Error during inference if patch size is larger than image

[fepegar]

If the patch size is larger than the image size during inference with TorchIO, an error is raised. This implies that this situation wasn't being covered by the tests.

To do:

• [ ] Write a unit test for this
• [ ] Constrain the patch size to the image size during inference

Error:

Traceback (most recent call last):
  File "InnerEyePrivate/ML/runner.py", line 50, in <module>
    main()
  File "InnerEyePrivate/ML/runner.py", line 46, in main
    post_cross_validation_hook=runner.default_post_cross_validation_hook)
  File "/mnt/azureml/cr/j/5a59e32ccd764940b8143a7327f35cb5/exe/wd/innereye-deeplearning/InnerEye/ML/runner.py", line 451, in run
    return runner.run()
  File "/mnt/azureml/cr/j/5a59e32ccd764940b8143a7327f35cb5/exe/wd/innereye-deeplearning/InnerEye/ML/runner.py", line 218, in run
    self.run_in_situ(azure_run_info)
  File "/mnt/azureml/cr/j/5a59e32ccd764940b8143a7327f35cb5/exe/wd/innereye-deeplearning/InnerEye/ML/runner.py", line 407, in run_in_situ
    self.ml_runner.run()
  File "/mnt/azureml/cr/j/5a59e32ccd764940b8143a7327f35cb5/exe/wd/innereye-deeplearning/InnerEye/ML/run_ml.py", line 351, in run
    self.run_inference(checkpoint_paths_for_testing, ModelProcessing.DEFAULT)
  File "/mnt/azureml/cr/j/5a59e32ccd764940b8143a7327f35cb5/exe/wd/innereye-deeplearning/InnerEye/ML/run_ml.py", line 474, in run_inference
    model_proc=model_proc)
  File "/mnt/azureml/cr/j/5a59e32ccd764940b8143a7327f35cb5/exe/wd/innereye-deeplearning/InnerEye/ML/run_ml.py", line 698, in model_inference_train_and_test
    model_proc=model_proc)
  File "/mnt/azureml/cr/j/5a59e32ccd764940b8143a7327f35cb5/exe/wd/innereye-deeplearning/InnerEye/ML/model_testing.py", line 72, in model_test
    return segmentation_model_test(config, data_split, checkpoint_paths, model_proc)
  File "/mnt/azureml/cr/j/5a59e32ccd764940b8143a7327f35cb5/exe/wd/innereye-deeplearning/InnerEye/ML/model_testing.py", line 102, in segmentation_model_test
    epoch_and_split=epoch_and_split)
  File "/mnt/azureml/cr/j/5a59e32ccd764940b8143a7327f35cb5/exe/wd/innereye-deeplearning/InnerEye/ML/model_testing.py", line 166, in segmentation_model_test_epoch
    voxel_spacing_mm=sample.metadata.image_header.spacing
  File "/mnt/azureml/cr/j/5a59e32ccd764940b8143a7327f35cb5/exe/wd/innereye-deeplearning/InnerEye/ML/pipelines/inference.py", line 47, in predict_and_post_process_whole_image
    return self.post_process(self.predict_whole_image(image_channels, voxel_spacing_mm, mask, patient_id))
  File "/azureml-envs/azureml_019d9b5c5814e5f53d425814bccd13a8/lib/python3.7/site-packages/torch/autograd/grad_mode.py", line 28, in decorate_context
    return func(*args, **kwargs)
  File "/mnt/azureml/cr/j/5a59e32ccd764940b8143a7327f35cb5/exe/wd/innereye-deeplearning/InnerEye/ML/pipelines/inference.py", line 272, in predict_whole_image
    padding_mode=self.model_config.padding_mode.value,
  File "/azureml-envs/azureml_019d9b5c5814e5f53d425814bccd13a8/lib/python3.7/site-packages/torchio/data/sampler/grid.py", line 77, in __init__
    self.locations = self._compute_locations(self.subject)
  File "/azureml-envs/azureml_019d9b5c5814e5f53d425814bccd13a8/lib/python3.7/site-packages/torchio/data/sampler/grid.py", line 102, in _compute_locations
    self._parse_sizes(*sizes)
  File "/azureml-envs/azureml_019d9b5c5814e5f53d425814bccd13a8/lib/python3.7/site-packages/torchio/data/sampler/grid.py", line 131, in _parse_sizes
    raise ValueError(message)
ValueError: Patch size (48, 64, 48) cannot be larger than image size (35, 51, 35)

AB#5114

[fepegar]

For some reason, this was not automatically corrected by

https://github.com/microsoft/InnerEye-DeepLearning/blob/821cb3be7a1b9cebbd06d49481e2024aa7cd0b06/InnerEye/ML/models/architectures/base_model.py#L77-L80