Auto-segmentation model in sample app not selectable from GUI #652
Unanswered
bhoeckendorf
asked this question in
Q&A
Replies: 12 comments 11 replies
-
|
we recently fixed a dependency issue on monailabel.. the NVIDIA pretrained models were not getting pulled.. so you can uninstall and install latest version of monailabel as.. or the most simple way is git clone git@github.com:Project-MONAI/MONAILabel.git
cd MONAILabel
pip install -r requirements.txt
./monailabel/scripts/monailabel start_server -a sample-apps/segmentation -s <some test folder of images>
|
Beta Was this translation helpful? Give feedback.
0 replies
-
|
Thanks for the suggestion, @SachidanandAlle. Unfortunately, this doesn't solve the problem. Following your suggestion, I'm using latest MONAILabel from git, as well as the latest segmentation sample app from git. Slicer is a new install (v4.13.0-2022-02-24) with the label plugin installed through Slicer's extension manager. Does that install the latest version of the plugin? The auto segmentation model is still not selectable (screenshot below). The models appear to get downloaded (simplified file tree below): There are no error messages in the logs. The training task from the active learning section runs fine and yields model.pt in addition to the already existing pretrained.pt. Thanks for any further suggestions and please let me know if more info would help pinpoint the cause of this. |
Beta Was this translation helpful? Give feedback.
0 replies
-
|
Delete or move the model folder.. and start the server and share the logs.. logs can also be accessed from browser http://127.0.0.1:8000/logs?lines=0 |
Beta Was this translation helpful? Give feedback.
1 reply
-
|
Here are the logs, @SachidanandAlle. I created a new app and deleted the models folder. I also trained a model on data, the training logs are truncated for brevity. The model combobox stays inactive. |
Beta Was this translation helpful? Give feedback.
-
|
Basically you should see something like this in the log when u start the app.. when nothing was downloaded.. Full console logs for me.. |
Beta Was this translation helpful? Give feedback.
1 reply
-
|
This is without running any training.. when u train u update the existing model.. r u saying after u run training it became disabled? |
Beta Was this translation helpful? Give feedback.
-
|
Beta Was this translation helpful? Give feedback.
0 replies
-
|
How about http://127.0.0.1:8000/info If you are seeing the segmentation model then you should see the same in 3D Slicer as well.. |
Beta Was this translation helpful? Give feedback.
0 replies
-
|
If all is good.. and still 3D slicer is not enabled for auto segmentation model.. i suggest.. uninstall the plugin.. and configure the plugin in developer mode.. |
Beta Was this translation helpful? Give feedback.
0 replies
-
|
Thank you for all these suggestions, @SachidanandAlle. Sadly, the issue persists. The model combobox remains inactive and no model can be selected for inference. Neither the downloaded pretrained model, nor the updated trained model. The fact that the training works seems to indicate to me that everything is setup correctly, incl that the pretained model is downloaded correctly. I have switched to the development version of the Slicer plugin, as you suggested, but it doesn't help, either. Here are again my logs. I apologize for only including a link earlier, it is what GitHub does if you drag-&-drop a file. I also wouldn't have clicked on it. Below that you will also find the output of The output of |
Beta Was this translation helpful? Give feedback.
0 replies
-
|
I don't know why I see training in your logs.. can u just try simple things.. without training.. also paste the commands from beginning that you have tried.. and logs for them.. This looks something related to your env.. @diazandr3s can you check.. |
Beta Was this translation helpful? Give feedback.
5 replies
-
|
Everything above the training logs is just starting the server with the sample app. The model is unavailable already then. I merely included the training to help illustrate that things are generally setup and working. I apologize if it made things more complicated. |
Beta Was this translation helpful? Give feedback.
-
|
Here is my conda env: |
Beta Was this translation helpful? Give feedback.
-
Thanks, @bhoeckendorf. My recommendation would be to download again the preview release of Slicer, create a clean virtual env and install the latest monailabel-weekly (https://pypi.org/project/monailabel-weekly/) or pull the latest changes if working directly with the repository. If this doesn't work, I'm happy to meet and talk more about this. Please let us know how this goes. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you @SachidanandAlle and @diazandr3s for your help verifying the server is running correctly and your suggestions to debug the client. It turns out that after re-installing Slicer and pulling server and Slicer plugin to latest git, things are working as expected now. It is a mystery to me, since I did these things just a few days ago, before asking for your help here. The only difference now is that I deleted Slicer completely first, whereas before I had allowed MacOS to "replace" it with the latest snapshot. The odds of some leftover config interfering like this seem remote to me, but then again, something must have been amiss. In any case, thank you again for your time debugging this, and even more so for working on this badly needed project. |
Beta Was this translation helpful? Give feedback.
-
|
Glad to hear it is working :) |
Beta Was this translation helpful? Give feedback.
-
|
That's the problem which needs to be fixed first.. I gave you a snapshot of exact log which needs to be there when I start fresh.. Delete everything in model folder and start the app.. and look for the pretrained model to get downloaded.. Also check: http://127.0.0.1:8000/info |
Beta Was this translation helpful? Give feedback.
3 replies
-
|
The logs I attached is exactly that, followed by training afterwards. I'd be happy to redo it and send the logs, but it will be the same as the first ~50 lines of the logs you already have. |
Beta Was this translation helpful? Give feedback.
-
|
There is this line, and the file is created after starting the app. |
Beta Was this translation helpful? Give feedback.
-
|
Here are the logs after deleting the model folder and without training. I hope it is more helpful. Thank you for spending the time to look into it and help me with this. |
Beta Was this translation helpful? Give feedback.
-
|
When starting Slicer from command line and activating MONAILabel, I get this output: The segmentation and DeepGrow models aren't selectable in the GUI. The Scribbles model is though. |
Beta Was this translation helpful? Give feedback.
1 reply
-
|
From this.. server is all fine.. all the models are listed. I have no idea what happened to your slicer.. why it's showing something in disabled mode.. |
Beta Was this translation helpful? Give feedback.
-
|
Are you still facing the issue? Have u tried the latest version of MONAILabel and 3D Slicer clients? |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
I'm using the "segmentation" sample app practically as-is and expected to be able to select and use a model under the "Auto Segmentation" GUI section for inference. Instead, the model combobox in that section stays inactive (grayed out and doesn't react to mouse clicks).
Loading images from a custom dataset and running the active learning training task both work well. I'd be grateful for any pointers that help me debug or that clarify potential misconceptions.
Beta Was this translation helpful? Give feedback.
All reactions