# idata-ora/hiadapter architecture

> Architecture diagram of the GitHub repository idata-ora/hiadapter, 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/idata-ora/hiadapter
- Repository: https://github.com/idata-ora/hiadapter

## Overview

HiAdapter is a pathology-image classification research system for adapting pathology foundation models to downstream datasets. Its principal workflow prepares patch datasets, runs a UNI-based model with stain- and morphology-oriented adaptation, trains with classification and optional prototypical contrastive loss, and evaluates predictions. The README also documents five-fold data splitting and broader benchmark, unseen-stain/cancer, interpretability, and validation capabilities. Adapter wiring is only partly visible in the excerpts, so unsampled implementation details are not expanded.

Main entry points: [`core/function.py`](https://github.com/idata-ora/hiadapter/blob/main/core/function.py) (Training loop), [`train.py`](https://github.com/idata-ora/hiadapter/blob/main/train.py) (Training entry), [`model/uni/vision_transformer.py`](https://github.com/idata-ora/hiadapter/blob/main/model/uni/vision_transformer.py) (UNI transformer), [`myDatasets/myTileDataset.py`](https://github.com/idata-ora/hiadapter/blob/main/myDatasets/myTileDataset.py) (Patch dataset)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_data["Data preparation"]
  node_patch_data["Patch dataset<br/>[myTileDataset.py]"]
  node_fold_splits["Fold splitting<br/>[split_5fold.py]"]
  node_prompt_data["Prompt generation<br/>[get_prompt.py]"]
end

subgraph group_model["Adapted model"]
  node_uni["UNI transformer"]
  node_vit_layers["Transformer layers"]
  node_stain_adapter["Stain adaptation<br/>[cModule.py]"]
  node_morph_adapter["Morphology fusion<br/>[cfusion.py]"]
end

subgraph group_training["Training and evaluation"]
  node_train_entry["Training entry<br/>[train.py]"]
  node_train_loop["Training loop<br/>[function.py]"]
  node_classification_loss["Classification losses<br/>[losses.py]"]
  node_ppcl["PPCL objective<br/>[PPCL.py]"]
end

subgraph group_support["Training support"]
  node_metrics["Metrics and reporting<br/>[tools.py]"]
  node_runtime_utils["Runtime utilities<br/>[utils.py]"]
  node_parameters["Run parameters<br/>[params.py]"]
  node_optimizer_config["Optimizer configuration<br/>[optConfig.py]"]
  node_scheduler_config["Learning-rate schedule"]
end

node_researcher(("Researcher"))

node_researcher -->|"starts training"| node_train_entry
node_researcher -->|"prepares folds"| node_fold_splits
node_fold_splits -.->|"provides splits"| node_patch_data
node_train_entry -->|"loads options"| node_parameters
node_train_entry -->|"builds datasets"| node_patch_data
node_train_entry -->|"selects model"| node_uni
node_train_entry -.->|"runs training"| node_train_loop
node_patch_data -->|"uses utilities"| node_runtime_utils
node_patch_data -.->|"uses prompts"| node_prompt_data
node_uni -->|"uses layers"| node_vit_layers
node_uni -->|"uses module"| node_stain_adapter
node_uni -->|"uses fusion"| node_morph_adapter
node_train_loop -->|"computes loss"| node_classification_loss
node_train_loop -->|"updates proxies"| node_ppcl
node_train_loop -->|"uses metrics"| node_metrics
node_train_loop -.->|"uses utilities"| node_runtime_utils
node_train_loop -.->|"configures optimizer"| node_optimizer_config
node_train_loop -.->|"configures schedule"| node_scheduler_config
node_train_entry -->|"uses utilities"| node_runtime_utils

