Living Atlas for Deep Learning

[MariamJamilah]

I am trying to extract roads using the MultiTaskRoadExtractor() model. In order to train my model, can I use the wayback world imagery from Living Atlas as the input raster? Also, what is the resolution of the wayback imagery?

[BernSzukalski]

To find the resolution and accuracy of World Imagery in the Wayback archives, just display the date you are interested in, zoom to the desired extent, and click the imagery to view a pop-up with details.

I am checking with Deep Learning team on whether Wayback is appropriate, but can't see why it wouldn't be depending on your purposes.

[MariamJamilah]

Thank you for the response!

However with wayback imagery, at different zoom levels, the resolution is different. Hence how do I use a specific zoom level (that is,  the imagery at a specific resolution) to make training chips for my model (to train my deep learning model)? When using the 'Export Training Data for Deep Learning' Tool, how to use a specific resolution of the wayback imagery to export training chips?

[BernSzukalski · Accepted solution]

A good question, but perhaps mute at this point. After reviewing things with the Living Atlas Imagery team, there are caveats that will prevent you from using Wayback and World Imagery in general with respect to deep learning. We license much of that imagery from Maxar, and that brings along extra caveats for use. While deep learning is not specifically covered in the Terms Of Use item section (which you will find in the item overview) after conferring about this everyone is in agreement that, unfortunately, this use of Maxar imagery is not permitted. 

When you make a map from the Wayback archives, like World Imagery it is multi-resolutional and sourced depending on the resolution.

[MariamJamilah]

Thank you!