search

vanvalenlab / deepcell-label

Cloud-based data annotation tools for biological images
https://label.deepcell.org
Other
77 stars 15 forks source link

Cannot upload 16-bit, 4.6GB, 3.55X3.42mm, .tif of 29 channel multiplex image #527

Closed smith6jt-cop closed 1 year ago

smith6jt-cop commented 1 year ago

Describe the bug Getting "undefined" error after hitting upload button on website.

To Reproduce Steps to reproduce the behavior:

Upload by drag and drop or by choosing file. Choose any of the available dimension options in the dropdown menu.

Expected behavior The file to upload.

Screenshots The image I am trying to upload as it appears in FIJI: image image

Desktop (please complete the following information):

Additional context My image opens in QuPath as a single plane, single timepoint image with 29 channels.

ykevu commented 1 year ago

What is the dimension order of your image? It might be difficult to diagnose the problem without the file itself also, do you have a smaller example file of the same format?

smith6jt-cop commented 1 year ago

Thank you for your response. I believe the dimension order is XYC although sometimes different programs interpret dimension order differently (see https://forum.image.sc/t/multichannel-ome-tiff-import-order-xyzct-vs-xyczt/45960) so I tried uploading with each available option in the dropdown menu. 20_008-SP_CC2-B_11092020_reg001_chsNEWcropsmall.zip

rossbar commented 1 year ago

From the OP it looks like this is a 9.4k x 9k pixel image; currently the maximum supported size by deepcell-label is 2048 x 2048 pixels.

ykevu commented 1 year ago

Since the smaller cropped image seems to work, I think @rossbar is right that this is an issue of exceeding the supported size.

rossbar commented 1 year ago

Sorry, I got my wires crossed a bit - the explicit 2048x2048 limitation is for deepcell.org, not label.deepcell.org. Nevertheless, this is almost certainly a data size issue. In terms of workarounds, pre-tiling the image and uploading the tiles separately is probably the easiest workaround (quadrants should work). We'd like to support arbitrarily large images but it's not on the immediate roadmap.