# miagrouput/augmentations-for-the-unknown architecture

> Architecture diagram of the GitHub repository miagrouput/augmentations-for-the-unknown, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-10-02). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/miagrouput/augmentations-for-the-unknown
- Repository: https://github.com/miagrouput/augmentations-for-the-unknown

## Overview

This repository implements MRI image augmentations for out-of-distribution segmentation, plus nnU-Net training variants and a dataset-generation utility. A benchmark-generation workflow loads image/label pairs, applies configured transforms at severity levels, and writes transformed samples. The training branch adds augmentation and loss variants around nnU-Net. The README documents the research goal and pretrained models/testing data, but not a runtime pipeline; several trainer variants and transform implementations were not sampled, so their connections are omitted where unverified.

Main entry points: [`nnunet/training/network_training/nnUNetTrainerNoDAV2FLCE_MixUp_AFA.py`](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/network_training/nnUNetTrainerNoDAV2FLCE_MixUp_AFA.py) (AFA MixUp trainer), [`nnunet/training/network_training/nnUNetTrainerV2_CutMixUpAFA.py`](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/network_training/nnUNetTrainerV2_CutMixUpAFA.py) (CutMix trainer), [`medg/data/generator.py`](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/data/generator.py) (Dataset generator), [`medg/transforms/defaults.py`](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/transforms/defaults.py) (Default settings)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_generation["Dataset generation"]
  node_generator["Dataset generator<br/>[generator.py]"]
  node_data_utils["Data utilities<br/>[utils.py]"]
  node_misc["Attribute utilities<br/>[misc.py]"]
  node_writer["NIfTI writer<br/>[writer.py]"]
end

subgraph group_transforms["MRI augmentations"]
  node_transform_settings["Default settings<br/>[defaults.py]"]
  node_artifact_transforms["MRI artifacts<br/>[random_spike_n.py]"]
  node_kspace_sampling["K-space sampling<br/>[subsample.py]"]
  node_kspace_transforms["K-space transforms"]
  node_iso_downsample["Isotropic downsampling<br/>[iso_downsample.py]"]
  node_other_transforms["Additional transforms<br/>[defaults.py]"]
end

subgraph group_training["Segmentation training"]
  node_trainer["FLCE trainer"]
  node_trainer_afa_mixup["AFA MixUp trainer"]
  node_trainer_cutmix["CutMix trainer"]
  node_afa["AFA augmentation<br/>[afa.py]"]
  node_mixup["MixUp augmentation<br/>[mixup.py]"]
  node_cutmix["CutMix augmentation<br/>[cutmix.py]"]
  node_losses["Segmentation losses<br/>[dice_loss.py]"]
  node_deep_supervision["Deep supervision loss"]
  node_crossentropy["Cross entropy<br/>[crossentropy.py]"]
  node_focal_loss["Focal loss<br/>[focal_loss.py]"]
  node_unet["Generic U-Net<br/>[generic_unet.py]"]
  node_model_helper["Model helper<br/>[nnunet_helper.py]"]
  node_dubin["DuBIN normalization<br/>[dubin.py]"]
  node_dubn["Dual batch norm<br/>[dubn.py]"]
end

subgraph group_analysis["Model analysis"]
  node_model_surgery["Latent model analysis<br/>[model_surgery.py]"]
end

node_input(("Input scans"))
node_output[("Generated dataset")]
node_analyzer_input(("Scans and labels"))

node_input -->|"provides scans"| node_generator
node_generator -->|"loads defaults"| node_transform_settings
node_generator -->|"uses helpers"| node_misc
node_generator -->|"writes samples"| node_writer
node_transform_settings -->|"configures"| node_artifact_transforms
node_transform_settings -->|"configures"| node_other_transforms
node_other_transforms -->|"registers"| node_iso_downsample
node_other_transforms -->|"registers"| node_kspace_transforms
node_kspace_transforms -->|"uses"| node_kspace_sampling
node_writer -->|"saves NIfTI"| node_output
node_trainer_afa_mixup -->|"extends"| node_trainer
node_trainer_afa_mixup -->|"applies"| node_afa
node_trainer_afa_mixup -->|"applies"| node_mixup
node_trainer_afa_mixup -->|"uses"| node_focal_loss
node_trainer_cutmix -->|"applies"| node_afa
node_trainer_cutmix -->|"applies"| node_mixup
node_trainer_cutmix -->|"applies"| node_cutmix
node_trainer_cutmix -->|"uses"| node_losses
node_trainer -->|"uses"| node_deep_supervision
node_model_helper -->|"constructs"| node_unet
node_model_helper -->|"converts normalization"| node_dubin
node_dubin -->|"uses"| node_dubn
node_analyzer_input -->|"analyzes scans"| node_model_surgery
node_model_surgery -->|"loads model"| node_model_helper
node_model_surgery -->|"runs inference"| node_unet

click node_generator "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/data/generator.py"
click node_transform_settings "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/transforms/defaults.py"
click node_artifact_transforms "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/transforms/additional/random_spike_n.py"
click node_kspace_sampling "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/transforms/additional/fastMRI/subsample.py"
click node_kspace_transforms "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/transforms/additional/k_space_subsampling.py"
click node_iso_downsample "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/transforms/additional/iso_downsample.py"
click node_other_transforms "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/transforms/additional/defaults.py"
click node_data_utils "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/data/utils.py"
click node_misc "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/utils/misc.py"
click node_writer "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/utils/writer.py"
click node_trainer "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/network_training/nnUNetTrainerNoDAV2FLCE.py"
click node_trainer_afa_mixup "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/network_training/nnUNetTrainerNoDAV2FLCE_MixUp_AFA.py"
click node_trainer_cutmix "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/network_training/nnUNetTrainerV2_CutMixUpAFA.py"
click node_afa "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/data_augmentation/afa.py"
click node_mixup "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/data_augmentation/mixup.py"
click node_cutmix "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/data_augmentation/cutmix.py"
click node_losses "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/loss_functions/dice_loss.py"
click node_deep_supervision "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/loss_functions/deep_supervision.py"
click node_crossentropy "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/loss_functions/crossentropy.py"
click node_focal_loss "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/loss_functions/focal_loss.py"
click node_unet "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/ml/model/segmentation/nnunet/generic_unet.py"
click node_model_helper "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/ml/model/segmentation/nnunet/nnunet_helper.py"
click node_dubin "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/network_architecture/custom_modules/dubin.py"
click node_dubn "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/network_architecture/custom_modules/dubn.py"
click node_model_surgery "https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/model_surgery.py"

classDef toneNeutral fill:#f8fafc,stroke:#334155,stroke-width:1.5px,color:#0f172a
classDef toneBlue fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef toneAmber fill:#fef3c7,stroke:#d97706,stroke-width:1.5px,color:#78350f
classDef toneMint fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef toneRose fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
classDef toneIndigo fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef toneTeal fill:#ccfbf1,stroke:#0f766e,stroke-width:1.5px,color:#134e4a
class node_generator,node_data_utils,node_misc,node_writer toneBlue
class node_transform_settings,node_artifact_transforms,node_kspace_sampling,node_kspace_transforms,node_iso_downsample,node_other_transforms,node_output toneAmber
class node_trainer,node_trainer_afa_mixup,node_trainer_cutmix,node_afa,node_mixup,node_cutmix,node_losses,node_deep_supervision,node_crossentropy,node_focal_loss,node_unet,node_model_helper,node_dubin,node_dubn toneMint
class node_model_surgery toneRose
class node_input,node_analyzer_input toneIndigo
```

## Components

### Dataset generation

- [Dataset generator](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/data/generator.py): `medg/data/generator.py`
- [Data utilities](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/data/utils.py): `medg/data/utils.py`
- [Attribute utilities](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/utils/misc.py): `medg/utils/misc.py`
- [NIfTI writer](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/utils/writer.py): `medg/utils/writer.py`

### MRI augmentations

- [Default settings](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/transforms/defaults.py): `medg/transforms/defaults.py`
- [MRI artifacts](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/transforms/additional/random_spike_n.py): `medg/transforms/additional/random_spike_n.py`
- [K-space sampling](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/transforms/additional/fastMRI/subsample.py): `medg/transforms/additional/fastMRI/subsample.py`
- [K-space transforms](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/transforms/additional/k_space_subsampling.py): `medg/transforms/additional/k_space_subsampling.py`
- [Isotropic downsampling](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/transforms/additional/iso_downsample.py): `medg/transforms/additional/iso_downsample.py`
- [Additional transforms](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/medg/transforms/additional/defaults.py): `medg/transforms/additional/defaults.py`

### Segmentation training

- [FLCE trainer](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/network_training/nnUNetTrainerNoDAV2FLCE.py): `nnunet/training/network_training/nnUNetTrainerNoDAV2FLCE.py`
- [AFA MixUp trainer](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/network_training/nnUNetTrainerNoDAV2FLCE_MixUp_AFA.py): `nnunet/training/network_training/nnUNetTrainerNoDAV2FLCE_MixUp_AFA.py`
- [CutMix trainer](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/network_training/nnUNetTrainerV2_CutMixUpAFA.py): `nnunet/training/network_training/nnUNetTrainerV2_CutMixUpAFA.py`
- [AFA augmentation](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/data_augmentation/afa.py): `nnunet/training/data_augmentation/afa.py`
- [MixUp augmentation](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/data_augmentation/mixup.py): `nnunet/training/data_augmentation/mixup.py`
- [CutMix augmentation](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/data_augmentation/cutmix.py): `nnunet/training/data_augmentation/cutmix.py`
- [Segmentation losses](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/loss_functions/dice_loss.py): `nnunet/training/loss_functions/dice_loss.py`
- [Deep supervision loss](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/loss_functions/deep_supervision.py): `nnunet/training/loss_functions/deep_supervision.py`
- [Cross entropy](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/loss_functions/crossentropy.py): `nnunet/training/loss_functions/crossentropy.py`
- [Focal loss](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/training/loss_functions/focal_loss.py): `nnunet/training/loss_functions/focal_loss.py`
- [Generic U-Net](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/ml/model/segmentation/nnunet/generic_unet.py): `ml/model/segmentation/nnunet/generic_unet.py`
- [Model helper](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/ml/model/segmentation/nnunet/nnunet_helper.py): `ml/model/segmentation/nnunet/nnunet_helper.py`
- [DuBIN normalization](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/network_architecture/custom_modules/dubin.py): `nnunet/network_architecture/custom_modules/dubin.py`
- [Dual batch norm](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/nnunet/network_architecture/custom_modules/dubn.py): `nnunet/network_architecture/custom_modules/dubn.py`

### Model analysis

- [Latent model analysis](https://github.com/miagrouput/augmentations-for-the-unknown/blob/main/model_surgery.py): `model_surgery.py`

### Other components

- **Input scans**
- **Generated dataset**
- **Scans and labels**

## Connections

- Input scans → Dataset generator: provides scans (evidence: `medg/data/generator.py`)
- Dataset generator → Default settings: loads defaults (evidence: `medg/data/generator.py`)
- Dataset generator → Attribute utilities: uses helpers (evidence: `medg/data/generator.py`)
- Dataset generator → NIfTI writer: writes samples (evidence: `medg/data/generator.py`)
- Default settings → MRI artifacts: configures (evidence: `medg/transforms/defaults.py`)
- Default settings → Additional transforms: configures (evidence: `medg/transforms/defaults.py`)
- Additional transforms → Isotropic downsampling: registers (evidence: `medg/transforms/additional/defaults.py`)
- Additional transforms → K-space transforms: registers (evidence: `medg/transforms/additional/defaults.py`)
- K-space transforms → K-space sampling: uses (evidence: `medg/transforms/additional/k_space_subsampling.py`)
- NIfTI writer → Generated dataset: saves NIfTI (evidence: `medg/utils/writer.py`)
- AFA MixUp trainer → FLCE trainer: extends (evidence: `nnunet/training/network_training/nnUNetTrainerNoDAV2FLCE_MixUp_AFA.py`)
- AFA MixUp trainer → AFA augmentation: applies (evidence: `nnunet/training/network_training/nnUNetTrainerNoDAV2FLCE_MixUp_AFA.py`)
- AFA MixUp trainer → MixUp augmentation: applies (evidence: `nnunet/training/network_training/nnUNetTrainerNoDAV2FLCE_MixUp_AFA.py`)
- AFA MixUp trainer → Focal loss: uses (evidence: `nnunet/training/network_training/nnUNetTrainerNoDAV2FLCE_MixUp_AFA.py`)
- CutMix trainer → AFA augmentation: applies (evidence: `nnunet/training/network_training/nnUNetTrainerV2_CutMixUpAFA.py`)
- CutMix trainer → MixUp augmentation: applies (evidence: `nnunet/training/network_training/nnUNetTrainerV2_CutMixUpAFA.py`)
- CutMix trainer → CutMix augmentation: applies (evidence: `nnunet/training/network_training/nnUNetTrainerV2_CutMixUpAFA.py`)
- CutMix trainer → Segmentation losses: uses (evidence: `nnunet/training/network_training/nnUNetTrainerV2_CutMixUpAFA.py`)
- FLCE trainer → Deep supervision loss: uses (evidence: `nnunet/training/network_training/nnUNetTrainerNoDAV2FLCE.py`)
- Model helper → Generic U-Net: constructs (evidence: `ml/model/segmentation/nnunet/nnunet_helper.py`)
- Model helper → DuBIN normalization: converts normalization (evidence: `ml/model/segmentation/nnunet/nnunet_helper.py`)
- DuBIN normalization → Dual batch norm: uses (evidence: `nnunet/network_architecture/custom_modules/dubin.py`)
- Scans and labels → Latent model analysis: analyzes scans (evidence: `model_surgery.py`)
- Latent model analysis → Model helper: loads model (evidence: `model_surgery.py`)
- Latent model analysis → Generic U-Net: runs inference (evidence: `model_surgery.py`)
