Interest in adding metadata to the raster?

[jordancaraballo]

Hey Nick,

As part of my fork of Omnicloud for our operational surface reflectance product, I have been including some tags to the raster that include: class percentage (one per class), total cloud percentage, and cloud shadow percentage. Wondering if this is something you would like me to create a PR for if you think it would be useful for the overall package.

For us its useful because just by reading the metadata of the omnicloud outputs, we can exclude high cloud percentage images when needed, or simply to understand scenes before making mosaics of WorldView imagery. Let me know if this is useful, and I will package it into a PR if needed.

Thanks,
Jordan

[wrignj08]

Hey Jordan,

Yes, I can see how that would be useful, a PR would be a great starting point thanks!

Before you start, a couple of thoughts and questions to align on:

• Do you know if there is a standard structure or naming convention that we should be using for this meta data?
• Should we only perform this for classified output or also for the confidence outputs? Including it for confidence outputs would require performing an argmax first which would have a small performance hit.
• It may make sense to compute class statistics directly on the mosaic device immediately after the optimised argmax, using torch.bincount (and only if we are exporting to disk), as this is significantly faster than numpy.bincount. Happy to discuss placement, but that feels like a reasonable approach.

Thanks,

Nick

[wrignj08]

Hey @jordancaraballo

I have just published ClassCounter which should make this pretty easy to implement, did you still want to have a go at adding the functionality to OCM?