Seek-CAD: A Self-refined Generative Modeling for 3D Parametric CAD Using Local Inference via DeepSeek

arXiv 2025 52.5 method

TLDR

Seek-CAD uses locally deployed DeepSeek-R1 with visual and CoT feedback for training-free parametric CAD generation, introducing a new SSR dataset.

Reasoning

The paper presents a novel training-free approach using an open-source LLM for CAD generation, with a self-refinement mechanism incorporating visual and chain-of-thought feedback. Strengths include innovation and flexibility; weaknesses include potential reliance on specific LLM and incomplete abstract details.

Read-first score

Read-first score 52.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 70.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Methodology quality 25%

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

Topical relevance 42%

43.8

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

Reproducibility 25%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 7.

Keyword Scores

parametric CAD

CAD generation

AI for CAD

computer-aided design

generative CAD

sketch extrusion

neural CAD

CAD program

image-to-CAD

B-Rep

boundary representation

constructive solid geometry

CSG

CAD reconstruction

text-to-CAD

point cloud to CAD

Deep Analysis

Innovations

First exploration of locally deployed open-source LLM (DeepSeek-R1) for training-free CAD parametric model generation.
Self-refinement mechanism that integrates visual feedback from rendered step-wise images and Chain-of-Thought feedback.
Introduction of a new 3D CAD dataset based on the SSR (Sketch, Sketch-based feature, Refinements) triple design paradigm.

Methodology

Seek-CAD generates an initial CAD model, renders it into step-wise perspective images, then uses a Vision Language Model and Chain-of-Thought reasoning from DeepSeek-R1 to assess the model. The feedback is used by DeepSeek-R1 to iteratively refine the model in a training-free loop.

Key Results

Extensive experiments demonstrate the effectiveness of Seek-CAD across various metrics.

Resources

Paper