bioimage-io / collection-bioimage-io

(deprecated in favor of bioimage-io/collection) RDF collection for BioImage.IO
5 stars 9 forks source link

Pancreatic Phase Contrast Cell Segmentation (U-Net) #445

Open utterances-bot opened 2 years ago

utterances-bot commented 2 years ago

Pancreatic Phase Contrast Cell Segmentation (U-Net)

Bioimage.io -- an AI model repository for deep learning.

https://bioimage.io/?tags=pancreatic&id=10.5281%2Fzenodo.5914248&type=model

carlosuc3m commented 2 years ago

the link for the h5 weights is broken (https://zenodo.org/api/files/2d92e820-1b25-40f5-a2af-3219df97d7f7/unet_pancreatic_cell_segmentation.h5)

FynnBe commented 2 years ago

the file name is wrong. there is a keras_model.h5 file in the record. (keras weights are not validated (yet), so we didn't notice this). @iarganda could you please update the model accordingly?

esgomezm commented 2 years ago

Please point to the maintainers! I have corrected the specifications. Now it should work. Also, why does the model download say ZeroCostDL4Mic if the model is not compatible with ZeroCostDL4Mic?

esgomezm commented 2 years ago

@carlosuc3m can you check whether it works?

carlosuc3m commented 2 years ago

I will check it now with the BioEngine. however the yaml still points to a broken link: https://zenodo.org/api/files/2d92e820-1b25-40f5-a2af-3219df97d7f7/unet_pancreatic_cell_segmentation.h5 @esgomezm

carlosuc3m commented 2 years ago

I just checked the model and it does not work on the BioEngine. I guess it is because of the incorrect link in the yaml file, the source should be pointing to the correct file @esgomezm @oeway

esgomezm commented 2 years ago

@carlosuc3m it should work now! I have just updated the model manually in zenodo because the updater does not update the rdf.yaml file correctly in the zenodo repository. @oeway I remember this was already a problem a while ago and seems that it's still not working properly. Any idea?

carlosuc3m commented 2 years ago

it still does not work. The Bioimage.io collections rdf.yaml files keep pointing towards the incorrect keras weight source link. This model runs on keras on the BioEngine, so the keras link should not be broken and should point to a correct Keras model @esgomezm

esgomezm commented 2 years ago

Hi I cannot help with this one. The RDF file in the zenodo repository is correct and points to the keras model. There's something when updating the model that seems to be broken as it does not update correctly the model specifications. @oeway is probably not able to fix at the moment. @FynnBe do you know how to solve this?

esgomezm commented 2 years ago

Hi there! @oeway I'm trying to load this model using the bioimageio.core and still have the same problem: the system does not update the link to the keras weights. It still points to https://zenodo.org/api/files/2d92e820-1b25-40f5-a2af-3219df97d7f7/unet_pancreatic_cell_segmentation.h5

Any idea about how to solve this one?

oeway commented 2 years ago

Hi there! @oeway I'm trying to load this model using the bioimageio.core and still have the same problem: the system does not update the link to the keras weights. It still points to https://zenodo.org/api/files/2d92e820-1b25-40f5-a2af-3219df97d7f7/unet_pancreatic_cell_segmentation.h5

Any idea about how to solve this one?

I just checked, it appears that the latest version is not generated which make sense, since we decided to block it. Here, we said that @esgomezm will update with a newer version.

Here is the resource.yaml: https://github.com/bioimage-io/collection-bioimage-io/blob/main/collection/10.5281/zenodo.5914248/resource.yaml

And here are the available versions, note that the latest two versions in the resource.yaml are blocked. https://github.com/bioimage-io/collection-bioimage-io/tree/gh-pages/rdfs/10.5281/zenodo.5914248

Therefore, @esgomezm you should perhaps upload with a newer version (although I don't remember what needed to be changed for this one ;)

carlosuc3m commented 2 years ago

This model is still pointing to invalid keras weights @esgomezm @lmoyasans

carlosuc3m commented 1 year ago

Still broken

carlosuc3m commented 1 year ago

still invalid keras weights