Back to Wiki

[50] CADTransformer: Panoptic Symbol Spotting Transformer for CAD Drawings

EBSCO – Screened BIM Classification & Semantic Enrichment 2022

Paper Description extracted from PDF · abstract · 131 words

Understanding 2D computer-aided design (CAD) drawings plays a crucial role for creating 3D prototypes in architecture, engineering and construction (AEC) industries. The task of automated panoptic symbol spotting, i.e., to spot and parse both countable object instances (windows, doors, tables, etc.) and uncountable stuff (wall, railing, etc.) from CAD drawings, has recently drawn interests from the computer vision community. Unfortunately, the highly irregular ordering and orientations set major roadblocks for this task. Existing methods, based on convolutional neural networks (CNNs) and/or graph neural networks (GNNs), regress instance bounding boxes in the pixel domain and then convert the predictions into symbols. In this paper, we present a novel framework named CAD Transformer, that can painlessly modify existing vision transformer (ViT) backbones to tackle the above limitations for the panoptic symbol spotting task.

AuthorsFan, Zhiwen and Chen, Tianlong and Wang, Peihao and Wang, Zhangyang
Year2022
Venue
DOI10.1109/cvpr52688.2022.01071
Source DatabaseEBSCO (IEEE Xplore Digital Library)
Bridge-to-GNNCategory Borderline
GNN ArchitectureGNN (heuristic — verify)
Graph Encoding2D Drawing Graph (inferred from title)
AEC TaskNot yet extracted from PDF
CohortPeak Adoption (2022)

Extracted Methodology heuristic — verify before citing

Key Findingnovel framework named CAD Transformer, that can painlessly modify existing vision transformer (ViT) backbones to tackle the above limitations for the panoptic symbol spotting task (heuristic — verify)
Quality Assessment1/6 · rigor: Low ·
Code/Data: ✓ (heuristic — verify)

Detected Technical Stack auto-detected from PDF · regex catalogue match

GNN architectures detectedGNN (general)
Code repositories found in texthttps://github.com/VITA-Group/
Quality (heuristic, 6-flag)1/6 · rigor: Low
Code/Data: ✓Benchmark: —Baseline: —CV/Split: —Ablation: —Reproducible: —

Lineage Families

bim-classification

Enabled By / Precursors

[15]

Extended By

None in corpus

Infrastructure Dependencies

IFC BIM

Research Frontier Tags

graph-ml-aec

Part of the GML/GNN in AEC Systematic Review (PRISMA 2020) — 112 papers, 2020–2026