Building Roofprints 50cm/px¶

Included on a trial basis with ORCA™ HUNTR is a version of Building Roofprints 50cm/px.

You can find the model already in the list of available trained models in the ORCA HUNTR - MAP view or in the list of available models when starting a run from the ORCA HUNTR - SCENES view.

choose model — Find the Building Roofprints 50cm/px here.

The general input requirements for ORCA HUNTR must be met. Additionally, this model has the following input requirements:

The image must be georeferenced
The image must have at least 3 channels (RGB)
The imagery should have a similar resolution to the training imagery (~50cm). This means detection quality will degrade significantly when out of the range between 30 and 70cm resolution.

The output is georeferenced. The vector output is available as polygons only.

Model Metrics¶

For this models and its metrics, a roofprint and its features, as presented in the satellite or aerial images, is defined as an intended target object meeting the following specifications: any permanent or semi-permanent man-made structure that has a footprint area equal to or greater than 6m² and is covered by a roof.

Examples of intended roofprints include single-family homes, commercial buildings, industrial buildings, warehouses, stadiums, row houses, and public buildings.

Examples of intended excluded features include natural features, small sheds, water features (pools), roads, bridges, statues, and playing fields.

Product quality may vary for different parts of the globe as it can be affected by various factors like building shape, density distribution, image quality, and occlusions. Complex roof structures (e.g., gabled mixed with flat) when present in industrial zones may lead to lower quality detection.

To provide a comprehensive way to show the differing product quality, five different zones are used. The zoning types used are central business district (CBD), commercial, industrial, residential, and rural.

The following metrics are the observed metrics for the vectorized output of the Building Roofprints 50cm/px model as tested on the Maxar Vivid 50cm dataset.

Accuracy of Roofprints per Region and Zone (Vectorized)
REGION	CBD	Commercial	Industrial	Residential	Rural
Australia and New Zealand	84	94	92	92	99
Caribbean	93	94	94	92	99
Central America	90	92	94	88	99
Central Asia	82	91	95	93	99
Eastern Africa	80	90	90	93	97
Eastern Asia	89	91	89	92	93
Eastern Europe	88	95	89	91	97
Melanesia	97	95	99	93	97
Micronesia	91	95	-	92	99
Middle Africa	87	93	95	90	97
Northern Africa	90	92	95	88	99
Northern America	90	93	94	91	99
Northern Europe	88	92	94	91	98
Polynesia	-	96	97	91	97
South America	87	94	95	85	98
South-Eastern Asia	86	94	94	92	99
Southern Africa	88	94	88	88	98
Southern Asia	85	90	90	93	96
Southern Europe	83	94	90	85	98
Western Africa	90	95	95	90	97
Western Asia	90	92	93	90	98
Western Europe	88	96	98	91	98

F1 Score of Roofprints per Region and Zone (Vectorized)
REGION	CBD	Commercial	Industrial	Residential	Rural
Australia and New Zealand	.67	.89	.75	.83	.85
Caribbean	.88	.93	.79	.81	.78
Central America	.85	.88	.80	.83	.74
Central Asia	.72	.78	.87	.84	.73
Eastern Africa	.75	.88	.84	.76	.51
Eastern Asia	.79	.87	.89	.86	.71
Eastern Europe	.81	.91	.83	.81	.79
Melanesia	.82	.88	.80	.76	.59
Micronesia	.85	.89	-	.85	.84
Middle Africa	.74	.83	.82	.79	.61
Northern Africa	.82	.90	.84	.82	.77
Northern America	.84	.90	.71	.83	.79
Northern Europe	.74	.89	.83	.85	.86
Polynesia	-	.89	.93	.81	.82
South America	.79	.91	.82	.79	.75
South-Eastern Asia	.82	.91	.92	.85	.85
Southern Africa	.84	.89	.82	.78	.43
Southern Asia	.81	.88	.84	.85	.77
Southern Europe	.75	.89	.72	.77	.79
Western Africa	.80	.89	.82	.81	.62
Western Asia	.68	.77	.78	.81	.71
Western Europe	.75	.92	.95	.81	.81