Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SymPoint Revolutionized: Boosting Panoptic Symbol Spotting with Layer Feature Enhancement

Wenlong Liu, Tianyu Yang, Qizhi Yu, Lei Zhang

TL;DR

SymPoint-V2 tackles panoptic symbol spotting in CAD drawings by addressing two key drawbacks of the prior approach: neglect of graphical layer information and slow early-stage convergence. It introduces a Layer Feature-Enhanced (LFE) module to fuse layer context into primitive features, and a Position-Guided Training (PGT) strategy that uses center queries to stabilize and accelerate learning in the transformer decoder. On FloorPlanCAD, SPv2 achieves state-of-the-art performance, notably a panoptic quality (PQ) of 90.1 and substantial gains in semantic and instance spotting over SPv1, while maintaining reasonable training and inference efficiency. The work advances CAD parsing for BIM and engineering workflows by delivering more accurate, layer-aware symbol spotting with improved training dynamics and practical deployment potential.

Abstract

SymPoint is an initial attempt that utilizes point set representation to solve the panoptic symbol spotting task on CAD drawing. Despite its considerable success, it overlooks graphical layer information and suffers from prohibitively slow training convergence. To tackle this issue, we introduce SymPoint-V2, a robust and efficient solution featuring novel, streamlined designs that overcome these limitations. In particular, we first propose a Layer Feature-Enhanced module (LFE) to encode the graphical layer information into the primitive feature, which significantly boosts the performance. We also design a Position-Guided Training (PGT) method to make it easier to learn, which accelerates the convergence of the model in the early stages and further promotes performance. Extensive experiments show that our model achieves better performance and faster convergence than its predecessor SymPoint on the public benchmark. Our code and trained models are available at https://github.com/nicehuster/SymPointV2.

SymPoint Revolutionized: Boosting Panoptic Symbol Spotting with Layer Feature Enhancement

TL;DR

SymPoint-V2 tackles panoptic symbol spotting in CAD drawings by addressing two key drawbacks of the prior approach: neglect of graphical layer information and slow early-stage convergence. It introduces a Layer Feature-Enhanced (LFE) module to fuse layer context into primitive features, and a Position-Guided Training (PGT) strategy that uses center queries to stabilize and accelerate learning in the transformer decoder. On FloorPlanCAD, SPv2 achieves state-of-the-art performance, notably a panoptic quality (PQ) of 90.1 and substantial gains in semantic and instance spotting over SPv1, while maintaining reasonable training and inference efficiency. The work advances CAD parsing for BIM and engineering workflows by delivering more accurate, layer-aware symbol spotting with improved training dynamics and practical deployment potential.

Abstract

SymPoint is an initial attempt that utilizes point set representation to solve the panoptic symbol spotting task on CAD drawing. Despite its considerable success, it overlooks graphical layer information and suffers from prohibitively slow training convergence. To tackle this issue, we introduce SymPoint-V2, a robust and efficient solution featuring novel, streamlined designs that overcome these limitations. In particular, we first propose a Layer Feature-Enhanced module (LFE) to encode the graphical layer information into the primitive feature, which significantly boosts the performance. We also design a Position-Guided Training (PGT) method to make it easier to learn, which accelerates the convergence of the model in the early stages and further promotes performance. Extensive experiments show that our model achieves better performance and faster convergence than its predecessor SymPoint on the public benchmark. Our code and trained models are available at https://github.com/nicehuster/SymPointV2.
Paper Structure (35 sections, 8 equations, 11 figures, 5 tables, 1 algorithm)

This paper contains 35 sections, 8 equations, 11 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Panoptic Symbol Spotting
  4. Ease Training for DETRs
  5. Approach
  6. Preliminaries
  7. Task Formulation.
  8. SPv1.
  9. Baseline.
  10. Layer Feature Enchanced Module
  11. Pool Function.
  12. Fusion Function.
  13. Position-Guided Training
  14. Class Embedding.
  15. Positional Encoding.
  16. ...and 20 more sections

Figures (11)

  • Figure 1: A 2D floorplan (a) and its panoptic symbol spotting results (b), in which the semantics of segments are indicated through different color and instances are highlighted by semi-transparent rectangles. The BIM model (c) with complete semantic and precise geometry can be reconstructed from such an annotated floor plan. We only present the 3D model of windows, doors, and walls for clarity.
  • Figure 1: Results of SPv2 on FloorPlanCAD.
  • Figure 2: (a) A CAD drawing is composed of multiple graphical layers. (b) Comparision curves of with and without center queries.
  • Figure 2: Results of SPv2 on FloorPlanCAD.
  • Figure 3: The overview of our framework.
  • ...and 6 more figures