ERROR: model not working anymore #52
Comments
|
I am also planning to release a new model soon, let do let me know what I need to change in the |
|
So, I restarted Fiji and I ran it again..
Its like the axes got flipped.. |
|
HEllo @pr4deepr I hae downloaed the model at: https://zenodo.org/records/10460434/files/2D_enteric_ganglia_model_QA.zip?download=1 I have tried the model ´D_Ganglia_RGB_v2.bioimage.io.model` from that folder. It did not work for you at first because the output of the model has only one channel, where as the rdf.yaml specifies 3. Change the following: by this: Or this With this change, the model works. And produces the same output as the |
|
@pr4deepr Sorry for this! REgards, |
|
HEllo again @pr4deepr |
Having a packager of some sort is super handy.. Its quite overwhelming to figure out how to manually edit/create rdf.yaml file otherwise.. In the plugin it was really handy to be able to turn pre or post processing on or off. Is it possible to have like a
|
|
Hello @pr4deepr Does the model work now as expected? Carlos |
|
Thanks for all the suggestions
No, it doesn't. When I try:
Image on left is prediction and image on right is what I expect to get:
which is incorrect prediction.. Looks like a really low threshold was preformed?WIth this change
I get:
|
|
HEllo @pr4deepr |
|
@pr4deepr I have used the input .npy as the input image. Maybe try it yourself with the keras model to see if the results are what you expect. If they are, there must have been an error with the conversion from keras to tf1, if not the keras model might be corrupted. |
|
@carlosuc3m I am also running into the same channel error described above: [SERVICE-2] {“task”:“0bae35a2-4113-4b8c-9568-8b6737ea9a8b”,“requestType”:“EXECUTE”,“inputs”:{“modelSource”:“/Applications/Fiji.app/models/InstanSeg_brightfield_nuclei/instanseg.pt”,“modelFolder”:“/Applications/Fiji.app/models/InstanSeg_brightfield_nuclei”},“script”:“loadModel”} I am using the following model: https://github.com/instanseg/instanseg/releases/tag/instanseg_models_v1 I tried running on an image before and after splitting channels and still get the same error. Any insight into what needs to be amended? |
|
@pr4deepr |
|
@carlosuc3m Hmm. I downloaded the brightfield model and placed it in the models folder. It shows up and when I click run on the following image (https://drive.google.com/file/d/1USnhzYxHRQTWyxrBUsXPoE4IHTnejQm3/view?usp=sharing) I get the error above. I tried it on the native image and following splitting of channels. It sounds like it could be something with the image input if it works for you on the sample images? |
|
Hi In the zip file from zenodo, the model for deepimagej is located at: This is the one I currently have in the model:
using newest deepimagej version (3.1.0 - SNAPSHOT) I've updated today, on a test image: rgb image:
test image: works fine, but incorrect prediction rgb image: The model has been used successfully on older deepimagej version for past 7-8 months I believe, so the model is not corrupt. I'm not sure if the preprocessing is being performed in deepimagej currently using the macro: Cheers |
|
The models should run ok now. The pre- and post-processing were not being applied. Also the plugin was not converting directly from rgb color to rdg stack, that is why it seemed to have only one channel, instead of three. All that is fixed! If you update the plugin, it should work now. Sorry for the inconvenience! |
|
As you mentioned the pre-processing was not being applied. And specifically for your model, the result I get is not the same as the sample image, because the sample image was obtained without halo (without tiling). It seems that your model is quite reactive to tiling, so depending on the halo and image size,one result or another will be obtained. Please let me know if i can help you with anything else. Also regarding the macro, the plugin is not recordable anymore, but there is a new plugin that automatically generates the macro if you want. Regards, |
|
Hi Deepimagej 3.0.3 Image 1
Image 2 (download)
DeepImage 3.1.0 Image 1
