Awesome AI4CAD Hub Papers · Datasets · Projects
← Back to papers

Text-to-CAD Retrieval: a Strong Baseline

arXiv 2026 54.7 method

TLDR

Introduces text-to-CAD retrieval as a new task and proposes a strong baseline using multi-modal embeddings from procedural sequences and point clouds.

Reasoning

Strengths: formalizes an underexplored task, establishes a benchmark, and proposes a novel multi-modal alignment framework. Weaknesses: limited to retrieval without generative results, and no quantitative results are reported in the abstract.

Read-first score

Read-first score 54.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 57.

Recency 8%
100

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

Methodology quality 25%
80

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

Reproducibility 25%
46

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%
35.6

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 32.

Keyword Scores

computer-aided design
10
text-to-CAD
10
neural CAD
8
AI for CAD
6
CAD program
6
parametric CAD
5
point cloud to CAD
5
generative CAD
4
CAD generation
3
B-Rep
0
boundary representation
0
constructive solid geometry
0
CSG
0
sketch extrusion
0
CAD reconstruction
0
image-to-CAD
0

Deep Analysis

Innovations

  • Formal introduction of text-to-CAD retrieval as a new cross-modal retrieval task
  • Unified framework learning multi-modal CAD embeddings from both procedural sequences and geometric point clouds
  • Novel feature decoder that reconstructs masked sequence features via cross-attention with text and point features for implicit multi-modal alignment
  • Efficient inference by removing the auxiliary decoder and using concatenated sequence-point features

Methodology

The framework encodes CAD models using a sequence encoder for construction logic and a point encoder for geometric point clouds, and text queries via a text encoder. During training, a feature decoder reconstructs masked sequence features through cross-attention with text and point features to align modalities. At inference, the decoder is discarded and retrieval is performed using concatenated sequence-point features.

Key Results

The proposed method establishes a strong baseline for text-to-CAD retrieval, enabling efficient cross-modal retrieval and supporting downstream tasks like retrieval-augmented generation.

Tags

CV