Monai Pathology Nuclick Training

[OmarAshkar]

Hello,
I have few problems with the sample pathology app and many parameters seems hard-coded here and here and hard for me to trace.

I want to train a nuclick for easier labeling. I create some annotations, make rectangle and submit it. Then try to train nuclick

1. The created images are 1024x1024. where the patches submitted are way smaller. (I submit like 100x100) the reminder of the image is just black.
2. There is a label mismatch, the created patches in train_ds always have label 1, even if the annotated images has object that should be labeled 3.
3. On training Nuclick, I get the following error :

Traceback (most recent call last):
  File "/opt/monai/monai/utils/module.py", line 260, in instantiate
    return component(**kwargs)
  File "/opt/monai/monai/data/dataloader.py", line 91, in __init__
    super().__init__(dataset=dataset, num_workers=num_workers, **kwargs)
  File "/usr/local/lib/python3.8/dist-packages/torch/utils/data/dataloader.py", line 352, in __init__
    sampler = RandomSampler(dataset, generator=generator)  # type: ignore[arg-type]
  File "/usr/local/lib/python3.8/dist-packages/torch/utils/data/sampler.py", line 107, in __init__
    raise ValueError("num_samples should be a positive integer "
ValueError: num_samples should be a positive integer value, but got num_samples=0
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
  File "/opt/monai/monai/bundle/config_item.py", line 292, in instantiate
    return instantiate(modname, mode, **args)
  File "/opt/monai/monai/utils/module.py", line 270, in instantiate
    raise RuntimeError(
RuntimeError: Failed to instantiate component 'monai.data.dataloader.DataLoader' with kwargs: {'dataset': <monai.data.dataset.CacheDataset object at 0x2abe37fa61f0>, 'batch_size': 64, 'shuffle': True, 'num_workers': 4}
 set '_mode_=debug' to enter the debugging mode.
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
  File "/usr/lib/python3.8/runpy.py", line 194, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/usr/lib/python3.8/runpy.py", line 87, in _run_code
    exec(code, run_globals)
  File "/usr/local/lib/python3.8/dist-packages/monailabel/interfaces/utils/app.py", line 128, in <module>
    run_main()
  File "/usr/local/lib/python3.8/dist-packages/monailabel/interfaces/utils/app.py", line 113, in run_main
    result = a.train(request)
  File "/usr/local/lib/python3.8/dist-packages/monailabel/interfaces/app.py", line 423, in train
    result = task(request, self.datastore())
  File "/usr/local/lib/python3.8/dist-packages/monailabel/tasks/train/bundle.py", line 335, in __call__
    self.run_single_gpu(request, overrides)
  File "/usr/local/lib/python3.8/dist-packages/monailabel/tasks/train/bundle.py", line 344, in run_single_gpu
    monai.bundle.run(
  File "/opt/monai/monai/utils/deprecate_utils.py", line 223, in _wrapper
    return func(*args, **kwargs)
  File "/opt/monai/monai/bundle/scripts.py", line 695, in run
    workflow.run()
  File "/opt/monai/monai/bundle/workflows.py", line 262, in run
    return self._run_expr(id=self.run_id)
  File "/opt/monai/monai/bundle/workflows.py", line 296, in _run_expr
    return self.parser.get_parsed_content(id, **kwargs) if id in self.parser else None
  File "/opt/monai/monai/bundle/config_parser.py", line 291, in get_parsed_content
    return self.ref_resolver.get_resolved_content(id=id, **kwargs)
  File "/opt/monai/monai/bundle/reference_resolver.py", line 190, in get_resolved_content
    return self._resolve_one_item(id=id, **kwargs)
  File "/opt/monai/monai/bundle/reference_resolver.py", line 160, in _resolve_one_item
    self._resolve_one_item(id=d, waiting_list=waiting_list, **kwargs)
  File "/opt/monai/monai/bundle/reference_resolver.py", line 160, in _resolve_one_item
    self._resolve_one_item(id=d, waiting_list=waiting_list, **kwargs)
  File "/opt/monai/monai/bundle/reference_resolver.py", line 160, in _resolve_one_item
    self._resolve_one_item(id=d, waiting_list=waiting_list, **kwargs)
  File "/opt/monai/monai/bundle/reference_resolver.py", line 168, in _resolve_one_item
    self.resolved_content[id] = item.instantiate() if kwargs.get("instantiate", True) else item
  File "/opt/monai/monai/bundle/config_item.py", line 294, in instantiate
    raise RuntimeError(f"Failed to instantiate {self}.") from e
RuntimeError: Failed to instantiate {'_target_': 'DataLoader', 'dataset': <monai.data.dataset.CacheDataset object at 0x2abe37fa61f0>, 'batch_size': 64, 'shuffle': True, 'num_workers': 4}.
[2023-08-21 16:50:01,768] [119801] [ThreadPoolExecutor-2_0] [INFO] (monailabel.utils.async_tasks.utils:83) - Return code: 1

There is something mentioned called debug-mode by also don't know how to activate it.

Thanks alot~

[diazandr3s]

Hi @OmarAshkar,

From the logs, I can see there are no training samples or the moe is not recognizing them.

ValueError: num_samples should be a positive integer value, but got num_samples=0

As you can see in the config file of the nuclick model, we use the pathology_nuclick_annotation bundle: https://github.com/Project-MONAI/MONAILabel/blob/main/sample-apps/pathology/lib/configs/nuclick.py#L32C24-L32C52

Here is the Readme and more information about that bundle: https://github.com/Project-MONAI/model-zoo/tree/dev/models/pathology_nuclick_annotation

FYI: You could run the pathology app as a typical Python main script and without the viewer. For this, please check the main function within the pathology app: https://github.com/Project-MONAI/MONAILabel/blob/main/sample-apps/pathology/main.py#L178-L183

Using your preferred IDE, you could also add breakpoints and debug any errors.

I hope this helps,