Image 2
Cheers |
|
HEllo @pr4deepr |
|
Thanks, @carlosuc3m! I tried and when it runs all I get is a black screen as an output: 
When I try on the sample image, I also don't get a masked segmentation as an output: 
Any insight into where I may be going wrong? |
|
have you tried updating the plugin? |
|
@carlosuc3m yes that is why I don't get the tiling error anymore but now I'm running into the blank screen with no segmentation. |
|
could you provide the image please? @abadgerw |
|
@carlosuc3m here is my image: https://drive.google.com/file/d/1USnhzYxHRQTWyxrBUsXPoE4IHTnejQm3/view?usp=sharing I also see the same thing with the test image that comes with the instanseg model: https://github.com/instanseg/instanseg/releases/tag/instanseg_models_v1 |
|
This is the result i get with your sample image @abadgerw |
|
Just downloading your image and runnning with the latest updated version of deepIamgeJ. |
|
Thanks, @carlosuc3m. I uninstalled and reinstalled deepimageJ and got the same black screen. I also tried the adjustment of brightness/contrast which didn't seem to change the outcome. One thought: How should I be loading this custom model? I originally placed the unzipped file in the following folder: /Applications/Fiji.app/models. It shows up when running deepimageJ. However, in case that is the issue, instead, I tried using the install model manually command and that doesn't seem to work. I am using an Apple M1 Pro in case that is useful information. This is what I see in the console: Model loading: 1 And this is what my screen looks like: 
|
|
HEllo @abadgerw Maybe @cfusterbarcelo can try to reproduce this? |
This is an older model I used in older deepimagej version: link to model_yaml. in this file I think the settings were similar. Cheers |
|
Thanks, @carlosuc3m and @cfusterbarcelo! Do not hesitate to let me know if you require anything on my end as you troubleshoot. |
|
It should work now @abadgerw . |
|
Thanks, @carlosuc3m! I tried again and look to be still having the same issue. The console outputs the following: [SERVICE-0] {"task":"cb059d92-fff1-47c5-9539-bfbcec019f2d","requestType":"EXECUTE","inputs":{"modelSource":"/Applications/Fiji.app/models/brightfield_nuclei/instanseg.pt","modelFolder":"/Applications/Fiji.app/models/brightfield_nuclei"},"script":"loadModel"} And the output file is the following: 
|
|
@abadgerw |
|
what about pytorch 2? |
|
This comment was marked as resolved.
This comment was marked as resolved.
|
@carlosuc3m apologies for the delay. Here are the console outputs when running the updated plugin (the screen still show a black screen after running the model): [SERVICE-0] {"task":"12f3f646-08e8-473f-8e3b-0ff620a21f0f","requestType":"EXECUTE","inputs":{"modelSource":"/Applications/Fiji.app/models/brightfield_nuclei/instanseg.pt","modelFolder":"/Applications/Fiji.app/models/brightfield_nuclei"},"script":"loadModel"} |
|
HEllo @abadgerw I am back!! I need to update the deepImageJ update site and once it is done, I will let you know and if you want you can try it! Thanks a lot! |
|
Also @pr4deepr sorry for the delay with your issue. Now that this is fixed, I am fully back on the tiling thing! |
|
Thanks for letting me know! |
|
@abadgerw it should work now!! |
|
HEllo @pr4deepr issue identified! |
|
Thanks, @carlosuc3m! I now see an output! 
However, I was wondering why the image has a gradient (dark at the top with nothing detected)? It seems to be a different output compared to when I use this same model directly inside QuPath. |
|
HEllo @pr4deepr. If no more issues arise, a new dij version will be released today or tomorrow. Thanks for your patience and happy new year!! Regards, |
Hi
After the recent update, my deepimageJ bundled model as part of a software package is not working anymore.
You can download the model here:
https://zenodo.org/records/10460434and get the file:2D_enteric_ganglia_model_QA.zipThe model is under:
2D_Ganglia_RGB_v2.bioimage.io.model.zip
I get the following error when I run it on the test image:
The text was updated successfully, but these errors were encountered: