# annegnx/pnp-flow architecture

> Architecture diagram of the GitHub repository annegnx/pnp-flow, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-11-19). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/annegnx/pnp-flow
- Repository: https://github.com/annegnx/pnp-flow

## Overview

PnP-Flow trains or loads image-generative priors, applies a selected plug-and-play or flow-based solver to degraded images, and evaluates generated/restored results. The README also documents a separate score-based image-generation training/evaluation stack. Edges follow the sampled calls, source-index references, or README statements; unsampled wiring is omitted rather than inferred.

Main entry points: [`main.py`](https://github.com/annegnx/pnp-flow/blob/main/main.py) (Experiment CLI), [`pnpflow/image_generation/run_lib.py`](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/run_lib.py) (Score training runtime), [`pnpflow/train_flow_matching.py`](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/train_flow_matching.py) (Flow matching), [`pnpflow/compute_metric.py`](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/compute_metric.py) (Image metrics)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_workflow["Restoration workflow"]
  node_cli["Experiment CLI<br/>[main.py]"]
  node_datasets["Image loaders<br/>[dataloaders.py]"]
  node_degradation["Image degradations<br/>[degradations.py]"]
  node_prior["Generative priors<br/>[models.py]"]
end

subgraph group_methods["Inverse solvers"]
  node_pnpflow["PnP-Flow<br/>[pnp_flow.py]"]
  node_otode["OT-ODE<br/>[ot_ode.py]"]
  node_dflow["D-Flow<br/>[d_flow.py]"]
  node_flowpriors["Flow Priors<br/>[flow_priors.py]"]
  node_pnpdiff["PnP-Diff<br/>[pnp_diff.py]"]
  node_pnpgs["PnP-GS<br/>[pnp_gs.py]"]
end

subgraph group_training["Prior training"]
  node_flowtrain["Flow matching"]
  node_denoisertrain["Denoiser training<br/>[train_denoiser.py]"]
end

subgraph group_generation["Score generation"]
  node_scoretrain["Score training runtime<br/>[run_lib.py]"]
  node_scoremodels["Score architectures<br/>[ncsnpp.py]"]
  node_scoreutils["Model registry<br/>[utils.py]"]
  node_sde["Flow and SDEs<br/>[sde_lib.py]"]
  node_sampling["Sampling<br/>[sampling.py]"]
  node_losses["Generative losses<br/>[losses.py]"]
  node_scoredatasets["Generation datasets<br/>[datasets.py]"]
end

subgraph group_evaluation["Evaluation"]
  node_scoreeval["Generation evaluation<br/>[evaluation.py]"]
  node_likelihood["Likelihood estimation<br/>[likelihood.py]"]
  node_metrics["Image metrics<br/>[compute_metric.py]"]
  node_fid["FID scoring<br/>[fid_score.py]"]
end

node_user(("Researcher"))

node_user -->|"configures runs"| node_cli
node_cli -->|"loads images"| node_datasets
node_cli -->|"selects degradation"| node_degradation
node_cli -->|"defines prior"| node_prior
node_cli -->|"dispatches method"| node_pnpflow
node_cli -->|"dispatches method"| node_otode
node_cli -->|"dispatches method"| node_dflow
node_cli -->|"dispatches method"| node_flowpriors
node_cli -->|"dispatches method"| node_pnpdiff
node_cli -->|"dispatches method"| node_pnpgs
node_cli -->|"starts training"| node_flowtrain
node_cli -->|"starts training"| node_denoisertrain
node_cli -->|"computes metrics"| node_metrics
node_pnpgs -->|"uses operators"| node_degradation
node_flowtrain -->|"loads batches"| node_datasets
node_flowtrain -->|"scores samples"| node_fid
node_metrics -->|"loads models"| node_prior
node_metrics -->|"computes FID"| node_fid
node_scoretrain -->|"creates models"| node_scoremodels
node_scoretrain -->|"creates models"| node_scoreutils
node_scoretrain -->|"builds objectives"| node_losses
node_scoretrain -->|"samples images"| node_sampling
node_scoretrain -->|"loads datasets"| node_scoredatasets
node_scoretrain -->|"evaluates samples"| node_scoreeval
node_scoretrain -->|"estimates likelihood"| node_likelihood
node_scoreutils -->|"uses SDE types"| node_sde
node_sampling -->|"uses dynamics"| node_sde
node_losses -->|"uses dynamics"| node_sde
node_likelihood -->|"uses model functions"| node_scoreutils

