SweepNet: Unsupervised Learning Shape Abstraction via Neural Sweepers
This work addresses shape abstraction for computer graphics and CAD applications, presenting an incremental improvement through a novel neural approach.
The paper tackles shape abstraction by introducing SweepNet, an unsupervised method that predicts sweep surface representations using a differentiable neural sweeper and encoder-decoder architecture, achieving effective simplification with as few as 14 parameters.
Shape abstraction is an important task for simplifying complex geometric structures while retaining essential features. Sweep surfaces, commonly found in human-made objects, aid in this process by effectively capturing and representing object geometry, thereby facilitating abstraction. In this paper, we introduce \papername, a novel approach to shape abstraction through sweep surfaces. We propose an effective parameterization for sweep surfaces, utilizing superellipses for profile representation and B-spline curves for the axis. This compact representation, requiring as few as 14 float numbers, facilitates intuitive and interactive editing while preserving shape details effectively. Additionally, by introducing a differentiable neural sweeper and an encoder-decoder architecture, we demonstrate the ability to predict sweep surface representations without supervision. We show the superiority of our model through several quantitative and qualitative experiments throughout the paper. Our code is available at https://mingrui-zhao.github.io/SweepNet/