Browsing by Author "Lombardi, Marco"

Now showing 1 - 3 of 3
  • No Thumbnail Available
    Item
    3DReg-i-Net: Improving Deep Learning Based 3D Registration for a Robust Real-time Alignment of Small-scale Scans
    (The Eurographics Association, 2019) Lombardi, Marco; Riccardi, Andrea; Savardi, Mattia; Signoroni, Alberto; Agus, Marco and Corsini, Massimiliano and Pintus, Ruggero
    We present 3DReg-i-Net, an improved deep learning solution for pairwise registration of 3D scans, which evolves the recently proposed 3DRegNet technique by Pais et al. This is one of the very first learning based algorithm aiming at producing the co-registration of two 3D views starting solely from a set of point correspondences, which is able to perform outlier rejection and to recover the registration matrix. We evolve the original method to face the challenging scenario of quick 3D modelling at small scales through the alignment of dense 3D views acquired at video frame-rate with a handheld scanner. We improve the system tracking robustness and alignment performance with a generalized input data augmentation. Moreover, working on suboptimal aspects of the original solution, we propose different improvements that lead to a redefinition of the training loss function. When tested on the considered scenario, the proposed 3DReg-i-Net significantly outperforms the prior solution in terms of accuracy of the estimated aligning transforms.
  • No Thumbnail Available
    Item
    Deep Tracking for Robust Real-time Object Scanning
    (The Eurographics Association, 2022) Lombardi, Marco; Savardi, Mattia; Signoroni, Alberto; Cabiddu, Daniela; Schneider, Teseo; Allegra, Dario; Catalano, Chiara Eva; Cherchi, Gianmarco; Scateni, Riccardo
    Nowadays, a high-fidelity 3d model representation can be obtained easily by means of handheld optical scanners, which offer a good level of reconstruction quality, portability, and low latency in scan-to-data. However, it is well known that the tracking process can be critical for such devices: sub-optimal lighting conditions, smooth surfaces in the scene, or occluded views and repetitive patterns are all sources of error. In this scenario, recent disruptive technologies such as sparse convolutional neural networks have been tailored to address common problems in 3D vision and analysis. Our work aims to integrate the most promising solutions into an operating framework which can then be used to achieve compelling results in 3D real-time reconstruction. Several scenes from a dataset containing dense views of objects are tested using our proposed pipeline and compared with the current state-of-the-art of online reconstruction.
  • No Thumbnail Available
    Item
    Deep-learning Alignment for Handheld 3D Acquisitions: A new Densematch Dataset for an Extended Comparison
    (The Eurographics Association, 2020) Lombardi, Marco; Savardi, Mattia; Signoroni, Alberto; Biasotti, Silvia and Pintus, Ruggero and Berretti, Stefano
    Promising solutions for the alignment of 3D views based on representation learning approaches have been proposed very recently. The potentials of these solutions that could positively affect the 3D object registration has yet to be extensively tested. In fact, a direct comparison among advisable technologies is still lacking, especially if the focus is on different data types and real-time application requirements. This work is a first contribution in this direction since we perform an independent extended comparison among prominent deep learning-driven 3D view alignment solutions by considering two relevant setups: 1) data coming from commodity 3D sensors, and 2) denser data coming from a handheld 3D optical scanner. While for the first scenario reference datasets exist, we collect and release the new benchmark dataset DenseMatch for the second setup.