click node_cli "https://github.com/annegnx/pnp-flow/blob/main/main.py"
click node_datasets "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/dataloaders.py"
click node_degradation "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/degradations.py"
click node_prior "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/models.py"
click node_pnpflow "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/methods/pnp_flow.py"
click node_otode "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/methods/ot_ode.py"
click node_dflow "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/methods/d_flow.py"
click node_flowpriors "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/methods/flow_priors.py"
click node_pnpdiff "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/methods/pnp_diff.py"
click node_pnpgs "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/methods/pnp_gs.py"
click node_flowtrain "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/train_flow_matching.py"
click node_denoisertrain "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/train_denoiser.py"
click node_scoretrain "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/run_lib.py"
click node_scoremodels "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/models/ncsnpp.py"
click node_scoreutils "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/models/utils.py"
click node_sde "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/sde_lib.py"
click node_sampling "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/sampling.py"
click node_losses "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/losses.py"
click node_scoredatasets "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/datasets.py"
click node_scoreeval "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/evaluation.py"
click node_likelihood "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/likelihood.py"
click node_metrics "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/compute_metric.py"
click node_fid "https://github.com/annegnx/pnp-flow/blob/main/pnpflow/fid_score.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_cli,node_datasets,node_degradation,node_prior toneBlue
class node_pnpflow,node_otode,node_dflow,node_flowpriors,node_pnpdiff,node_pnpgs toneAmber
class node_flowtrain,node_denoisertrain toneMint
class node_scoretrain,node_scoremodels,node_scoreutils,node_sde,node_sampling,node_losses,node_scoredatasets toneRose
class node_scoreeval,node_likelihood,node_metrics,node_fid,node_user toneIndigo
```

## Components

### Restoration workflow

- [Experiment CLI](https://github.com/annegnx/pnp-flow/blob/main/main.py): `main.py`
- [Image loaders](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/dataloaders.py): `pnpflow/dataloaders.py`
- [Image degradations](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/degradations.py): `pnpflow/degradations.py`
- [Generative priors](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/models.py): `pnpflow/models.py`

### Inverse solvers

- [PnP-Flow](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/methods/pnp_flow.py): `pnpflow/methods/pnp_flow.py`
- [OT-ODE](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/methods/ot_ode.py): `pnpflow/methods/ot_ode.py`
- [D-Flow](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/methods/d_flow.py): `pnpflow/methods/d_flow.py`
- [Flow Priors](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/methods/flow_priors.py): `pnpflow/methods/flow_priors.py`
- [PnP-Diff](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/methods/pnp_diff.py): `pnpflow/methods/pnp_diff.py`
- [PnP-GS](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/methods/pnp_gs.py): `pnpflow/methods/pnp_gs.py`

### Prior training

- [Flow matching](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/train_flow_matching.py): `pnpflow/train_flow_matching.py`
- [Denoiser training](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/train_denoiser.py): `pnpflow/train_denoiser.py`

### Score generation

- [Score training runtime](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/run_lib.py): `pnpflow/image_generation/run_lib.py`
- [Score architectures](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/models/ncsnpp.py): `pnpflow/image_generation/models/ncsnpp.py`
- [Model registry](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/models/utils.py): `pnpflow/image_generation/models/utils.py`
- [Flow and SDEs](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/sde_lib.py): `pnpflow/image_generation/sde_lib.py`
- [Sampling](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/sampling.py): `pnpflow/image_generation/sampling.py`
- [Generative losses](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/losses.py): `pnpflow/image_generation/losses.py`
- [Generation datasets](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/datasets.py): `pnpflow/image_generation/datasets.py`

### Evaluation

- [Generation evaluation](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/evaluation.py): `pnpflow/image_generation/evaluation.py`
- [Likelihood estimation](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/image_generation/likelihood.py): `pnpflow/image_generation/likelihood.py`
- [Image metrics](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/compute_metric.py): `pnpflow/compute_metric.py`
- [FID scoring](https://github.com/annegnx/pnp-flow/blob/main/pnpflow/fid_score.py): `pnpflow/fid_score.py`

### Other components

- **Researcher**

## Connections

- Researcher → Experiment CLI: configures runs
- Experiment CLI → Image loaders: loads images (evidence: `main.py`)
- Experiment CLI → Image degradations: selects degradation (evidence: `main.py`)
- Experiment CLI → Generative priors: defines prior (evidence: `main.py`)
- Experiment CLI → PnP-Flow: dispatches method (evidence: `main.py`)
- Experiment CLI → OT-ODE: dispatches method (evidence: `main.py`)
- Experiment CLI → D-Flow: dispatches method (evidence: `main.py`)
- Experiment CLI → Flow Priors: dispatches method (evidence: `main.py`)
- Experiment CLI → PnP-Diff: dispatches method (evidence: `main.py`)
- Experiment CLI → PnP-GS: dispatches method (evidence: `main.py`)
- Experiment CLI → Flow matching: starts training (evidence: `main.py`)
- Experiment CLI → Denoiser training: starts training (evidence: `main.py`)
- Experiment CLI → Image metrics: computes metrics (evidence: `main.py`)
- PnP-GS → Image degradations: uses operators (evidence: `pnpflow/methods/pnp_gs.py`)
- Flow matching → Image loaders: loads batches (evidence: `pnpflow/train_flow_matching.py`)
- Flow matching → FID scoring: scores samples (evidence: `pnpflow/train_flow_matching.py`)
- Image metrics → Generative priors: loads models (evidence: `pnpflow/compute_metric.py`)
- Image metrics → FID scoring: computes FID (evidence: `pnpflow/compute_metric.py`)
- Score training runtime → Score architectures: creates models (evidence: `pnpflow/image_generation/run_lib.py`)
- Score training runtime → Model registry: creates models (evidence: `pnpflow/image_generation/run_lib.py`)
- Score training runtime → Generative losses: builds objectives (evidence: `pnpflow/image_generation/run_lib.py`)
- Score training runtime → Sampling: samples images (evidence: `pnpflow/image_generation/run_lib.py`)
- Score training runtime → Generation datasets: loads datasets (evidence: `pnpflow/image_generation/run_lib.py`)
- Score training runtime → Generation evaluation: evaluates samples (evidence: `pnpflow/image_generation/run_lib.py`)
- Score training runtime → Likelihood estimation: estimates likelihood (evidence: `pnpflow/image_generation/run_lib.py`)
- Model registry → Flow and SDEs: uses SDE types (evidence: `pnpflow/image_generation/models/utils.py`)
- Sampling → Flow and SDEs: uses dynamics (evidence: `pnpflow/image_generation/sampling.py`)
- Generative losses → Flow and SDEs: uses dynamics (evidence: `pnpflow/image_generation/losses.py`)
- Likelihood estimation → Model registry: uses model functions (evidence: `pnpflow/image_generation/likelihood.py`)