click node_patch_data "https://github.com/idata-ora/hiadapter/blob/main/myDatasets/myTileDataset.py"
click node_fold_splits "https://github.com/idata-ora/hiadapter/blob/main/myDatasets/split_5fold.py"
click node_prompt_data "https://github.com/idata-ora/hiadapter/blob/main/myDatasets/get_prompt.py"
click node_uni "https://github.com/idata-ora/hiadapter/blob/main/model/uni/vision_transformer.py"
click node_vit_layers "https://github.com/idata-ora/hiadapter/tree/main/model/uni/layers"
click node_stain_adapter "https://github.com/idata-ora/hiadapter/blob/main/utils/cModule.py"
click node_morph_adapter "https://github.com/idata-ora/hiadapter/blob/main/utils/cfusion.py"
click node_train_entry "https://github.com/idata-ora/hiadapter/blob/main/train.py"
click node_train_loop "https://github.com/idata-ora/hiadapter/blob/main/core/function.py"
click node_classification_loss "https://github.com/idata-ora/hiadapter/blob/main/core/losses.py"
click node_ppcl "https://github.com/idata-ora/hiadapter/blob/main/core/PPCL.py"
click node_metrics "https://github.com/idata-ora/hiadapter/blob/main/utils/tools.py"
click node_runtime_utils "https://github.com/idata-ora/hiadapter/blob/main/utils/utils.py"
click node_parameters "https://github.com/idata-ora/hiadapter/blob/main/utils/params.py"
click node_optimizer_config "https://github.com/idata-ora/hiadapter/blob/main/utils/optConfig.py"
click node_scheduler_config "https://github.com/idata-ora/hiadapter/blob/main/utils/lrSchedulearConfig.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_patch_data,node_fold_splits,node_prompt_data toneBlue
class node_uni,node_vit_layers,node_stain_adapter,node_morph_adapter toneAmber
class node_train_entry,node_train_loop,node_classification_loss,node_ppcl toneMint
class node_metrics,node_runtime_utils,node_parameters,node_optimizer_config,node_scheduler_config toneRose
class node_researcher toneIndigo
```

## Components

### Data preparation

- [Patch dataset](https://github.com/idata-ora/hiadapter/blob/main/myDatasets/myTileDataset.py): `myDatasets/myTileDataset.py`
- [Fold splitting](https://github.com/idata-ora/hiadapter/blob/main/myDatasets/split_5fold.py): `myDatasets/split_5fold.py`
- [Prompt generation](https://github.com/idata-ora/hiadapter/blob/main/myDatasets/get_prompt.py): `myDatasets/get_prompt.py`

### Adapted model

- [UNI transformer](https://github.com/idata-ora/hiadapter/blob/main/model/uni/vision_transformer.py): `model/uni/vision_transformer.py`
- [Transformer layers](https://github.com/idata-ora/hiadapter/tree/main/model/uni/layers): `model/uni/layers`
- [Stain adaptation](https://github.com/idata-ora/hiadapter/blob/main/utils/cModule.py): `utils/cModule.py`
- [Morphology fusion](https://github.com/idata-ora/hiadapter/blob/main/utils/cfusion.py): `utils/cfusion.py`

### Training and evaluation

- [Training entry](https://github.com/idata-ora/hiadapter/blob/main/train.py): `train.py`
- [Training loop](https://github.com/idata-ora/hiadapter/blob/main/core/function.py): `core/function.py`
- [Classification losses](https://github.com/idata-ora/hiadapter/blob/main/core/losses.py): `core/losses.py`
- [PPCL objective](https://github.com/idata-ora/hiadapter/blob/main/core/PPCL.py): `core/PPCL.py`

### Training support

- [Metrics and reporting](https://github.com/idata-ora/hiadapter/blob/main/utils/tools.py): `utils/tools.py`
- [Runtime utilities](https://github.com/idata-ora/hiadapter/blob/main/utils/utils.py): `utils/utils.py`
- [Run parameters](https://github.com/idata-ora/hiadapter/blob/main/utils/params.py): `utils/params.py`
- [Optimizer configuration](https://github.com/idata-ora/hiadapter/blob/main/utils/optConfig.py): `utils/optConfig.py`
- [Learning-rate schedule](https://github.com/idata-ora/hiadapter/blob/main/utils/lrSchedulearConfig.py): `utils/lrSchedulearConfig.py`

### Other components

- **Researcher**

## Connections

- Researcher → Training entry: starts training (evidence: `README.md`)
- Researcher → Fold splitting: prepares folds (evidence: `README.md`)
- Fold splitting → Patch dataset: provides splits
- Training entry → Run parameters: loads options (evidence: `train.py`)
- Training entry → Patch dataset: builds datasets (evidence: `train.py`)
- Training entry → UNI transformer: selects model (evidence: `train.py`)
- Training entry → Training loop: runs training (evidence: `train.py`)
- Patch dataset → Runtime utilities: uses utilities (evidence: `myDatasets/myTileDataset.py`)
- Patch dataset → Prompt generation: uses prompts
- UNI transformer → Transformer layers: uses layers (evidence: `model/uni/vision_transformer.py`)
- UNI transformer → Stain adaptation: uses module (evidence: `model/uni/vision_transformer.py`)
- UNI transformer → Morphology fusion: uses fusion (evidence: `model/uni/vision_transformer.py`)
- Training loop → Classification losses: computes loss (evidence: `core/function.py`)
- Training loop → PPCL objective: updates proxies (evidence: `core/function.py`)
- Training loop → Metrics and reporting: uses metrics (evidence: `core/function.py`)
- Training loop → Runtime utilities: uses utilities
- Training loop → Optimizer configuration: configures optimizer
- Training loop → Learning-rate schedule: configures schedule
- Training entry → Runtime utilities: uses utilities (evidence: `train.py`)
