Open utterances-bot opened 1 year ago
This model still has the sample_input.tif and sample_output.tif saved with incorrect dimensions. The Z and X dimensions are interchanged. The conversion from .npy to .tif was corrected at https://github.com/bioimage-io/core-bioimage-io-python/pull/275
Also zenodo links take surprisingly long, the downloader seems to be doing nothing
It now works better. It seems that the problem was on zenodo. It would be interesting to notify problems with the connection or zenodo.
It now works better. It seems that the problem was on zenodo. It would be interesting to notify problems with the connection or zenodo.
do you mean to zenodo or when using the bioimageio.core library?
"To zenodo" yess, sorry for the typo.
we'll likely host model packages separately from zenodo in the future to minimize downtime. This was discussed before and we decided for now to rely on zenodo for simplicity and out of lack of readily available alternatives.
The sample input tif image axes are still transposed @constantinpape
Segmentation outputs black image on own data from 3D zebrafish brain nuclei (.tif). Algorithm not suitable for data or genuine bug?
Hi! Are you trying it in Fiji? Please, use Fiji > Image > Adjust > Brightness/Contrast to display the values 0 and 1. The output is a binary mask and often, the display by default considers it to have values in the range from 0 to 255. Also, you can calculate the histogram of the image to make sure that there are no other values than 0 in your image
Hi! Are you trying it in Fiji? Please, use Fiji > Image > Adjust > Brightness/Contrast to display the values 0 and 1. The output is a binary mask and often, the display by default considers it to have values in the range from 0 to 255. Also, you can calculate the histogram of the image to make sure that there are no other values than 0 in your image
I have used Fiji to try to adjust contrast, but the binary mask output only shows NaN.
Can you please share some screenshots?
@esgomezm Here are the screenshots. The input file is called 64cubed.tif
Ok, so this is the Test run API in the browser that uses ImageJ.JS rather than Fiji ;)
I think the image has too many z-planes and the model run seems to collapse (please @oeway @Nanguage @cfusterbarcelo maybe you can check this one).
Meanwhile, @postnubilaphoebus, if you cut your image to have less z-planes (32 for example), it will work (I tried with a similar image and it worked). Please, note that this image seems to be quite different with respect to the image data type used to train this model.
Yet another solution if you want to process the entire volume is to use deepImageJ in your local computer using Fiji
I just talked to Weize (@Nanguage) he will look into that issue when he have time (he is in the deepimagej hackathon right now).
Thank you!
Results will not be displayed if the model crashes or fails, but according to the screenshot this is not the case. And after I used the test data to conduct experiments, I found no problems, so I tend to think that the model is not suitable for the characteristics of the data.
Hello! I just tried this model and noticed that it has 85M parameters. It might be helpful to add this information to the model card for the end user. Some people may prefer to try lightweight models, and currently, there is no way to see this info at a quick glance before using the model.
Neuron Segmentation in EM (Membrane Prediction)
Bioimage.io -- an AI model repository for deep learning.
https://bioimage.io/?id=10.5281%2Fzenodo.5874741