clic

[ashmoy]

hello @allemangD I think that now here is only my commit, sorry

[allemangD]

Generally I like the structure of this, but I'm not sure about lazy_import.

It also looks like the conda environment is unused... in the other modules, the workers run in separate processes in the conda environment. Doing that would generally be more portable, but I think it may be incompatible with the threading model you have here.

Since you check dependencies and prompt before changing the environment, I'd say this could be merged as is, but I encourage you to look into using SlicerConda like the other modules instead of the main process for better portability.

[ashmoy]

Hi @allemangD, I’ve implemented the requested changes in CLIC. Thanks for your feedback!

[allemangD]

I think I did not explain my prior comment about the conda environment well. Apologies for any confusion. See details in the new inline comment.

[allemangD]

I think this does not prepare the environment as you expect. You should be able to self.conda.condaCreateEnv to prepare a conda environment with torch and the other dependencies - it won't modify the slicer environment, where the UI and _on_predict run. You'd need to move the processing code into a different script, and use self.conda.condaRunFilePython to run that script in the conda environment.

The threading model that you had before seemed fairly robust - it may be easier to simply revert these conda changes and use that instead, but I worry about cross-platform support and support on future versions of Slicer.

So I defer to you on which route to take:

• either switch back to the threading model you had before, removing the conda usage entirely
• or try to move the computation into a script that runs in the Conda environment that you prepared in the latest changes.

For reference, the "threading model" I refer to is this, from the prior revision of this PR.

SlicerAutomatedDentalTools/CLIC/CLIC.py

Lines 261 to 310 in 1efceef

	def onPredictButton(self):
	if not self.check_dependencies():
	return
	if not self.input_path:
	qt.QMessageBox.warning(self.parent, 'Warning', 'Please select an input file/folder')
	return
	if not self.model_folder:
	qt.QMessageBox.warning(self.parent, 'Warning', 'Please select a model folder')
	return
	param = {
	"input_path": self.input_path,
	"model_folder": self.model_folder,
	"output_dir": self.output_folder,
	"suffix": self.ui.suffixLineEdit.text
	}
	def update_progress(progress):
	self.ui_queue.put(("progress", progress))
	def update_log(message):
	self.ui_queue.put(("log", message))
	def display_callback(action, file_path):
	self.ui_queue.put((action, file_path))
	self.segmentationLoadedEvent = threading.Event()
	try:
	self.RunningUI(True)
	self.processThread = threading.Thread(
	target=self.logic.process,
	args=(param, update_progress, update_log, display_callback, self.segmentationLoadedEvent)
	)
	self.processThread.start()
	import time
	start_time = time.time()
	while self.processThread.is_alive():
	slicer.app.processEvents()
	current_time = time.time()
	self.ui.TimerLabel.setText(f"Time elapsed: {current_time - start_time:.2f}s")
	self.process_ui_queue()
	time.sleep(0.1)
	self.process_ui_queue()
	self.ui_queue.put(("log", "Segmentation completed successfully!"))
	except Exception as e:
	self.ui_queue.put(("log", f"An error occurred during segmentation: {str(e)}"))
	finally:
	self.RunningUI(False)

If you choose to move things to the Conda environment, you could use the ALI module as a reference. These sections in particular:

https://github.com/DCBIA-OrthoLab/SlicerAutomatedDentalTools/blob/main/ALI/ALI.py#L1107-L1191

https://github.com/DCBIA-OrthoLab/SlicerAutomatedDentalTools/blob/main/ALI/ALI.py#L1817-L1851