[BUG] Cellfinder 2D viewer not working on windows

[FedeClaudi]

Describe the bug
The 2D viewer to inspect the results of cellfinder doesn't start on windows [gets stuck while loading the data].

[adamltyson]

Thanks to @FedeClaudi, this seems to be related to napari/napari#733. Should be an easy workaround.

[FedeClaudi]

As per the issue above:

Calling v.add_image(images, contrast_limits =[0.0, 1000.0]) works,
The upper bound of the contrast limits should be computed for each dataset though.

[adamltyson]

Thanks.
Temporary fix could be:

• Compute histogram on "some" images (e.g. every nth, or the middle n)
• Calculate a "liberal" limit (e.g. 0->max*2)
• v.add_image with these contrast limits

[adamltyson]

Alternatively, just pass the "theoretical" max from dtype and allow the user to change the limits. This is probably not ideal though.

[adamltyson]

N.B. If this issue persists, it will need fixing for the curation interface too.

[FedeClaudi]

Can confirm that the same problem comes up with curation interface.

[adamltyson]

Fixed in a627661.

Not sure if this should be made smarter though. Currently it just takes 2x max of the middle slice.

[sofroniewn]

@FedeClaudi @adamltyson I'm trying to chase down and eliminate any remaining contrast limit estimation bugs so that if users don't explicitly pass contrast_limits we don't choke. Can you confirm if you're still having any issues if you don't pass contrast_limits with the 0.2.12 napari release.

[sofroniewn]

Compute histogram on "some" images (e.g. every nth, or the middle n)

This is now the approach napari employs - see https://github.com/napari/napari/blob/dec5b9965d4a3fcea2d8314064465af14a97fec7/napari/layers/layer_utils.py#L25-L65 - it would be good to know if we are failing still

[adamltyson]

@sofroniewn I couldn't replicate this issue, @FedeClaudi is this something you can test? Perhaps by loading raw data into the amap viewer (assuming you can actually access your data currently)?

I'd like to remove my hacky fix and rely on napari dealing with this more gracefully if possible.

[adamltyson]

Looks to be working (eda3c10), but I could never fully reproduce this.