Skip to content

Inoutseg fix #25

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Hendrik-code wants to merge 8 commits into
base: main
Choose a base branch
from
Open

Inoutseg fix #25

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
7f72308
fixed _apply_region_mode logic by cloning the segmentation so as not …
Hendrik-code Jan 14, 2026
95a2a7c
fixed ordering of mix_prob and added it to inverse and histogram equa…
Hendrik-code Jan 14, 2026
d721980
embedded new arguments
Hendrik-code Jan 14, 2026
4fa9840
updated config file to newest observations
Hendrik-code Jan 14, 2026
20c93e8
add RandomChooseXTransformsGPU
NathanMolinier Jan 15, 2026
859c229
removed zscorenormalization as training and validation transform at t…
Hendrik-code Jan 21, 2026
2f1165e
Merge branch 'inoutseg_fix' of github.com:neuropoly/AugLab into inout…
Hendrik-code Jan 21, 2026
cd4330b
Account for transfrom probability in RandomChooseXTransformsGPU
NathanMolinier Jan 21, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
114 changes: 53 additions & 61 deletions auglab/configs/transform_params_gpu.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,9 +3,9 @@
"kernel_type": "Scharr",
"absolute": true,
"retain_stats": true,
"mix_prob": 0.90,
"in_seg": 0.5,
"out_seg": 0.5,
"mix_prob": 0.50,
"in_seg": 0.1,
"out_seg": 0.1,
"mix_in_out": true,
"probability": 0.15
},
Expand All @@ -14,18 +14,18 @@
"sigma": 1.0,
"retain_stats": false,
"mix_prob": 0.50,
"in_seg": 0.5,
"out_seg": 0.5,
"mix_in_out": false,
"probability": 0.20
"in_seg": 0.1,
"out_seg": 0.1,
"mix_in_out": true,
"probability": 0.15
},
"UnsharpMaskTransform": {
"kernel_type": "UnsharpMask",
"sigma": 1.0,
"unsharp_amount": 1.5,
"retain_stats": false,
"in_seg": 0.5,
"out_seg": 0.5,
"in_seg": 0.1,
"out_seg": 0.1,
"mix_in_out": true,
"mix_prob": 0.50,
"probability": 0.10
Expand All @@ -34,83 +34,85 @@
"kernel_type": "RandConv",
"kernel_sizes": [3,5,7],
"retain_stats": true,
"in_seg": 0.5,
"out_seg": 0.5,
"in_seg": 0.1,
"out_seg": 0.1,
"mix_in_out": true,
"mix_prob": 0.50,
"probability": 0.10
},
"RedistributeSegTransform": {
"in_seg": 0.2,
"in_seg": 0.25,
"retain_stats": true,
"probability": 0.5
"probability": 0.1
},
"GaussianNoiseTransform": {
"mean": 0.0,
"std": 0.1,
"std": 1.0,
"in_seg": 0.5,
"out_seg": 0.5,
"mix_in_out": true,
"probability": 0.10
"probability": 0.15
},
"ClampTransform": {
"max_clamp_amount": 0.2,
"retain_stats": false,
"in_seg": 0.5,
"out_seg": 0.5,
"retain_stats": true,
"in_seg": 0.1,
"out_seg": 0.1,
"mix_in_out": true,
"probability": 0.40
"probability": 0.05
},
"BrightnessTransform": {
"brightness_range": [0.75, 1.25],
"in_seg": 0.0,
"out_seg": 0.0,
"mix_in_out": false,
"probability": 0.15
"in_seg": 0.1,
"out_seg": 0.1,
"mix_in_out": true,
"probability": 0.5
},
"GammaTransform": {
"gamma_range": [0.7, 1.5],
"retain_stats": true,
"in_seg": 0.5,
"out_seg": 0.5,
"in_seg": 0.1,
"out_seg": 0.1,
"mix_in_out": true,
"probability": 0.30
"probability": 0.50
},
"InvGammaTransform": {
"gamma_range": [0.7, 1.5],
"retain_stats": true,
"in_seg": 0.5,
"out_seg": 0.5,
"in_seg": 0.1,
"out_seg": 0.1,
"mix_in_out": true,
"probability": 0.10
"probability": 0.30
},
"ContrastTransform": {
"contrast_range": [0.75, 1.25],
"retain_stats": false,
"in_seg": 0.0,
"out_seg": 0.0,
"mix_in_out": false,
"probability": 0.15
"in_seg": 0.1,
"out_seg": 0.1,
"mix_in_out": true,
"probability": 0.5
},
"FunctionTransform": {
"retain_stats": true,
"in_seg": 0.5,
"out_seg": 0.5,
"mix_in_out": true,
"probability": 0.05
"in_seg": 0,
"out_seg": 0,
"mix_in_out": false,
"probability": 0.025
},
"InverseTransform": {
"retain_stats": true,
"in_seg": 0.5,
"out_seg": 0.5,
"in_seg": 0.1,
"out_seg": 0.1,
"mix_in_out": true,
"probability": 0.05
"mix_prob": 0.50,
"probability": 0.1
},
"HistogramEqualizationTransform": {
"retain_stats": false,
"in_seg": 0.5,
"out_seg": 0.5,
"retain_stats": true,
"in_seg": 0,
"out_seg": 0,
"mix_in_out": true,
"mix_prob": 0.80,
"probability": 0.10
},
"SimulateLowResTransform": {
Expand All @@ -123,14 +125,14 @@
"scale": [0.6, 1.0],
"crop": [1.0, 1.0],
"same_on_batch": false,
"probability": 0.40
"probability": 0.05
},
"BiasFieldTransform": {
"retain_stats": false,
"coefficients": 0.5,
"in_seg": 0.5,
"out_seg": 0.5,
"mix_in_out": false,
"retain_stats": true,
"coefficients": 0.2,
"in_seg": 0.1,
"out_seg": 0.1,
"mix_in_out": true,
"probability": 0.10
},
"FlipTransform": {
Expand All @@ -145,19 +147,9 @@
"scale": [0.7, 1.4],
"shear": [-5, 5, -5, 5, -5, 5],
"resample": "bilinear",
"probability": 0
},
"nnUNetSpatialTransform": {
"patch_center_dist_from_border": 80,
"random_crop": true,
"p_elastic_deform": 0.2,
"p_rotation": 0.5,
"p_scaling": 0.5,
"scaling": [0.7, 1.4],
"p_synchronize_scaling_across_axes": 1,
"bg_style_seg_sampling": false
"probability": 0.5
},
"ZscoreNormalizationTransform": {
"probability": 0.3
"probability": 0.0
}
Copy link

