HoW-3D: Holistic 3D Wireframe Perception from a Single Image

arXiv 2022 44.9 method, benchmark

TLDR

Proposes holistic 3D wireframe perception from single images, introducing ABC-HoW benchmark and Deep Spatial Gestalt model.

Reasoning

Strengths include a novel task, large-scale benchmark, and strong performance on invisible geometry inference. Weaknesses are reliance on synthetic data and focus on wireframes rather than full CAD reconstruction.

Read-first score

Read-first score 44.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 41.

Methodology quality 25%

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

Recency 8%

56.5

Uses a gentle age decay so recent papers surface without erasing older foundations. 2022

Reproducibility 25%

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

Topical relevance 42%

25.6

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

Methodology anchor

Rank sensitivity

Stability: volatile; rank range: 38.

Keyword Scores

B-Rep

boundary representation

image-to-CAD

AI for CAD

CAD reconstruction

computer-aided design

point cloud to CAD

neural CAD

generative CAD

parametric CAD

CAD generation

CAD program

constructive solid geometry

CSG

sketch extrusion

text-to-CAD

Deep Analysis

Innovations

Introduction of the holistic 3D wireframe perception (HoW-3D) task, requiring inference of both visible and invisible (NLOS) 3D wireframes from a single image.
ABC-HoW benchmark: a large-scale dataset of 12k single-view images with holistic 3D wireframe annotations derived from ABC CAD models.
Deep Spatial Gestalt (DSG) model that detects visible junctions and line segments and then infers NLOS 3D structures using Gestalt principles.

Methodology

The paper proposes the HoW-3D task and creates the ABC-HoW benchmark from ABC dataset CAD models. A Deep Spatial Gestalt model is designed to first learn visible 3D wireframe elements (junctions, line segments) from a single image, then infer the non-line-of-sight 3D geometry by applying Gestalt-inspired reasoning on the visible cues.

Key Results

The DSG model outperforms previous wireframe detectors in detecting invisible line geometry from single-view images and is competitive with methods that use high-fidelity point cloud inputs for 3D wireframe reconstruction.

Resources

Paper