Object detection is arguably one of the most important and complex tasks to enable the advent of next-generation autonomous systems. Recent advancements in deep learning techniques allowed a significant improvement in detection accuracy and latency of modern neural networks, allowing their adoption in automotive, avionics and industrial embedded systems, where performances are required to meet size, weight and power constraints. Multiple benchmarks and surveys exist to compare stateof-the-art detection networks, profiling important metrics, like precision, latency and power efficiency on Commercial-off-the-Shelf (COTS) embedded platforms. However, we observed a fundamental lack of fairness in the existing comparisons, with a number of implicit assumptions that may significantly bias the metrics of interest. This includes using heterogeneous settings for the input size, training dataset, threshold confidences, and, most importantly, platform-specific optimizations, that are especially important when assessing latency and energy-related values. The lack of uniform comparisons is mainly due to the significant effort required to re-implement network models, whenever openly available, on the specific platforms, to properly configure the available acceleration engines for optimizing performance, and to retrain the model using a homogeneous dataset. This paper aims at filling this gap, providing a comprehensive and fair comparison of the best-in-class Convolution Neural Networks (CNNs) for real-time embedded systems, detailing the effort made to achieve an unbiased characterization on cutting-edge system-on-chips. Multi-dimensional trade-offs are explored for achieving a proper configuration of the available programmable accelerators for neural inference, adopting the best available software libraries. To stimulate the adoption of fair benchmarking assessments, the framework is released to the public in an open source repository.

Combining Natural Language with Vision represents a unique and interesting challenge in the domain of Artificial Intelligence. The AI City Challenge Track 5 for Natural Language-Based Vehicle Retrieval focuses on the problem of combining visual and textual information, applied to a smart-city use case. In this paper, we present All You Can Embed (AYCE), a modular solution to correlate singlevehicle tracking sequences with natural language. The main building blocks of the proposed architecture are (i) BERT to provide an embedding of the textual descriptions, (ii) a convolutional backbone along with a Transformer model to embed the visual information. For the training of the retrieval model, a variation of the Triplet Margin Loss is proposed to learn a distance measure between the visual and language embeddings. The code is publicly available at https://github.com/cscribano/AYCE_2021.

A major limitation of the classical random Boolean network model of gene regulatory networks is its synchronous updating, which implies that all the proteins decay at the same rate. Here a model is discussed, where the network is composed of two different sets of nodes, labelled G and P with reference to "genes" and "proteins". Each gene corresponds to a protein (the one it codes for), while several proteins can simultaneously affect the expression of a gene. Both kinds of nodes take Boolean values. If we look at the genes only, it is like adding some memory terms, so the new state of the gene subnetwork network does no longer depend upon its previous state only. In general, these terms tend to make the dynamics of the network more ordered than that of the corresponding memoryless network. The analysis is focused here mostly on dynamical critical states. It has been shown elsewhere that the usual way of computing the Derrida parameter, starting from purely random initial conditions, can be misleading in strongly non-ergodic systems. So here the effects of perturbations on both genes' and proteins' levels is analysed, using both the canonical Derrida procedure and an "extended" one. The results are discussed. Moreover, the stability of attractors is also analysed, measured by counting the fraction of perturbations where the system eventually falls back onto the initial attractor.

Convolutional Neural Networks are widely known and used architectures in image processing contexts, in particular for medical images. These Deep Learning techniques, known for their ability to extract high-level features, almost always require a labeled dataset, a process that can be computationally expensive. Most of the time in the biomedical context, when images are used they are noisy and the ground-truth is unknown. For this reason, and in the context of Green Artificial Intelligence, recently, an unsupervised method that employs Convolutional Neural Networks, or more precisely autoencoders, has appeared in the panorama of Deep Learning. This technique, called Deep Image Prior (DIP) by the authors, can be used in areas such as denoising, superresolution, and inpainting. Starting from these assumptions, this work analyses the robustness of these networks with respect to different types of initialization. First of all, we analyze the different types of parameters: related to the Batch Norm and the Convolutional layers. For the results, we focus on the speed of convergence and the maximum performance obtained. However, this paper aims to apply acquired information on Computer Tomography noised images. In fact, the final purpose is to test the best initializations of the first phase on a phantom image and then on a real Computer Tomography one. In fact, Computer Tomography together with Magnetic Resonance Imaging and Positron Emission Tomography are some of the diagnostic tools currently available to neuroscientists and oncologists. This work shows how initializations affect final performances and, in addition, how they should be used in the medical image reconstruction field. The section on numerical experiments shows results that on the one hand confirm the importance of a good initialization to obtain fast convergence and high performance; on the other hand, it shows how the method is robust to the processing of different image types: natural and medical. Not a single good initialization is discovered, but many of them could be chosen, according to specific necessities of the single problem.

Object pose estimation underwater allows an autonomous system to perform tracking and intervention tasks. Nonetheless, underwater target pose estimation is remarkably challenging due to, among many factors, limited visibility, light scattering, cluttered environments, and constantly varying water conditions. An approach is to employ sonar or laser sensing to acquire 3D data, but besides being costly, the resulting data is normally noisy. For this reason, the community has focused on extracting pose estimates from RGB input. However, the literature is scarce and exhibits low detection accuracy. In this work, we propose an approach consisting of a 2D object detection and a 6D pose estimation that reliably obtains object poses in different underwater scenarios. To test our pipeline, we collect and make available a dataset of 4 objects in 10 different real scenes with annotations for object detection and pose estimation. We test our proposal in real and synthetic settings and compare its performance with similar end-to-end methodologies for 6D object pose estimation. Our dataset contains some challenging objects with symmetrical shapes and poor texture. Regardless of such object characteristics, our proposed method outperforms stateof-the-art pose accuracy by ∼ 8%. We finally demonstrate the reliability of our pose estimation pipeline by doing experiments with an underwater manipulation in a reaching task.

The luxury car market has demanding product development standards aimed at providing state-of-the-art features in the automotive domain. Handling performance is amongst the most important properties that must be assessed when developing a new car model. In this work, we analyse the problem of predicting subjective evaluations of automobiles handling performances from objective records of driving sessions. A record is a multi-dimensional time series describing the temporal evolution of the mechanical state of an automobile. A categorical variable quantifies the evaluations of handling properties. We describe an original deep learning system, featuring a denoising autoencoder and hierarchical attention mechanisms, that we designed to solve this task. Attention mechanisms intrinsically compute probability distributions over their inputs’ components. Combining this feature with the saliency maps technique, our system can compute heatmaps that provide a visual aid to identify the physical events conditioning its predictions.

No abstract

No abstract