SR-CurvANN: Advancing 3D Surface Reconstruction through Curvature-Aware Neural Networks

Marina Hernández-Bautista,Francisco J. Melero

2024-09-26

Abstract:Incomplete or missing data in three-dimensional (3D) models can lead to erroneous or flawed renderings, limiting their usefulness in applications such as visualization, geometric computation, and 3D printing. Conventional surface-repair techniques often fail to infer complex geometric details in missing areas. Neural networks successfully address hole-filling tasks in 2D images using inpainting techniques. The combination of surface reconstruction algorithms, guided by the model's curvature properties and the creativity of neural networks in the inpainting processes should provide realistic results in the hole completion task. In this paper, we propose a novel method entitled SR-CurvANN (Surface Reconstruction Based on Curvature-Aware Neural Networks) that incorporates neural network-based 2D inpainting to effectively reconstruct 3D surfaces. We train the neural networks with images that represent planar representations of the curvature at vertices of hundreds of 3D models. Once the missing areas have been inferred, a coarse-to-fine surface deformation process ensures that the surface fits the reconstructed curvature image. Our proposal makes it possible to learn and generalize patterns from a wide variety of training 3D models, generating comprehensive inpainted curvature images and surfaces. Experiments conducted on 959 models with several holes have demonstrated that SR-CurvANN excels in the shape completion process, filling holes with a remarkable level of realism and precision.

Graphics,Artificial Intelligence

What problem does this paper attempt to address?

The problems that this paper attempts to solve are: in three - dimensional (3D) models, rendering errors, geometric calculation errors and 3D printing problems caused by incomplete or missing data. Specifically, traditional surface repair techniques are often unable to infer complex geometric details in missing regions, especially when dealing with 3D models with complex shapes and details.

To address these problems, the author proposes a new method, called SR - CurvANN (Surface Reconstruction based on Curvature - Aware Neural Network). This method combines surface reconstruction algorithms and deep - learning techniques, especially by using the curvature characteristics of the model to guide the neural network for inpainting, thereby achieving more accurate and realistic hole - filling tasks.

The following are the main contributions of this paper:
1. **Proposing a fully - automatic method**: SR - CurvANN combines image inpainting with virtual sculpture and can reconstruct surfaces with a large number of polygons and complex shapes, surpassing the limitations of existing methods for simple models.
2. **Demonstrating that a specifically - trained neural network is superior to a general inpainting network**: On the curvature image data set, a specially - trained neural network can provide more accurate and precise surface completion effects.
3. **Defining a process for creating training images containing holes**: A large number of training images are generated from real 3D models, ensuring the diversity and authenticity of the training data.

Through these improvements, SR - CurvANN can fill holes with higher realism and precision when dealing with 3D models with complex geometric structures.