DreamCAD: Scaling Multi-modal CAD Generation using Differentiable Parametric Surfaces

arXiv 2026 66.7 method

TLDR

DreamCAD generates editable BReps from point-level supervision using differentiable parametric surfaces, achieving multi-modal CAD generation with state-of-the-art results.

Reasoning

The paper introduces a novel differentiable tessellation method for BRep generation from unannotated meshes, enabling scalable multi-modal CAD generation. Strengths include a large new captioning dataset and strong benchmark results; weaknesses are not explicitly discussed in the abstract but may include reliance on synthetic data or limited evaluation of editability.

Read-first score

Read-first score 66.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 113.

Recency 8%

100

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

Topical relevance 42%

70.6

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

Methodology quality 25%

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

Reproducibility 25%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 22.

Keyword Scores

computer-aided design

generative CAD

parametric CAD

B-Rep

boundary representation

CAD generation

text-to-CAD

neural CAD

image-to-CAD

point cloud to CAD

AI for CAD

CAD reconstruction

constructive solid geometry

CSG

sketch extrusion

CAD program

Deep Analysis

Innovations

Multi-modal generative framework that directly produces editable BReps from point-level supervision without CAD-specific annotations
Representation of BRep as parametric patches (Bézier surfaces) with differentiable tessellation for mesh generation, enabling training on unannotated 3D data
Introduction of CADCap-1M, a large CAD captioning dataset with over 1M descriptions generated using GPT-5
State-of-the-art performance on ABC and Objaverse benchmarks across text, image, and point modalities

Methodology

DreamCAD represents each BRep as a set of parametric patches (e.g., Bézier surfaces) and uses a differentiable tessellation method to generate meshes, allowing training with point-level supervision on large unannotated 3D datasets. It is a multi-modal framework that can generate editable CAD models from text, image, or point inputs. They also introduce CADCap-1M, a dataset of 1M+ CAD captions generated by GPT-5 for text-to-CAD research.

Key Results

DreamCAD achieves state-of-the-art performance on ABC and Objaverse benchmarks across text, image, and point modalities, improving geometric fidelity and surpassing 75% user preference.

Resources

Paper