STEP-Parts: Geometric Partitioning of Boundary Representations for Large-Scale CAD Processing

arXiv 2026 56.1 method

TLDR

STEP-Parts deterministically extracts geometric instance partitions from B-Reps for large-scale CAD processing, providing stable labels for downstream learning.

Reasoning

Strengths include a deterministic, efficient pipeline that preserves analytic surface structure and topological adjacency, with robust boundaries under tessellation changes. Weaknesses are its limitation to B-Rep inputs and lack of direct evaluation on generative or reconstruction tasks beyond two downstream probes.

Read-first score

Read-first score 56.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 53.

Recency 8%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=analysis,dataset,evaluation,metric,result

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code,dataset

Topical relevance 42%

33.1

Uses existing LLM keyword relevance scores normalized to 0-100. AI for CAD,computer-aided design,neural CAD,generative CAD,parametric CAD,B-Rep,boundary representation,constructive solid geometry,CSG,sketch extrusion,CAD generation,CAD reconstruction,text-to-CAD,image-to-CAD,point cloud to CAD,CAD program

Field roles

FrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 39.

Keyword Scores

B-Rep

boundary representation

computer-aided design

AI for CAD

neural CAD

CAD reconstruction

parametric CAD

generative CAD

CAD generation

constructive solid geometry

CSG

point cloud to CAD

sketch extrusion

text-to-CAD

image-to-CAD

CAD program

Deep Analysis

Innovations

Deterministic CAD-to-supervision toolchain extracting geometric instance partitions from raw STEP B-Reps, preserving analytic surface and topology.
Face merging based on same analytic primitive type and near-tangent continuity, exploiting bimodal dihedral angles for threshold-insensitive part extraction.
Transfer of partitions to tessellated meshes via retained source-face correspondence, ensuring tessellation-robust instance labels.
Efficient large-scale processing: 180k models in <6 hours on a consumer CPU.

Methodology

STEP-Parts processes raw STEP B-Rep files by merging adjacent faces that share the same analytic primitive type and satisfy a near-tangent continuity criterion, using a bimodal dihedral angle distribution to set a threshold-insensitive low-angle regime. The resulting geometric instance partitions are transferred to triangle meshes through retained source-face correspondence, producing instance labels and metadata.

Key Results

The pipeline processes approximately 180,000 models from the DeepCAD subset of ABC in under six hours on a consumer CPU, and the partitions serve as tessellation-robust supervision for an implicit reconstruction-segmentation network and a point-based backbone.

Resources

Paper