced Neural Network Applications
- Identity, Memory, and Therapy
- Human Pose and Action Recognition
- Optical measurement and interference techniques
- Industrial Vision Systems and Defect Detection
- 3D Surveying and Cultural Heritage
- Multimodal Machine Learning Applications
- Video Analysis and Summarization
- Machine Learning and ELM
- Advanced Graph Neural Networks
- Image and Object Detection Techniques
- Cardiovascular Health and Disease Prevention
- Advanced Vision and Imaging
- Optical Polarization and Ellipsometry
- Image Processing and 3D Reconstruction
- Medical Imaging and Analysis
- Machine Learning and Data Classification
- Adversarial Robustness in Machine Learning
- Neural Networks and Applications
- Ergonomics and Musculoskeletal Disorders
Polytechnic University of Turin
2020-2024
Turin Polytechnic University
2020
Point clouds are an increasingly relevant geometric data type but they often corrupted by noise and affected the presence of outliers. We propose a deep learning method that can simultaneously denoise point cloud remove outliers in single model. The core proposed is graph-convolutional neural network able to efficiently deal with irregular domain permutation invariance problem typical clouds. fully-convolutional build complex hierarchies features dynamically constructing neighborhood graphs...
Our comprehension of video streams depicting human activities is naturally multifaceted: in just a few moments, we can grasp what happening, identify the relevance and interactions objects scene, forecast will happen soon, everything all at once. To endow autonomous systems with such holistic perception, learning how to correlate concepts, abstract knowledge across diverse tasks, leverage tasks synergies when novel skills essential. A significant step this direction EgoPack, unified...
Continual learning (CL) is the sub-field of machine concerned with accumulating knowledge in dynamic environments. So far, CL research has mainly focused on incremental classification tasks, where models learn to classify new categories while retaining previously learned ones. Here, we argue that maintaining such a focus limits both theoretical development and practical applicability methods. Through detailed analysis concrete examples - including multi-target classification, robotics...
Deep neural networks for graphs have emerged as a powerful tool learning on complex non-euclidean data, which is becoming increasingly common variety of different applications.Yet, although their potential has been widely recognised in the machine community, graph largely unexplored downstream tasks such robotics applications.To fully unlock potential, hence, we propose review architectures from perspective.The paper covers fundamentals graph-based models, including architecture, training...
Existing end-to-end signal compression schemes using neural networks are largely based on an autoencoder-like structure, where a universal encoding function creates compact latent space and the representation in this is quantized stored. Recently, advances from field of 3D graphics have shown possibility building implicit networks, i.e., returning value at given query coordinate. In paper, we propose representations as novel paradigm for with defined by very weights network. We discuss how...
Surface normal estimation is a basic task for many point cloud processing algorithms. However, it can be challenging to capture the local geometry of data, especially in presence noise. Recently, deep learning approaches have shown promising results. Nevertheless, applying convolutional neural networks clouds not straightforward, due irregular positioning points. In this paper, we propose method based on graph-convolutional deal with such domain. The layers build hierarchies localized...
Recent transformer-based architectures have shown impressive results in the field of image segmentation. Thanks to their flexibility, they obtain outstanding performance multiple segmentation tasks, such as semantic and panoptic, under a single unified framework. To achieve performance, these employ intensive operations require substantial computational resources, which are often not available, especially on edge devices. fill this gap, we propose Prototype-based Efficient MaskFormer (PEM),...
Human comprehension of a video stream is naturally broad: in few instants, we are able to understand what happening, the relevance and relationship objects, forecast will follow near future, everything all at once. We believe that - effectively transfer such an holistic perception intelligent machines important role played by learning correlate concepts abstract knowledge coming from different tasks, synergistically exploit them when novel skills. To accomplish this, seek for unified...
Neural Networks design is a complex and often daunting task, particularly for resource-constrained scenarios typical of mobile-sized models. Architecture Search promising approach to automate this process, but existing competitive methods require large training time computational resources generate accurate To overcome these limits, paper contributes with: i) novel training-free metric, named Entropic Score, estimate model expressivity through the aggregated element-wise entropy its...
Deep neural networks for graphs have emerged as a powerful tool learning on complex non-euclidean data, which is becoming increasingly common variety of different applications. Yet, although their potential has been widely recognised in the machine community, graph largely unexplored downstream tasks such robotics To fully unlock potential, hence, we propose review architectures from perspective. The paper covers fundamentals graph-based models, including architecture, training procedures,...
Neural Networks design is a complex and often daunting task, particularly for resource-constrained scenarios typical of mobile-sized models. Architecture Search promising approach to automate this process, but existing competitive methods require large training time computational resources generate accurate To overcome these limits, paper contributes with: i) novel training-free metric, named Entropic Score, estimate model expressivity through the aggregated element-wise entropy its...
Point clouds are an increasingly relevant data type but they often corrupted by noise. We propose a deep neural network based on graph-convolutional layers that can elegantly deal with the permutation-invariance problem encountered learning-based point cloud processing methods. The is fully-convolutional and build complex hierarchies of features dynamically constructing neighborhood graphs from similarity among high-dimensional feature representations points. When coupled loss promoting...