Low-quality pictures from the past are going to be lost if not digitalized and restored. This work focuses on the restoration of exploded view diagrams, which represent a particular subset of those images. The peculiarity of these images is their simplicity, which is both an advantage and a disadvantage since the little information they provide is easier to handle but harder to understand. The final goal of the study is to both reconstruct their individual components and rebuild the composite object. To do this, we need to generate a dataset of diagrams and their relative information that will be used for the application of reverse-engineering models which can restore the images in 2D or 3D. Most of the studies on the depth estimation of a scene have been developed on real pictures or realistic simulations of scenes which include non-uniform lighting. The understanding of the depth in an image or a video, in those cases, heavily relies on the different shades that the light creates when hitting objects and shadows. However, we totally miss this information in the exploded views, and this opens a new type of challenge. From segmented 3D models, we reconstruct their components’ connections and explode them, proposing an algorithm for the generation of the diagrams. We study the possibilities and choose them with an eye on scalability and generalization, trying to adapt existing studies to ours. In conclusion, we recognize how a set of ad-hoc techniques may outperform a general universal algorithm.
Senza la digitalizzazione e il restauro, vecchie immagini andranno perse. Questo lavoro é focalizzato sul restauro di diagrammi esplosi, che rappresentano un particolare sottoinsieme di queste immagini. La peculiaritá di questi diagrammi sta nella loro semplicitá, che é allo stesso tempo un vantaggio e uno svantaggio, poiché la poca informazione che forniscono é facile da gestire ma piú complessa da comprendere. L'obiettivo finale di questo studio é sia di ricostruire le singole componenti sia di ricomporre l'oggetto. Per far questo, dobbiamo generare un dataset di diagrammi e loro relative informazioni che verrá usato per l'applicazione di modelli di reverse engineering in grado di ricostruire le immagini in 2D o 3D. La maggior parte degli studi sulla stima della profonditá di una scena sono stati svolti su forografie o simulazioni realistiche con illuminazione non uniforme. La comprensione di un'immagine o un video, in questi casi, si basa fortemente sulle diverse sfumature e obre create dalla luce. Tuttavia, quest'informazione é del tutto assente nei diagrammi esplosi, e questo porta ad un nuovo tipo di sfida. Da modelli 3D segmentati ricostruiamo le connessioni tra i loro componenti e li esplodiamo, proponendo un algoritmo per la generazione dei diagrammi. Studiamo le varie possibilitá e le scegliamo prestando cura alla scabilitá e alla generalizzazione, provando ad adattare gli studi esistenti al nostro lavoro. In conclusione, riconosciamo come delle tecniche su misura possano ottenere risultati migliori rispetto a un algoritmo generico e universale.
Exploration of exploded view diagrams for schematic 3D analysis : a study on PartNet dataset
FAVARO, MARZIA
2021/2022
Abstract
Low-quality pictures from the past are going to be lost if not digitalized and restored. This work focuses on the restoration of exploded view diagrams, which represent a particular subset of those images. The peculiarity of these images is their simplicity, which is both an advantage and a disadvantage since the little information they provide is easier to handle but harder to understand. The final goal of the study is to both reconstruct their individual components and rebuild the composite object. To do this, we need to generate a dataset of diagrams and their relative information that will be used for the application of reverse-engineering models which can restore the images in 2D or 3D. Most of the studies on the depth estimation of a scene have been developed on real pictures or realistic simulations of scenes which include non-uniform lighting. The understanding of the depth in an image or a video, in those cases, heavily relies on the different shades that the light creates when hitting objects and shadows. However, we totally miss this information in the exploded views, and this opens a new type of challenge. From segmented 3D models, we reconstruct their components’ connections and explode them, proposing an algorithm for the generation of the diagrams. We study the possibilities and choose them with an eye on scalability and generalization, trying to adapt existing studies to ours. In conclusion, we recognize how a set of ad-hoc techniques may outperform a general universal algorithm.File | Dimensione | Formato | |
---|---|---|---|
Exploration_of_exploded_view_diagrams_for_schematic_3D_analysis__a_study_on_PartNet_dataset.pdf
accessibile in internet per tutti
Descrizione: tesi
Dimensione
6.89 MB
Formato
Adobe PDF
|
6.89 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/188825