In this paper, we present SCENE-Net V2, a new resource-efficient, gray-box model for multiclass 3D scene understanding. SCENE-Net V2 leverages Group Equivariant Non-Expansive Operators (GENEOs) to incorporate fundamental geometric priors as inductive biases, offering a more transparent alternative to the prevalent black-box models in the domain. This model addresses the limitations of its white-box predecessor, SCENE-Net, by expanding its applicability from pole-like structures to a wider range of datasets with detailed 3D elements. Our model achieves the sweet-spot between application and transparency: SCENENet V2 is a general method for object identification with interpretability guarantees. Our experimental results demonstrate that SCENE-Net V2 achieves competitive performance with a significantly lower parameter count. Furthermore, we propose the use of GENEO-based architectures as a feature extraction tool for black-box models, enabling an increase in performance by adding a minimal number of meaningful parameters.

SCENE-Net V2: Interpretable Multiclass 3D Scene Understanding with Geometric Priors / D. Lavado, C. Soares, A. Micheletti (PROCEEDINGS OF MACHINE LEARNING RESEARCH). - In: Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM) at ICML 2024, 29 July 2024, Vienna, Austria / [a cura di] S. Vadgama, E. Bekkers, A. Pouplin, S.O. Kaba, R. Walters, H. Lawrence, T. Emerson, H. Kvinge, J. Tomczak, S. Jegelka. - [s.l] : PMLR, 2024 Dec. - pp. 222-232 (( convegno ELLIS Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM) tenutosi a Wien nel 2024.

SCENE-Net V2: Interpretable Multiclass 3D Scene Understanding with Geometric Priors

D. Lavado
Primo
;
A. Micheletti
Ultimo
2024

Abstract

In this paper, we present SCENE-Net V2, a new resource-efficient, gray-box model for multiclass 3D scene understanding. SCENE-Net V2 leverages Group Equivariant Non-Expansive Operators (GENEOs) to incorporate fundamental geometric priors as inductive biases, offering a more transparent alternative to the prevalent black-box models in the domain. This model addresses the limitations of its white-box predecessor, SCENE-Net, by expanding its applicability from pole-like structures to a wider range of datasets with detailed 3D elements. Our model achieves the sweet-spot between application and transparency: SCENENet V2 is a general method for object identification with interpretability guarantees. Our experimental results demonstrate that SCENE-Net V2 achieves competitive performance with a significantly lower parameter count. Furthermore, we propose the use of GENEO-based architectures as a feature extraction tool for black-box models, enabling an increase in performance by adding a minimal number of meaningful parameters.
Semantic segmentation; GENEO; Multiclass classification
Settore INFO-01/A - Informatica
Settore MATH-03/B - Probabilità e statistica matematica
Settore STAT-01/A - Statistica
dic-2024
European Laboratory for Learning and Intelligent Systems
European Network of AI Excellence Centers
International Conference on Machine Learning
https://raw.githubusercontent.com/mlresearch/v251/main/assets/lavado24a/lavado24a.pdf
Book Part (author)
File in questo prodotto:
File Dimensione Formato  
lavado24a.pdf

accesso aperto

Descrizione: Main paper
Tipologia: Publisher's version/PDF
Dimensione 1.44 MB
Formato Adobe PDF
1.44 MB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/2434/1127775
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact