Fusion 360 gallery

Abstract

Parametric computer-aided design (CAD) is a standard paradigm used to design manufactured objects, where a 3D shape is represented as a program supported by the CAD software. Despite the pervasiveness of parametric CAD and a growing interest from the research community, currently there does not exist a dataset of realistic CAD models in a concise programmatic form. In this paper we present the Fusion 360 Gallery , consisting of a simple language with just the sketch and extrude modeling operations, and a dataset of 8,625 human design sequences expressed in this language. We also present an interactive environment called the Fusion 360 Gym , which exposes the sequential construction of a CAD program as a Markov decision process, making it amendable to machine learning approaches. As a use case for our dataset and environment, we define the CAD reconstruction task of recovering a CAD program from a target geometry. We report results of applying state-of-the-art methods of program synthesis with neurally guided search on this task.

Clustering of Sequential Cad Modelling Data

“…Indeed, the here presented data show that if only models with at least two CAD modelling operations are considered, the proportion of the models containing the extrusion feature increases to approximately 88%. They also claim the modelling operations the most used after the sketch and extrude are -fillet and chamfer (edge cut operations) (Willis et al, 2021), which our study has confirmed with the share of 26.8% of the total number of operations used in the dataset. Furthermore, second-level clustering analysis has answered the second research question.…”

“…In contrast to all these works and despite the research progress in using CAD data for learning-based purposes, only a limited number of research focuses on sequences of CAD modelling operations. Hence, we focus our research on understanding how the CAD model's design came about by using a fuller array of human-designed CAD modelling operations beyond a sketch and extrude (Willis et al, 2021;Wu et al, 2021), sweep (Li et al, 2020) or boolean operations (Xu et al, 2021). As a first step in automating user interaction with CAD systems, the characteristic transitions between feature-based operation needs to be understood.…”

“…Furthermore, CAD reconstruction (e.g., feature recognition) is the detection of geometric primitives and the correspondence of primitives holistically modelled as a modelling sequence. The learning-based approaches are also mainly applied to potentially learn the rules that can automate scenarios requiring user input and generalize when confronted with unfamiliar geometry (Willis et al, 2021), both being compared with the traditional methods that cannot completely rebuild the model operation sequences, such as rule-based or grammar-based. Thus, Fusion 360 Gallery (Willis et al, 2021), Zone Graphs (Xu et al, n.d.), DeepCAD (Wu et al, 2021), and CSGNet (Sharma et al, 2017) are the recent research learning-based attempts to reconstruct the 3D models from CAD modelling sequences generated from B-reps or Constructive Solid Geometry (CSG) representations.…”

Neurosymbolic Models for Computer Graphics

“…Such a program could include continuous parameters that can be varied to produce a range of possible visual outputs. Many CAD programs fall into this category [WPL*21, WXZ21]. Probabilistic Program : a single program that makes random choices, producing a different visual output each time it is executed (e.g.…”

“…Procedural modeling as applied to CAD: using feature‐based modeling to produce solid geometry as a composition of operations on 2D sketches extruded into boundary representations [WPL*21]. …”