3264 Enhance DynUNet structure

[yiheng-wang-nv]

Fixes #3264 .

Description

Changes:

1. Make activation layers and filters configurable (in nnUNet, filters are determined via a rule, but for Brats21 1st place solution, this parameter is different)
2. Add a parameter to control the bias of transposed conv layers.
3. Make deep supervision heads more lightly, since only needed (depends on deep_supr_num ) heads are useful, and when deep_supervision=False, there is no need to keep deep_supervision_heads, thus an empty nn.ModuleList() will be created (see: https://github.com/Project-MONAI/MONAI/pull/3265/files#diff-0095d268268e7ee415e079c1f0ccabed72f2c722f92cb0780d2ba8133a088334R341).

Status

Ready

Types of changes

• Breaking change (fix or new feature that would cause existing functionality to change).
• New tests added to cover the changes.
• Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.
• Quick tests passed locally by running ./runtests.sh --quick --unittests.
• In-line docstrings updated.
• Documentation updated, tested make html command in the docs/ folder.

[wyli]

Thanks, to get this prepared for the upcoming release, could you please update AssertionError with either ValueError or TypeError so that the error types are more informative? and also remove DynUNetV1. cc @Nic-Ma @diazandr3s @SachidanandAlle

[madil90]

/build

[wyli]

/build

[wyli]

Thanks, it looks good to me.

[Nic-Ma]

I think these errors are related to this PR:

https://pipelines.actions.githubusercontent.com/l4HC8ZQBySk3IyeX2xuNSGAMiwZfqm8fOwamutFAAMLyYg6ITN/_apis/pipelines/1/runs/29280/signedlogcontent/4?urlExpires=2021-11-08T15%3A29%3A12.1656696Z&urlSigningMethod=HMACV1&urlSignature=LiJwnk6BaHBQMiwHaA3kHbqSm6FwXznYpytVSRDUxko%3D
======================================================================
ERROR: test_network_consistency_0_Dynunet (__main__.TestNetworkConsistency)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/opt/conda/lib/python3.8/site-packages/parameterized/parameterized.py", line 533, in standalone_func
    return func(*(a + p.args), **p.kwargs)
  File "tests/test_network_consistency.py", line 57, in test_network_consistency
    model.load_state_dict(loaded_data["model"])
  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1413, in load_state_dict
    raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for DynUNet:
        Missing key(s) in state_dict: "deep_supervision_heads.0.conv.conv.weight", "deep_supervision_heads.0.conv.conv.bias", "deep_supervision_heads.1.conv.conv.weight", "deep_supervision_heads.1.conv.conv.bias". 

======================================================================
ERROR: test_network_consistency_1_Unet (__main__.TestNetworkConsistency)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/opt/conda/lib/python3.8/site-packages/parameterized/parameterized.py", line 533, in standalone_func
    return func(*(a + p.args), **p.kwargs)
  File "tests/test_network_consistency.py", line 56, in test_network_consistency
    model = nets.__dict__[net_name](**model_params)
TypeError: __init__() missing 1 required positional argument: 'spatial_dims'

----------------------------------------------------------------------

Could you please help fix it here?

Thanks in advance.

[Nic-Ma]

/build

[yiheng-wang-nv]

I think these errors are related to this PR:

https://pipelines.actions.githubusercontent.com/l4HC8ZQBySk3IyeX2xuNSGAMiwZfqm8fOwamutFAAMLyYg6ITN/_apis/pipelines/1/runs/29280/signedlogcontent/4?urlExpires=2021-11-08T15%3A29%3A12.1656696Z&urlSigningMethod=HMACV1&urlSignature=LiJwnk6BaHBQMiwHaA3kHbqSm6FwXznYpytVSRDUxko%3D
======================================================================
ERROR: test_network_consistency_0_Dynunet (__main__.TestNetworkConsistency)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/opt/conda/lib/python3.8/site-packages/parameterized/parameterized.py", line 533, in standalone_func
    return func(*(a + p.args), **p.kwargs)
  File "tests/test_network_consistency.py", line 57, in test_network_consistency
    model.load_state_dict(loaded_data["model"])
  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1413, in load_state_dict
    raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for DynUNet:
        Missing key(s) in state_dict: "deep_supervision_heads.0.conv.conv.weight", "deep_supervision_heads.0.conv.conv.bias", "deep_supervision_heads.1.conv.conv.weight", "deep_supervision_heads.1.conv.conv.bias". 

======================================================================
ERROR: test_network_consistency_1_Unet (__main__.TestNetworkConsistency)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/opt/conda/lib/python3.8/site-packages/parameterized/parameterized.py", line 533, in standalone_func
    return func(*(a + p.args), **p.kwargs)
  File "tests/test_network_consistency.py", line 56, in test_network_consistency
    model = nets.__dict__[net_name](**model_params)
TypeError: __init__() missing 1 required positional argument: 'spatial_dims'

----------------------------------------------------------------------

Could you please help fix it here?

Thanks in advance.

Hi @Nic-Ma , @wyli the first error will be fixed if merging this PR (since the network structure of DynUNet changes). The second error seems due to Unet changes (use spatial_dims to replace dimensions). I think https://github.com/Project-MONAI/MONAI-extra-test-data/blob/main/unet_0/Unet_0.json should be modified.

[Nic-Ma]

OK, let's try again the blossom test in PR #3250 when this PR merged.

Thanks.

[wyli]

/build

[wyli]

/build

[wyli]

/build

[wyli]

/build

[wyli]

/build