# algolzw/daclip-uir architecture

> Architecture diagram of the GitHub repository algolzw/daclip-uir, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-11-10). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/algolzw/daclip-uir
- Repository: https://github.com/algolzw/daclip-uir

## Overview

DA-CLIP is a controlled vision-language image-restoration system, with both the original DA-CLIP-SDE configuration and the later wild-IR variant. The user workflow takes a degraded image, extracts image and degradation context with DA-CLIP, and conditions an SDE-based restoration model to produce an output image. The repository also includes CLIP training/modeling, restoration training, and degradation/data utilities. Edges reflect the README and shown call sites; unsampled subsystems are included where documented, without assuming their internal wiring.

Main entry points: [`universal-image-restoration/config/wild-ir/models/denoising_model.py`](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/models/denoising_model.py) (Denoising Model), [`universal-image-restoration/config/wild-ir/inference.py`](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/inference.py) (Wild-IR Inference), [`universal-image-restoration/config/wild-ir/app.py`](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/app.py) (Wild-IR Demo), [`universal-image-restoration/config/daclip-sde/app.py`](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/daclip-sde/app.py) (DA-CLIP-SDE Demo)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_inference["Restoration Inference"]
  node_wild_app["Wild-IR Demo<br/>[app.py]"]
  node_sde_app["DA-CLIP-SDE Demo<br/>[app.py]"]
  node_wild_infer["Wild-IR Inference<br/>[inference.py]"]
end

subgraph group_clip["Vision Language Model"]
  node_clip_factory["CLIP Factory<br/>[factory.py]"]
  node_clip_model["CLIP Encoders<br/>[model.py]"]
  node_clip_transformer["Transformer Layers<br/>[transformer.py]"]
  node_clip_pretrained["Pretrained Weights<br/>[pretrained.py]"]
  node_clip_coca["CoCa Model<br/>[coca_model.py]"]
  node_clip_metadata["Zero-Shot Metadata"]
end

subgraph group_restore["Restoration Models"]
  node_restore_model["Denoising Model<br/>[denoising_model.py]"]
  node_restore_network["Restoration Network<br/>[networks.py]"]
  node_restore_sde["SDE Sampling<br/>[sde_utils.py]"]
  node_restore_loss["Matching Loss<br/>[loss.py]"]
  node_restore_optimizer["Optimizer Setup<br/>[optimizer.py]"]
  node_restore_scheduler["Learning-Rate Schedule<br/>[lr_scheduler.py]"]
  node_restore_options["Restoration Options<br/>[options.py]"]
end

subgraph group_training["Training and Data"]
  node_datasets["Paired Image Datasets<br/>[LQGT_dataset.py]"]
  node_degradation["Degradation Utilities<br/>[deg_util.py]"]
  node_clip_train_data["CLIP Training Data<br/>[data.py]"]
  node_clip_train["CLIP Training Loop<br/>[train.py]"]
  node_clip_train_main["CLIP Training Entry<br/>[main.py]"]
  node_clip_train_dist["Distributed Training<br/>[distributed.py]"]
  node_clip_train_params["Training Parameters<br/>[params.py]"]
end

node_user(("Image User"))

node_user -->|"submits image"| node_wild_app
node_wild_app -->|"loads DA-CLIP"| node_clip_factory
node_wild_app -->|"encodes image"| node_clip_model
node_wild_app -->|"adds noise"| node_restore_sde
node_wild_app -->|"feeds and tests"| node_restore_model
node_sde_app -->|"loads DA-CLIP"| node_clip_factory
node_sde_app -->|"encodes image"| node_clip_model
node_sde_app -->|"adds noise"| node_restore_sde
node_sde_app -->|"feeds and tests"| node_restore_model
node_wild_infer -->|"loads DA-CLIP"| node_clip_factory
node_wild_infer -->|"encodes image"| node_clip_model
node_wild_infer -->|"adds noise"| node_restore_sde
node_wild_infer -->|"feeds and tests"| node_restore_model
node_wild_infer -->|"loads test data"| node_datasets
node_clip_model -->|"uses layers"| node_clip_transformer
node_restore_model -->|"builds network"| node_restore_network
node_restore_model -->|"uses loss"| node_restore_loss
node_restore_model -->|"sets optimizer"| node_restore_optimizer
node_restore_model -->|"sets schedule"| node_restore_scheduler
node_datasets -->|"uses utilities"| node_degradation
node_clip_train_main -->|"imports data"| node_clip_train_data
node_clip_train_main -->|"imports training loop"| node_clip_train
node_clip_train_main -->|"imports distributed setup"| node_clip_train_dist
node_clip_train_main -->|"imports parameters"| node_clip_train_params
node_wild_app -->|"configures model"| node_restore_model
node_sde_app -->|"configures model"| node_restore_model
node_wild_infer -->|"parses options"| node_restore_options

