New Transformer Model Reconstructs 3D Curves from Sparse Images

Researchers have developed CGGT, a Curve-Grounded Geometry Transformer, designed to reconstruct editable 3D parametric curves from sparse, unposed multi-view images. This new method bypasses the need for dense camera views or costly per-scene optimization common in existing NeRF and 3DGS approaches. CGGT integrates a transformer encoder for feature learning and a masked-attention decoder for instance association, enabling it to predict camera parameters, depth maps, and curve masks in a single pass. The system is trained on Wireframe-100K, a large dataset of 100,000 CAD models, and demonstrates strong generalization to real-world images, effectively distinguishing structural edges from view-dependent ones. AI

IMPACT This research advances 3D reconstruction from limited visual data, potentially improving CAD modeling and computer graphics applications.

RANK_REASON The cluster describes a new research paper detailing a novel method for 3D curve reconstruction. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• 3D Gaussian splatting
• 3D Parametric Curve Reconstruction
• arXiv
• computer-aided design
• Geometry Transformer
• Hugging Face
• NeRF
• Wireframe-100K

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →