Custom ImageType for factory

[neda-dtu]

Hi,

I first want to thank you for the great collection of tools. I have started a discussion for this as I am not sure if the fix should be here or in rio_tiler. We are currently investigating using titiler for the next generation of our Global Wind Atlas portal, and it is going quite well. As part of this tool, we are loading numerical tiff files in the browser to handle styling on the client side. One limitation to this approach is that the Tiff files are not compressed, and also don't have the nodata information included in them.

I have implemented a hack to add the compression to rio_tiler/profiles.py by copying an updated file to my docker container after installing it, and thereby adding compression to all generated Tiff Tiles. However, it would be nice to have a way to override these output profiles directly in TiTiler, e.g. by creating a new ImageType class that is added to the Factory, and could then allow for overwriting the profile function to provide whatever we would like. An even better option for us would be able to directly configure the profile argument, as we would like to be able to set nodata values, and perhaps the compression predictor based on the original data.

Thanks,
Neil

[vincentsarago]

👋 Hi @neda-dtu,

This is an interesting use case.

First the compression. By default all responses should be compressed using either brotli or gzip (if accepted by the client).

I don't think there are one good solution. I've got an idea, I'll start a PR and we can continue the discussion there!

[vincentsarago]

@neda-dtu

without changing much of code you could get it to work by using titiler master version.

The idea is to subclass the RenderParams dependency and add what options you want in obj.kwargs

see: https://github.com/developmentseed/titiler/blob/master/tests/test_CustomRender.py

[neda-dtu]

Great thanks, I will work on setting this up later this weekend.

[vincentsarago]

@neda-dtu quick question, have you checked NumpyTile ? https://developmentseed.org/titiler/concepts/output_format/#numpytile

It seems that it could also be an interesting format for your use case

[neda-dtu]

We haven't yet. There is another project we are starting work on this year, where I plan on investigating NumpyTile.