click node_wild_app "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/app.py"
click node_sde_app "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/daclip-sde/app.py"
click node_wild_infer "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/inference.py"
click node_clip_factory "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/open_clip/factory.py"
click node_clip_model "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/open_clip/model.py"
click node_clip_transformer "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/open_clip/transformer.py"
click node_clip_pretrained "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/open_clip/pretrained.py"
click node_clip_coca "https://github.com/algolzw/daclip-uir/blob/main/da-clip/src/open_clip/coca_model.py"
click node_clip_metadata "https://github.com/algolzw/daclip-uir/blob/main/da-clip/src/open_clip/zero_shot_metadata.py"
click node_restore_model "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/models/denoising_model.py"
click node_restore_network "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/models/networks.py"
click node_restore_sde "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/utils/sde_utils.py"
click node_restore_loss "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/models/modules/loss.py"
click node_restore_optimizer "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/models/optimizer.py"
click node_restore_scheduler "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/models/lr_scheduler.py"
click node_restore_options "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/options.py"
click node_datasets "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/data/LQGT_dataset.py"
click node_degradation "https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/data/deg_util.py"
click node_clip_train_data "https://github.com/algolzw/daclip-uir/blob/main/da-clip/src/training/data.py"
click node_clip_train "https://github.com/algolzw/daclip-uir/blob/main/da-clip/src/training/train.py"
click node_clip_train_main "https://github.com/algolzw/daclip-uir/blob/main/da-clip/src/training/main.py"
click node_clip_train_dist "https://github.com/algolzw/daclip-uir/blob/main/da-clip/src/training/distributed.py"
click node_clip_train_params "https://github.com/algolzw/daclip-uir/blob/main/da-clip/src/training/params.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_wild_app,node_sde_app,node_wild_infer,node_user toneBlue
class node_clip_factory,node_clip_model,node_clip_transformer,node_clip_pretrained,node_clip_coca,node_clip_metadata toneAmber
class node_restore_model,node_restore_network,node_restore_sde,node_restore_loss,node_restore_optimizer,node_restore_scheduler,node_restore_options toneMint
class node_datasets,node_degradation,node_clip_train_data,node_clip_train,node_clip_train_main,node_clip_train_dist,node_clip_train_params toneRose
```

## Components

### Restoration Inference

- [Wild-IR Demo](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/app.py): `universal-image-restoration/config/wild-ir/app.py`
- [DA-CLIP-SDE Demo](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/daclip-sde/app.py): `universal-image-restoration/config/daclip-sde/app.py`
- [Wild-IR Inference](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/inference.py): `universal-image-restoration/config/wild-ir/inference.py`

### Vision Language Model

- [CLIP Factory](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/open_clip/factory.py): `universal-image-restoration/open_clip/factory.py`
- [CLIP Encoders](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/open_clip/model.py): `universal-image-restoration/open_clip/model.py`
- [Transformer Layers](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/open_clip/transformer.py): `universal-image-restoration/open_clip/transformer.py`
- [Pretrained Weights](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/open_clip/pretrained.py): `universal-image-restoration/open_clip/pretrained.py`
- [CoCa Model](https://github.com/algolzw/daclip-uir/blob/main/da-clip/src/open_clip/coca_model.py): `da-clip/src/open_clip/coca_model.py`
- [Zero-Shot Metadata](https://github.com/algolzw/daclip-uir/blob/main/da-clip/src/open_clip/zero_shot_metadata.py): `da-clip/src/open_clip/zero_shot_metadata.py`

### Restoration Models

- [Denoising Model](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/models/denoising_model.py): `universal-image-restoration/config/wild-ir/models/denoising_model.py`
- [Restoration Network](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/models/networks.py): `universal-image-restoration/config/wild-ir/models/networks.py`
- [SDE Sampling](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/utils/sde_utils.py): `universal-image-restoration/utils/sde_utils.py`
