ABC: A Big CAD Model Dataset for Geometric Deep Learning
TLDR
Introduces ABC-Dataset, one million CAD models with parametric surfaces for geometric deep learning research and benchmarking.
Reasoning
The paper's strength is providing a large, high-quality dataset with ground truth for various geometric tasks, enabling fair comparisons. Weakness is that it focuses solely on dataset contribution without novel methods or applications beyond a normal estimation benchmark.
Read-first score
Read-first score 49.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 57.
Field roles
Rank sensitivity
Stability: volatile; rank range: 80.
Keyword Scores
Deep Analysis
Innovations
- Introduction of ABC-Dataset: one million CAD models with explicit parametric curves and surfaces for geometric deep learning.
- Parametric descriptions provide ground truth for differential quantities, patch segmentation, feature detection, and shape reconstruction, and allow data generation in multiple formats and resolutions.
- Large-scale benchmark for surface normal estimation comparing data-driven methods against ground truth and traditional methods.
Methodology
The dataset consists of one million CAD models with explicitly parametrized curves and surfaces. These parametric models can be sampled to produce data in various formats (e.g., point clouds) and resolutions. A benchmark for surface normal estimation is conducted using this dataset, evaluating existing data-driven methods against both the ground truth normals and traditional estimation techniques.
Key Results
The benchmark provides a large-scale evaluation of surface normal estimation methods, comparing data-driven approaches to traditional methods and ground truth, but specific quantitative results are not detailed in the abstract.