Support for remote images

[jmiguelv]

Currently, BVQA only supports local image files stored in the ./data/images/ directory. Which means images need to be copied into local storage before processing.

Would it make sense to support remote images, particularly those available via IIIF servers?

[geoffroy-noel-ddh]

You can specify a different data folder path using the -R argument. This would help with images from github or other packaged dataset.

I'm less sure about processing directly from non-local copies. So far our use cases involved iteratively trying questions or models over the same collection of images. It never worked well enough with the first attempt. So latency would be an issue for repeated runs.

I see the point in not having to copy a whole collection locally if it's really large. That's convenient. But as soon as there is more than one pass on the same set of images, multiple downloads becomes expensive. And if we introduce some local caching then how is that different from a separate script fetching a selection of a IIIF collection locally?

I'm thinking this requirement might better be externalised.

Do you have any particular real use cases in mind for IIIF or other APIs?

[jmiguelv]

I don't have a specific use case at the moment, but was thinking of applying it to a corpus that has a IIIF endpoint.

I can see the point about externalising it to a script though. That makes sense given the workflow you described.