- [Matching Loss](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/models/modules/loss.py): `universal-image-restoration/config/wild-ir/models/modules/loss.py`
- [Optimizer Setup](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/models/optimizer.py): `universal-image-restoration/config/wild-ir/models/optimizer.py`
- [Learning-Rate Schedule](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/models/lr_scheduler.py): `universal-image-restoration/config/wild-ir/models/lr_scheduler.py`
- [Restoration Options](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/config/wild-ir/options.py): `universal-image-restoration/config/wild-ir/options.py`

### Training and Data

- [Paired Image Datasets](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/data/LQGT_dataset.py): `universal-image-restoration/data/LQGT_dataset.py`
- [Degradation Utilities](https://github.com/algolzw/daclip-uir/blob/main/universal-image-restoration/data/deg_util.py): `universal-image-restoration/data/deg_util.py`
- [CLIP Training Data](https://github.com/algolzw/daclip-uir/blob/main/da-clip/src/training/data.py): `da-clip/src/training/data.py`
- [CLIP Training Loop](https://github.com/algolzw/daclip-uir/blob/main/da-clip/src/training/train.py): `da-clip/src/training/train.py`
- [CLIP Training Entry](https://github.com/algolzw/daclip-uir/blob/main/da-clip/src/training/main.py): `da-clip/src/training/main.py`
- [Distributed Training](https://github.com/algolzw/daclip-uir/blob/main/da-clip/src/training/distributed.py): `da-clip/src/training/distributed.py`
- [Training Parameters](https://github.com/algolzw/daclip-uir/blob/main/da-clip/src/training/params.py): `da-clip/src/training/params.py`

### Other components

- **Image User**

## Connections

- Image User → Wild-IR Demo: submits image (evidence: `universal-image-restoration/config/wild-ir/app.py`)
- Wild-IR Demo → CLIP Factory: loads DA-CLIP (evidence: `universal-image-restoration/config/wild-ir/app.py`)
- Wild-IR Demo → CLIP Encoders: encodes image (evidence: `universal-image-restoration/config/wild-ir/app.py`)
- Wild-IR Demo → SDE Sampling: adds noise (evidence: `universal-image-restoration/config/wild-ir/app.py`)
- Wild-IR Demo → Denoising Model: feeds and tests (evidence: `universal-image-restoration/config/wild-ir/app.py`)
- DA-CLIP-SDE Demo → CLIP Factory: loads DA-CLIP (evidence: `universal-image-restoration/config/daclip-sde/app.py`)
- DA-CLIP-SDE Demo → CLIP Encoders: encodes image (evidence: `universal-image-restoration/config/daclip-sde/app.py`)
- DA-CLIP-SDE Demo → SDE Sampling: adds noise (evidence: `universal-image-restoration/config/daclip-sde/app.py`)
- DA-CLIP-SDE Demo → Denoising Model: feeds and tests (evidence: `universal-image-restoration/config/daclip-sde/app.py`)
- Wild-IR Inference → CLIP Factory: loads DA-CLIP (evidence: `universal-image-restoration/config/wild-ir/inference.py`)
- Wild-IR Inference → CLIP Encoders: encodes image (evidence: `universal-image-restoration/config/wild-ir/inference.py`)
- Wild-IR Inference → SDE Sampling: adds noise (evidence: `universal-image-restoration/config/wild-ir/inference.py`)
- Wild-IR Inference → Denoising Model: feeds and tests (evidence: `universal-image-restoration/config/wild-ir/inference.py`)
- Wild-IR Inference → Paired Image Datasets: loads test data (evidence: `universal-image-restoration/config/wild-ir/inference.py`)
- CLIP Encoders → Transformer Layers: uses layers (evidence: `universal-image-restoration/open_clip/model.py`)
- Denoising Model → Restoration Network: builds network (evidence: `universal-image-restoration/config/wild-ir/models/denoising_model.py`)
- Denoising Model → Matching Loss: uses loss (evidence: `universal-image-restoration/config/wild-ir/models/denoising_model.py`)
- Denoising Model → Optimizer Setup: sets optimizer (evidence: `universal-image-restoration/config/wild-ir/models/denoising_model.py`)
- Denoising Model → Learning-Rate Schedule: sets schedule (evidence: `universal-image-restoration/config/wild-ir/models/denoising_model.py`)
- Paired Image Datasets → Degradation Utilities: uses utilities (evidence: `universal-image-restoration/data/LQGT_dataset.py`)
- CLIP Training Entry → CLIP Training Data: imports data (evidence: `da-clip/src/training/main.py`)
- CLIP Training Entry → CLIP Training Loop: imports training loop (evidence: `da-clip/src/training/main.py`)
- CLIP Training Entry → Distributed Training: imports distributed setup (evidence: `da-clip/src/training/main.py`)
- CLIP Training Entry → Training Parameters: imports parameters (evidence: `da-clip/src/training/main.py`)
- Wild-IR Demo → Denoising Model: configures model (evidence: `universal-image-restoration/config/wild-ir/app.py`)
- DA-CLIP-SDE Demo → Denoising Model: configures model (evidence: `universal-image-restoration/config/daclip-sde/app.py`)
- Wild-IR Inference → Restoration Options: parses options (evidence: `universal-image-restoration/config/wild-ir/inference.py`)
