ArchCAD-400k: A Large-Scale CAD drawings Dataset and New Baseline for Panoptic Symbol Spotting

🌟 News

• 2025/10/16: Source code of DPSS is released !
• 2025/10/16: First round of open source of ArchCAD Dataset !
• 2025/10/9: Our project homepage is released !
• 2025/9/19: Our paper is accepted in NeurIPS 2025 ! 🎉

📖 Abstract

This repository contains the official implementation of Dual-Pathway Symbol Spotter (DPSS), a new baseline model for panoptic symbol spotting in architectural CAD drawings, which was introduced in our paper "ArchCAD-400K: An Open Large-Scale Architectural CAD Dataset and New Baseline for Panoptic Symbol Spotting". For more information about the project, please refer to our Project Page.

🔥 Highlights

• ArchCAD-400K: The first large-scale architectural CAD dataset with 400K+ symbols
• DPSS Model: Novel dual-pathway architecture for panoptic symbol spotting
• Comprehensive Evaluation: Extensive experiments and benchmarks
• Open Source: Both dataset and model code are publicly available

🔧Installation & Dataset

Environment

We provide an installation script to set up all dependencies:

bash install.sh

Download the required pretrained model weights:

cd svgnet/pretrained
bash download.sh

Dataset&Preprocess

ArchCAD-400K

We are pleased to announce the initial public release of the ArchCAD dataset. This first batch opens a curated subset of 40K high-quality samples, representing a more refined portion of the full collection. This release aims to facilitate preliminary research, with plans for subsequent releases in the future. Please visit our HuggingFace page and download the dataset.

FloorplanCAD

download dataset from floorplan website, and convert it to json format data for training and testing.

# download dataset
python dataset/download_data.py
# preprocess
## train, val, test
python dataset/parse_FpCAD_svg.py  --data_dir ./dataset/FloorplanCAD/train/svg_gt --output_dir ./dataset/FloorplanCAD/train/json
python dataset/parse_FpCAD_svg.py  --data_dir ./dataset/FloorplanCAD/val/svg_gt --output_dir ./dataset/FloorplanCAD/val/json
python dataset/parse_FpCAD_svg.py  --data_dir ./dataset/FloorplanCAD/test/svg_gt --output_dir ./dataset/FloorplanCAD/test/json

## dataset_split.json
python FloorplanCAD/dataset_split.py 

🚀Quick Start

# train
bash tools/train_dist.sh
# test
bash tools/test_dist.sh

TODO list:

• Release our tools CADParser for CAD processing.
• Release a highly optimized version of DPSS Framework.

Acknowledgement

We sincerely thank the authors of CADTransformer, SymPoint, SymPointV2 for their inspiring open-source contributions. We also thank all engineers and researchers involved in the data annotation, compilation, and review process.

📌 Citation

If you find this work useful in your research, please consider citing:

@article{luo2025archcad,
  title={ArchCAD-400K: An Open Large-Scale Architectural CAD Dataset and New Baseline for Panoptic Symbol Spotting},
  author={Luo, R and Liu, Z and Cheng, T and others},
  journal={arXiv preprint arXiv:2503.22346},
  year={2025}
}