Use the

Building Footprint Extraction Model (USA)

in ArcGIS Pro

Version 1.0
Table of contents
Introduction .................................................................................................................................................. 2
Licensing requirements ................................................................................................................................ 2
Building Footprint USA Model notes ............................................................................................................ 2
Access the model .......................................................................................................................................... 3
ArcGIS Living Atlas of the World ............................................................................................................... 3
Imagery properties ....................................................................................................................................... 3
Example 1—Processing Worldview-2 imagery ......................................................................................... 4
Use the model to extract building footprint features in ArcGIS Pro ............................................................ 7
Postprocessing ............................................................................................................................................ 13
References .................................................................................................................................................. 16

1
Introduction
This document explains how to use the building footprint extraction (USA) deep learning model available
within ArcGIS Living Atlas of the World.

If done manually, building footprint extraction is a complex and time-consuming task. You can use deep
learning to significantly optimize and automate this task. This building footprint extraction deep learning
package is a ready-to-use deep learning model that has been trained to extract building footprints from
high-resolution aerial/satellite imagery. You can use this model as is or fine-tune it to adapt to your own
data and geography. If you are interested in other ready-to-use models, see ArcGIS Living Atlas of the
World.

For those experienced with deep learning and those with data resources to train their models, ArcGIS
has all the tools built in to allow users to create their own deep learning models. Some of the workflows
to train deep learning models can be found on the ArcGIS Developers site.

Licensing requirements
ArcGIS Desktop—ArcGIS Image Analyst and ArcGIS 3D Analyst for ArcGIS Pro. The model cannot be used
in ArcMap. The ArcGIS 3D Analyst Extension is used for postprocessing by the Regularize Building
Footprint tool to improve the results.
ArcGIS Enterprise—ArcGIS Enterprise Map Viewer and ArcGIS Image Server with raster analytics
configured.
ArcGIS Online—Coming soon to ArcGIS Online Map Viewer using ArcGIS Online imagery.

Building Footprint USA Model notes

• Input—Raster datasets, raster products, mosaic datasets, or image services. You can find more
details in the Imagery properties section of this document.
• Output—A feature class containing building footprints.
• Processing and compute—This workflow is compute intensive, and a GPU with optimum CUDA
compute capability is recommended.
• Applicable geographies—This model is expected to work well in the United States.
• Architecture—This model uses the MaskRCNN model architecture implemented in ArcGIS API
for Python.
• Accuracy metrics—This model has an average precision score of 79.1 percent.

2
Access the model
The model can be downloaded from ArcGIS Living Atlas and then used, accessed directly from ArcGIS
Pro 2.7 and later, or used directly in ArcGIS Online Imagery (which currently is in beta).

ArcGIS Living Atlas of the World

To access the model in ArcGIS Living Atlas, do the following:
1. Browse to ArcGIS Living Atlas in a web browser.
2. Sign in with your ArcGIS Online ID.
3. Search for Building Footprint Extraction – USA and open the item page from the search results.
Alternatively, you can use this direct link:
http://www.arcgis.com/home/item.html?id=a6857359a1cd44839781a4f113cd5934
4. To download the model, click the Download button.
This downloads a DLPK file, which can be uploaded and used in ArcGIS Enterprise or in ArcGIS Pro.

Imagery properties
Use imagery with the following properties for optimum results:
• Resolution—High-resolution aerial/satellite imagery 10–40 cm.
• Dynamic range—8 bit
• Bands—Three bands, for example, Red(1), Green (2), and Blue (3).
• Orthorectified imagery (on-the-fly or persisted ortho products).
Off-nadir imagery or imagery with a high obliquity angle will not produce suitable results.

3
Example 1—Processing Worldview-2 imagery

1. In ArcGIS Pro, browse to the image product in the Contents pane.

2. Expand the product (.IMD file) and add Pansharpen layer to the map.

3. Right-click the newly added layer and select Edit Function Chain.
This opens the function chain window,

4. Click the Stretch Function button to edit stretch properties:

4
5. In the Stretch Properties dialog box, set Output Pixel Type to 8 Bit Unsigned:

5
6. Click the Apply button to apply the changes:

6
Use the model to extract building footprint features in ArcGIS Pro
To extract building footprints from the imagery, follow these steps:
1. Make sure you have downloaded the model and added the imagery layer in ArcGIS Pro.

2. Zoom to an area of interest.

3. Browse to Analysis > Tools.

The Geoprocessing pane appears.

7
4. Click Toolboxes > Image Analyst Tools > Deep Learning > Detect Objects Using Deep Learning.

5. Point to imagery and model to corresponding values and change the parameters as below.

8
 With ArcGIS Pro 2.7 and later, you can search the model from ArcGIS Living Atlas.

Click the Model Definition browse button. Search for the model using item ID in the browse window as
shown :

9
10
6. Set the Environments variables as shown:

a Set Cell Size to 0.3 (required). 0.3 is the raster resolution.

a. Set Processing Extent to the current display extent.

b. Set Processor Type to GPU if available.

11
7. Click Run to execute.

Once processing is complete, the layer is added to the map.

12
Postprocessing
To improve the visual appearance of the building footprint features, follow these steps:
1. Browse to Analysis > Tools.

The Geoprocessing pane appears.

2. Browse to Toolboxes > 3D Analyst Tools > Regularize Building Footprints.

13
 The input is the building footprint features that were detected by the inferencing tool.
3. Modify the parameters to match the screenshot.

14
4. Click Run.

15
References

Detect Objects Using Deep Learning

ArcGIS API for Python Samples Regularize Building Footprint

16