# prs-eth/marigold architecture

> Architecture diagram of the GitHub repository prs-eth/marigold, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-10-01). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/prs-eth/marigold
- Repository: https://github.com/prs-eth/marigold

## Overview

Marigold adapts pretrained diffusion models for three dense-image tasks: depth, surface normals, and intrinsic image decomposition (IID). The user-facing workflow is image input → task-specific prediction pipeline → task output; pipelines also support ensembling and uncertainty. Separate training, dataset, evaluation, and preprocessing code supports model development and assessment. The README documents the capabilities, while sampled pipeline and trainer excerpts show some implementation details; several script and dataset relationships are therefore left unconnected where source evidence is unavailable.

Main entry points: [`marigold/marigold_depth_pipeline.py`](https://github.com/prs-eth/marigold/blob/main/marigold/marigold_depth_pipeline.py) (Depth pipeline), [`marigold/marigold_normals_pipeline.py`](https://github.com/prs-eth/marigold/blob/main/marigold/marigold_normals_pipeline.py) (Normals pipeline), [`marigold/marigold_iid_pipeline.py`](https://github.com/prs-eth/marigold/blob/main/marigold/marigold_iid_pipeline.py) (IID pipeline), [`src/trainer/marigold_depth_trainer.py`](https://github.com/prs-eth/marigold/blob/main/src/trainer/marigold_depth_trainer.py) (Depth trainer)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_prediction["Prediction"]
  node_depth_pipeline["Depth pipeline"]
  node_normals_pipeline["Normals pipeline"]
  node_iid_pipeline["IID pipeline"]
end

subgraph group_development["Model development"]
  node_train_scripts["Training entrypoints"]
  node_depth_trainer["Depth trainer"]
  node_lcm_trainer["Depth LCM trainer"]
  node_iid_trainer["IID trainer"]
  node_normals_trainer["Normals trainer"]
end

subgraph group_data["Data and evaluation"]
  node_datasets["Task datasets"]
  node_mixed_sampler["Mixed sampler<br/>[mixed_sampler.py]"]
  node_preprocessing["Dataset preprocessing"]
  node_evaluation["Task evaluation"]
end

subgraph group_shared["Shared components"]
  node_ensemble["Prediction ensemble<br/>[ensemble.py]"]
  node_image_util["Image utilities<br/>[image_util.py]"]
  node_batchsize["Batch sizing<br/>[batchsize.py]"]
  node_training_utils["Training utilities"]
  node_losses["Training losses<br/>[loss.py]"]
  node_metrics["Task metrics<br/>[metric.py]"]
  node_data_loader["Loader utilities<br/>[data_loader.py]"]
end

node_user(("Image analyst"))
node_image["Input image"]
node_depth_result["Depth map"]
node_normals_result["Surface normals"]
node_iid_result["Intrinsic maps"]

node_user -->|"provides"| node_image
node_image -.->|"predicts depth"| node_depth_pipeline
node_image -.->|"predicts normals"| node_normals_pipeline
node_image -.->|"decomposes"| node_iid_pipeline
node_depth_pipeline -->|"returns depth"| node_depth_result
node_normals_pipeline -->|"returns normals"| node_normals_result
node_iid_pipeline -->|"returns components"| node_iid_result
node_depth_pipeline -->|"ensembles predictions"| node_ensemble
node_normals_pipeline -->|"ensembles predictions"| node_ensemble
node_iid_pipeline -->|"ensembles predictions"| node_ensemble
node_depth_pipeline -->|"uses image helpers"| node_image_util
node_normals_pipeline -->|"uses image helpers"| node_image_util
node_iid_pipeline -->|"uses image helpers"| node_image_util
node_depth_pipeline -->|"calculates batches"| node_batchsize
node_normals_pipeline -->|"calculates batches"| node_batchsize
node_iid_pipeline -->|"calculates batches"| node_batchsize
node_train_scripts -->|"starts training"| node_depth_trainer
node_train_scripts -->|"starts training"| node_iid_trainer
node_train_scripts -->|"starts training"| node_normals_trainer
node_train_scripts -->|"uses sampler"| node_mixed_sampler
node_depth_trainer -->|"trains model"| node_depth_pipeline
node_lcm_trainer -->|"extends trainer"| node_depth_trainer
node_depth_trainer -->|"uses losses"| node_losses
node_depth_trainer -->|"uses metrics"| node_metrics
node_depth_trainer -->|"uses loader helpers"| node_data_loader
node_iid_trainer -->|"trains model"| node_iid_pipeline
node_iid_trainer -->|"uses losses"| node_losses
node_iid_trainer -->|"uses metrics"| node_metrics
node_normals_trainer -->|"trains model"| node_normals_pipeline
node_normals_trainer -->|"uses losses"| node_losses
node_normals_trainer -->|"uses metrics"| node_metrics
node_preprocessing -.->|"prepares task data"| node_datasets
node_evaluation -.->|"computes scores"| node_metrics

