Archivio della ricerca - Fondazione Bruno Kessler
Not a member yet
21227 research outputs found
Sort by
mPickle: Pickle for MicroPython
No full textThe rapid deployment of AI and ML at the far edge, particularly on microcontrollers, requires efficient serialization of complex data structures. Python’s pickle module is widely used but lacks native support in MicroPython, hindering seamless data transfer. We introduce mPickle, a MicroPython-compatible library enabling memory-optimized serialization of complex Python objects, supporting TinyMLOps workflows that transfer model artifacts from CPython to MicroPython targets. mPickle implements the Pickle Protocol 4 with a module/function mapping layer that enables interoperable binary serialization and deserialization between CPython and MicroPython. The library is validated via unit tests, benchmarks, and examples covering built-in types, custom classes, numerical arrays, and model-weight dictionaries
Workflows for analysing and utilizing large-scale 3D datasets of cultural heritage
No full textFor 3D content, large-scale datasets, such as Objaverse or ShapeNet, and repositories, such as Sketchfab, have been compiled. Within the European 3DBigDataSpace project, a consortium of 10 partners assesses open licensed 3D models to select and retrieve those models particularly representing cultural heritage objects in Europe to aggregate them into the European Data Space. As key part of that work is the classification and geolocalization of 3D content, with mesh models viewable via different viewers and tested in different scenarios such as museum exhibitions, cultural tourism, or education. This article highlights the steps taken (1) to compile a large-scale pool of 3D assets of cultural heritage and ready-to-use viewer applications and (2) to enrich and (3) utilize them in various settings
Toward Detailed and Accurate Forest Inventory with Multi-Source Lidar Data
No full textAccurate forest inventories are crucial for ecological research and resource management. A novel workflow is introduced to fuse lidar point clouds from terrestrial, mobile, and unmanned aerial vehicle systems, significantly improving the characterization of forest structure. The proposed method uses a robust coarse-to-fine registration and a quantitative voxel-based metric to evaluate the scanning completeness of single and fused data sets to derive accurate individual tree segmentation and forest parameter results. The work is validated using a unique data set (released on paper acceptance), and it demonstrates that fusing multi-source lidar data provides a more complete and accurate representation of forest ecosystems, underscoring its potential for more effective ecological monitoring and sustainable forest management
Critical Aspects in the Modeling of Sub-GeV Calorimetric Particle Detectors: The Case Study of the High-Energy Particle Detector (HEPD-02) on Board the CSES-02 Satellite
No full textThe accurate simulation of sub-GeV particle detectors is essential for interpreting experimental data and optimizing detector design. This work identifies and addresses several critical aspects in modeling such detectors, taking as a case study the High-Energy Particle Detector (HEPD-02), a space-borne instrument developed within the CSES-02 mission to measure electrons in the ∼3–100 MeV range, protons and light nuclei in the ∼30–200 MeV/n. The HEPD-02 instrument consists of a silicon tracker, plastic and LYSO scintillator calorimeters, and anticoincidence systems, making it a representative example of a complex low-energy particle detector operating in Low Earth Orbit. Key challenges arise from replicating intricate detector geometries derived from CAD models, selecting appropriate hadronic physics lists for low-energy interactions, and accurately describing the detector response—particularly quenching effects in scintillators and digitization in solid-state tracking planes. Particular attention is given to three critical aspects: the precise CAD-level geometry implementation, the impact of hadronic physics models on the detector response, and the parameterization of scintillation quenching. In this study, we present original solutions to these challenges and provide data–MC comparisons using data from HEPD-02 beam tests
When Speaking of Risks Meets Responsibility: The European Ethical Framework for Shared Decision-Making in Transplantation
No full text
Combining 3D Urban Objects from All Around the World to Improve Object Classification and Semantic Segmentation
No full textGiven the growing number of applications in urban planning and large-scale digital twins, the development of effective solutions for urban point cloud classification is of extreme interest for the R&D community and commercial sector. State-of-the-art neural networks commonly lack adequate cross-dataset generalisation ability, mainly due to varying sensors and data collection platforms, object shape differences, as well as the presence of under-represented objects and imbalanced classes, especially in case of dense and high-resolution reality-based 3D data. This work demonstrates how the recently released ESTATE dataset (A large dataset of under-represented urban objects—https://github.com/3DOM-FBK/ESTATE), full of thousands of under-represented urban objects, such as traffic lights, electrical poles, pylons, and ventilation units, spread over 13 classes, can improve the performance of state-of-the-art point cloud classification algorithms. Experiments with different neural networks and several testing configurations with sensor-specific inputs (coordinate, intensity, and colour) show the effectiveness of this dataset in enhancing the classification capabilities and increasing cross-dataset generalisation. Moreover, reported results show not only the adaptation of object classification networks to the semantic segmentation pipeline, but also an improvement of semantic segmentation performance by increasing the distribution of under-represented classes with the ESTATE dataset
