Skip to the content.


(2021, August 13th) Initial release including 11,602 CAD drawings.


Access to large and diverse CAD drawings is critical for developing vector graphics and symbol related algorithms. Here, we present FloorPlanCAD, a large-scale real-world CAD drawing dataset containing over 10,000 floor plans, ranging from residential to commercial buildings. With line-grained annotations, we introduce the panoptic symbol spotting task, which requires spotting both instances of countable things and semantics of the uncountable stuff. Other tasks will be available in the future.

The rich semantic, accurate location and detailed 3D shape (middle) of windows (light blue), blind windows (blue), railings (orange) and walls (dark red) are faithfully encoded in the CAD drawings of a floor plan (left) and its south facade (right).


CAD drawings in the dataset are in the SVG format. An illustration of our FloorPlanCAD dataset.

Detailed properties and statistics can be found here


Fan, Zhiwen and Zhu, Lingjie and Li, Honghua and Zhu, Siyu and Tan, Ping


We are grateful to our partners for providing the CAD drawings and support.

Paper and Citation

Please cite our paper if you use the FloorPlanCAD dataset.

  author = {Fan, Zhiwen and Zhu, Lingjie and Li, Honghua and Zhu, Siyu and Tan, Ping},
  title = {FloorPlanCAD: A Large-Scale CAD Drawing Dataset for Panoptic Symbol},
  booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
  month = {October},
  year = {2021}


The copyright of the CAD drawings is owned by their creators.


By downloading the dataset you agree to the following terms:

The authors give no warranties regarding the dataset


The dataset is versioned to accommodate for future updates of the data chunks and tasks.

File Formats

Each drawing in the dataset consists of the following file formats:


Based on the CAD drawings from the FloorPlanCAD dataset, we provide the following benchmarks for the task of panoptic symbol spotting.

Panoptic symbol spotting task

We provide a GCN-based base-line and list the statistical results on the proposed dataset. For more details, please refer to the source code page.