click node_depth_pipeline "https://github.com/prs-eth/marigold/blob/main/marigold/marigold_depth_pipeline.py"
click node_normals_pipeline "https://github.com/prs-eth/marigold/blob/main/marigold/marigold_normals_pipeline.py"
click node_iid_pipeline "https://github.com/prs-eth/marigold/blob/main/marigold/marigold_iid_pipeline.py"
click node_ensemble "https://github.com/prs-eth/marigold/blob/main/marigold/util/ensemble.py"
click node_image_util "https://github.com/prs-eth/marigold/blob/main/marigold/util/image_util.py"
click node_batchsize "https://github.com/prs-eth/marigold/blob/main/marigold/util/batchsize.py"
click node_train_scripts "https://github.com/prs-eth/marigold/tree/main/script"
click node_depth_trainer "https://github.com/prs-eth/marigold/blob/main/src/trainer/marigold_depth_trainer.py"
click node_lcm_trainer "https://github.com/prs-eth/marigold/blob/main/src/trainer/marigold_depth_LCM_trainer.py"
click node_iid_trainer "https://github.com/prs-eth/marigold/blob/main/src/trainer/marigold_iid_trainer.py"
click node_normals_trainer "https://github.com/prs-eth/marigold/blob/main/src/trainer/marigold_normals_trainer.py"
click node_datasets "https://github.com/prs-eth/marigold/tree/main/src/dataset"
click node_mixed_sampler "https://github.com/prs-eth/marigold/blob/main/src/dataset/mixed_sampler.py"
click node_preprocessing "https://github.com/prs-eth/marigold/tree/main/script"
click node_evaluation "https://github.com/prs-eth/marigold/tree/main/script"
click node_training_utils "https://github.com/prs-eth/marigold/tree/main/src/util"
click node_losses "https://github.com/prs-eth/marigold/blob/main/src/util/loss.py"
click node_metrics "https://github.com/prs-eth/marigold/blob/main/src/util/metric.py"
click node_data_loader "https://github.com/prs-eth/marigold/blob/main/src/util/data_loader.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_depth_pipeline,node_normals_pipeline,node_iid_pipeline toneBlue
class node_train_scripts,node_depth_trainer,node_lcm_trainer,node_iid_trainer,node_normals_trainer toneAmber
class node_datasets,node_mixed_sampler,node_preprocessing,node_evaluation toneMint
class node_ensemble,node_image_util,node_batchsize,node_training_utils,node_losses,node_metrics,node_data_loader toneRose
class node_user,node_image,node_depth_result,node_normals_result,node_iid_result toneIndigo
```

## Components

### Prediction

- [Depth pipeline](https://github.com/prs-eth/marigold/blob/main/marigold/marigold_depth_pipeline.py): `marigold/marigold_depth_pipeline.py`
- [Normals pipeline](https://github.com/prs-eth/marigold/blob/main/marigold/marigold_normals_pipeline.py): `marigold/marigold_normals_pipeline.py`
- [IID pipeline](https://github.com/prs-eth/marigold/blob/main/marigold/marigold_iid_pipeline.py): `marigold/marigold_iid_pipeline.py`

### Model development

- [Training entrypoints](https://github.com/prs-eth/marigold/tree/main/script): `script`
- [Depth trainer](https://github.com/prs-eth/marigold/blob/main/src/trainer/marigold_depth_trainer.py): `src/trainer/marigold_depth_trainer.py`
- [Depth LCM trainer](https://github.com/prs-eth/marigold/blob/main/src/trainer/marigold_depth_LCM_trainer.py): `src/trainer/marigold_depth_LCM_trainer.py`
- [IID trainer](https://github.com/prs-eth/marigold/blob/main/src/trainer/marigold_iid_trainer.py): `src/trainer/marigold_iid_trainer.py`
- [Normals trainer](https://github.com/prs-eth/marigold/blob/main/src/trainer/marigold_normals_trainer.py): `src/trainer/marigold_normals_trainer.py`

### Data and evaluation

- [Task datasets](https://github.com/prs-eth/marigold/tree/main/src/dataset): `src/dataset`
- [Mixed sampler](https://github.com/prs-eth/marigold/blob/main/src/dataset/mixed_sampler.py): `src/dataset/mixed_sampler.py`
- [Dataset preprocessing](https://github.com/prs-eth/marigold/tree/main/script): `script`
- [Task evaluation](https://github.com/prs-eth/marigold/tree/main/script): `script`

### Shared components

- [Prediction ensemble](https://github.com/prs-eth/marigold/blob/main/marigold/util/ensemble.py): `marigold/util/ensemble.py`
- [Image utilities](https://github.com/prs-eth/marigold/blob/main/marigold/util/image_util.py): `marigold/util/image_util.py`
- [Batch sizing](https://github.com/prs-eth/marigold/blob/main/marigold/util/batchsize.py): `marigold/util/batchsize.py`
- [Training utilities](https://github.com/prs-eth/marigold/tree/main/src/util): `src/util`
- [Training losses](https://github.com/prs-eth/marigold/blob/main/src/util/loss.py): `src/util/loss.py`
- [Task metrics](https://github.com/prs-eth/marigold/blob/main/src/util/metric.py): `src/util/metric.py`
- [Loader utilities](https://github.com/prs-eth/marigold/blob/main/src/util/data_loader.py): `src/util/data_loader.py`

### Other components

- **Image analyst**
- **Input image**
- **Depth map**
- **Surface normals**
- **Intrinsic maps**

## Connections

- Image analyst → Input image: provides (evidence: `README.md`)
- Input image → Depth pipeline: predicts depth
- Input image → Normals pipeline: predicts normals
- Input image → IID pipeline: decomposes
- Depth pipeline → Depth map: returns depth (evidence: `marigold/marigold_depth_pipeline.py`)
- Normals pipeline → Surface normals: returns normals (evidence: `marigold/marigold_normals_pipeline.py`)
- IID pipeline → Intrinsic maps: returns components (evidence: `marigold/marigold_iid_pipeline.py`)
- Depth pipeline → Prediction ensemble: ensembles predictions (evidence: `marigold/marigold_depth_pipeline.py`)
- Normals pipeline → Prediction ensemble: ensembles predictions (evidence: `marigold/marigold_normals_pipeline.py`)
- IID pipeline → Prediction ensemble: ensembles predictions (evidence: `marigold/marigold_iid_pipeline.py`)
- Depth pipeline → Image utilities: uses image helpers (evidence: `marigold/marigold_depth_pipeline.py`)
- Normals pipeline → Image utilities: uses image helpers (evidence: `marigold/marigold_normals_pipeline.py`)
- IID pipeline → Image utilities: uses image helpers (evidence: `marigold/marigold_iid_pipeline.py`)
- Depth pipeline → Batch sizing: calculates batches (evidence: `marigold/marigold_depth_pipeline.py`)
- Normals pipeline → Batch sizing: calculates batches (evidence: `marigold/marigold_normals_pipeline.py`)
- IID pipeline → Batch sizing: calculates batches (evidence: `marigold/marigold_iid_pipeline.py`)
- Training entrypoints → Depth trainer: starts training
- Training entrypoints → IID trainer: starts training
- Training entrypoints → Normals trainer: starts training
- Training entrypoints → Mixed sampler: uses sampler (evidence: `script/depth/train.py`)
- Depth trainer → Depth pipeline: trains model (evidence: `src/trainer/marigold_depth_trainer.py`)
- Depth LCM trainer → Depth trainer: extends trainer (evidence: `src/trainer/marigold_depth_LCM_trainer.py`)
- Depth trainer → Training losses: uses losses (evidence: `src/trainer/marigold_depth_trainer.py`)
- Depth trainer → Task metrics: uses metrics (evidence: `src/trainer/marigold_depth_trainer.py`)
- Depth trainer → Loader utilities: uses loader helpers (evidence: `src/trainer/marigold_depth_trainer.py`)
- IID trainer → IID pipeline: trains model (evidence: `src/trainer/marigold_iid_trainer.py`)
- IID trainer → Training losses: uses losses (evidence: `src/trainer/marigold_iid_trainer.py`)
- IID trainer → Task metrics: uses metrics (evidence: `src/trainer/marigold_iid_trainer.py`)
- Normals trainer → Normals pipeline: trains model (evidence: `src/trainer/marigold_normals_trainer.py`)
- Normals trainer → Training losses: uses losses (evidence: `src/trainer/marigold_normals_trainer.py`)
- Normals trainer → Task metrics: uses metrics (evidence: `src/trainer/marigold_normals_trainer.py`)
- Dataset preprocessing → Task datasets: prepares task data (evidence: `script/depth/train.py`)
- Task evaluation → Task metrics: computes scores
