This deep learning model is used for extracting windows and doors in textured building data displayed in 3D views. Manually digitizing windows and doors from 3D building data can be a slow process. This model automates the extraction of these objects from a 3D view and can help in speeding up 3D editing and analysis workflows. Using this model, existing building data can be enhanced with additional information on location, size, and orientation of windows and doors. The extracted windows and doors can be further used to perform 3D visibility analysis using existing 3D geoprocessing tools in ArcGIS.
This model can be useful in many industries and workflows. National government and state-level law enforcement could use this model in security analysis scenarios. Local governments could use windows and door locations to help with tax assessments with CAMA (computer-aided mass appraisal) plus impact-studies for urban planning. Public safety personnel might be interested in regards to physical or visual access to restricted areas, or the ability to build evacuation plans. The commercial sector, with everyone from real estate agents to advertisers to office/interior designers, can benefit from knowing where windows and doors are located. Even utilities, especially mobile phone providers, could take advantage of knowing window sizes and positions. To be clear, this model doesn't solve these problems, but it does allow users to extract and collate some of the data they will need to do it.
This deep learning model was trained using images from the Open Images Dataset.
To complete this workflow, the following are the license requirements:
- An ArcGIS Pro Advanced license is required to use the Interactive Object Detection tool.
This model has the following characteristics:
- Input—Image (RGB).
- Output—Tuple with predictions, labels (window or door) and, optionally, confidence scores.
- Compute—This workflow is compute intensive and a GPU with compute capability of 6.0 or higher is recommended.
- Architecture—This model uses the Faster R-CNN model architecture implemented in ArcGIS API for Python.
- Accuracy metrics—This model has an average precision score of 0.36 for door class and 0.54 for window class.
Access and download the model
Download the Windows and Doors Extraction pre-trained model from ArcGIS Living Atlas of the World. Alternatively, access the model directly from ArcGIS Pro
The following are the release notes: