Research
I am broadly interested in computer vision and machine learning, especially in self-driving scenarios. The goal of my research is to combine geometry and learning for efficient, large-scale 3D understanding with cheap sensors.
|
|
Geometry and Learning for Efficient 3D Perception
Marco Orsingher
PhD Thesis, 2024
Thesis /
BibTeX
A collection of previous works, with additional details, visualizations and experiments, as well as the proper context on how they fit into my research goals.
|
|
Informative Rays Selection for Few-Shot Neural Radiance Fields
Marco Orsingher,
Anthony Dell'Eva,
Paolo Zani,
Paolo Medici,
Massimo Bertozzi
VISAPP, 2024
Paper /
Video /
Code /
BibTeX
We analyze the sampling efficiency of NeRF and present two strategies to optimize it, given a limited training budget.
|
|
Learning Neural Radiance Fields from Multi-View Geometry
Marco Orsingher,
Paolo Zani,
Paolo Medici,
Massimo Bertozzi
ECCV Workshop, 2022
Paper /
Video /
BibTeX
We use classical 3D reconstruction as pseudo-supervision for training NeRF and obtain much cleaner surfaces.
|
|
Arbitrary Point Cloud Upsampling with Spherical Mixture of Gaussians
Anthony Dell'Eva (*),
Marco Orsingher (*),
Massimo Bertozzi
(*Equal Contribution)
3DV, 2022   (Oral Presentation)
Project Page /
Paper /
Video /
Code /
BibTeX
We show that Transformers can learn to generate point clouds with arbitrary resolution from sparse raw data.
|
|
Revisiting PatchMatch Multi-View Stereo for Urban 3D Reconstruction
Marco Orsingher,
Paolo Zani,
Paolo Medici,
Massimo Bertozzi
IEEE IV, 2022
Paper /
BibTeX
We propose a framework for monocular 3D reconstruction that combines visual SLAM, multi-scale PatchMatch MVS with planar priors and confidence-based graph optimization.
|
|
Efficient View Clustering and Selection for City-Scale 3D Reconstruction
Marco Orsingher,
Paolo Zani,
Paolo Medici,
Massimo Bertozzi
ICIAP, 2021   (Oral Presentation)
Paper /
Video /
BibTeX
We design a view selection algorithm that scales linearly with the number of clusters and allows to reconstruct efficiently entire cities.
|
|