Copilot AI Jan 22, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The nnUNetSpatialTransform configuration section has been completely removed from the JSON config file. However, the code in nnUNetTrainerDAExt.py (line 214) still checks for this key with if 'nnUNetSpatialTransform' in config:. While the code handles the missing key gracefully by using default values, this removal represents a significant configuration change that could affect spatial augmentation behavior. Consider documenting why this section was removed or if spatial augmentations are now handled differently.

Suggested change
}
},
"_comment_nnUNetSpatialTransform": "The former 'nnUNetSpatialTransform' configuration block was intentionally removed. Spatial augmentations now rely on default parameters in nnUNetTrainerDAExt.py when this key is absent."

Copilot uses AI. Check for mistakes.
}
19 changes: 9 additions & 10 deletions auglab/trainers/nnUNetTrainerDAExt.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -210,7 +210,7 @@ def get_training_transforms(
else:
patch_size_spatial = patch_size
ignore_axes = None

if 'nnUNetSpatialTransform' in config:
spatial_params = config['nnUNetSpatialTransform']
else:
Expand All @@ -234,7 +234,7 @@ def get_training_transforms(

if do_dummy_2d_data_aug:
transforms.append(Convert2DTo3DTransform())

if use_mask_for_norm is not None and any(use_mask_for_norm):
transforms.append(MaskImageTransform(
apply_to_channels=[i for i in range(len(use_mask_for_norm)) if use_mask_for_norm[i]],
Expand Down Expand Up @@ -284,8 +284,8 @@ def get_training_transforms(
channel_in_seg=0
)
)
transforms.append(ZscoreNormalization())

# transforms.append(ZscoreNormalization())

# NOTE: DownsampleSegForDSTransform is now handled in train_step for GPU augmentations
# if deep_supervision_scales is not None:
Expand Down Expand Up @@ -323,13 +323,13 @@ def get_validation_transforms(
channel_in_seg=0
)
)
transforms.append(ZscoreNormalization())

# transforms.append(ZscoreNormalization())

if deep_supervision_scales is not None:
transforms.append(DownsampleSegForDSTransform(ds_scales=deep_supervision_scales))
return ComposeTransforms(transforms)

def train_step(self, batch: dict) -> dict:
data = batch['data']
target = batch['target']
Expand All @@ -350,13 +350,13 @@ def train_step(self, batch: dict) -> dict:
with autocast(self.device.type, enabled=True) if self.device.type == 'cuda' else dummy_context():
# Apply GPU augmentations to full-resolution data/target
data, target = self.transforms(data, target)

# Create multi-scale targets for deep supervision after augmentation
deep_supervision_scales = self._get_deep_supervision_scales()
if deep_supervision_scales is not None:
ds_transform = DownsampleSegForDSTransformCustom(ds_scales=deep_supervision_scales)
target = ds_transform(target)

output = self.network(data)
# del data
l = self.loss(output, target)
Expand Down Expand Up @@ -519,4 +519,3 @@ def train_step(self, batch: dict) -> dict:
torch.nn.utils.clip_grad_norm_(self.network.parameters(), 12)
self.optimizer.step()
return {'loss': l.detach().cpu().numpy()}

Loading
Loading