Sketch2Arti: Sketch-based Articulation Modeling of CAD Objects

2026-01-22 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Graphics & 3D Modeling · Depth: Expert, extended

Summary

Sketch2Arti is introduced as the first sketch-based system for articulation modeling of CAD objects, developed by researchers from the University of Edinburgh and Tsinghua University. This system allows users to define movable parts and their motion parameters on 3D objects using simple 2D sketches. It features category-agnostic training, enabling strong generalization to diverse objects beyond existing datasets like PartNet-Mobility. Sketch2Arti also supports controllable internal geometry completion for shell models, generating plausible structures consistent with existing geometry and predicted motion constraints. The system's effectiveness, controllability, and generalization are validated through comprehensive experiments and user evaluations, with code, dataset, and a prototype system available online. A new dataset, SketchMobility, comprising ~5K articulated shapes across 48 categories, was created for training and evaluation.

Key takeaway

For CAD designers and 3D artists seeking efficient and controllable articulation modeling, "Sketch2Arti" offers a powerful solution. You can rapidly define complex object movements and even generate missing internal structures using intuitive 2D sketches, significantly streamlining your workflow. This approach reduces reliance on limited datasets and enhances generalization, allowing you to animate novel or out-of-distribution objects with high fidelity and user control.

Key insights

Sketch2Arti enables intuitive, precise articulation modeling and internal geometry completion for CAD objects using simple 2D user sketches.

Principles

• Local geometric cues enable category-agnostic articulation inference.
• Hierarchical segmentation improves part identification accuracy.
• Geometric snapping refines predicted motion parameters for plausibility.

Method

Sketch2Arti uses a UNet-based CNN for 2D mask and motion parameter prediction from local depth/normal maps, then maps to 3D parts via hierarchical Partfield segmentation, followed by geometric snapping and iterative, masked 3D generative model completion.

In practice

• Specify diverse articulation intents for the same part by changing sketches.
• Iteratively model multiple articulations on complex objects.
• Generate missing internal structures for shell models.

Topics

• Sketch-based Modeling
• Articulation Modeling
• CAD Objects
• 3D Generative Models
• Shape Completion
• Part Segmentation
• SketchMobility Dataset

Best for: Research Scientist, AI Scientist, AI Engineer, Product Designer

Related on AIssential

• Geometrically Consistent Multi-View Scene Generation from Freehand Sketches — A new framework generates geometrically consistent multi-view scenes from single freehand sketches using novel data and architectural components.
• STEP-Parts: Geometric Partitioning of Boundary Representations for Large-Scale CAD Processing — STEP-Parts extracts stable geometric instance partitions from B-Reps, preserving analytic structure for CAD learning.
• From edges to meaning: Semantic line sketches as a cognitive scaffold for ancient pictograph invention — Pictographic writing likely emerged from the brain's intrinsic visual compression into boundary-based abstractions.
• I made an AI system that turns 2D floor plans into 3D, kinda surprised it works — Mitigating "AI hallucinations" in generative models often requires breaking down complex tasks into constrained, sequential steps.
• Diff-SBSR: Learning Multimodal Feature-Enhanced Diffusion Models for Zero-Shot Sketch-Based 3D Shape Retrieval — Diffusion models, enhanced with multimodal features, can effectively perform zero-shot sketch-based 3D shape retrieval.
• VoxelDiffusionCut: Non-destructive Internal-part Extraction via Iterative Cutting and Structure Estimation — Iterative internal structure estimation via diffusion models enables non-destructive part extraction from unknown products.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.