# jiaxi-jiang/fbcnn architecture

> Architecture diagram of the GitHub repository jiaxi-jiang/fbcnn, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-11-21). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/jiaxi-jiang/fbcnn
- Repository: https://github.com/jiaxi-jiang/fbcnn

## Overview

FBCNN is a PyTorch system for blind JPEG-artifact removal, with a quality-factor estimate that lets restoration strength be adjusted. The main workflow trains from clean images using synthetic single- or double-JPEG degradation, then runs color, grayscale, double-JPEG, or real-world image restoration. The graph includes the documented use cases and the sampled training/model layers; unsampled test-script details and internal network wiring are not expanded beyond documented capabilities and indexed references.

Main entry points: [`models/model_fbcnn.py`](https://github.com/jiaxi-jiang/fbcnn/blob/main/models/model_fbcnn.py) (Training model), [`main_train_fbcnn.py`](https://github.com/jiaxi-jiang/fbcnn/blob/main/main_train_fbcnn.py) (Training runner), [`models/network_fbcnn.py`](https://github.com/jiaxi-jiang/fbcnn/blob/main/models/network_fbcnn.py) (FBCNN network), [`utils/utils_image.py`](https://github.com/jiaxi-jiang/fbcnn/blob/main/utils/utils_image.py) (Image processing)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_workflow["Training and inference"]
  node_trainer["Training runner"]
  node_train_options["Training options"]
  node_test_color["Color testing"]
  node_test_gray["Grayscale testing"]
  node_test_double["Double-JPEG testing"]
  node_test_real["Real-image testing"]
end

subgraph group_data["Degradation data"]
  node_dataset_select["Dataset selector<br/>[select_dataset.py]"]
  node_dataset_color["Color JPEG dataset<br/>[dataset_jpeg.py]"]
  node_dataset_gray["Grayscale JPEG dataset"]
  node_dataset_double["Double-JPEG dataset"]
end

subgraph group_model["FBCNN model"]
  node_model_select["Model selector<br/>[select_model.py]"]
  node_model_fbcnn["Training model<br/>[model_fbcnn.py]"]
  node_model_base["Model base<br/>[model_base.py]"]
  node_network_select["Network selector<br/>[select_network.py]"]
  node_network["FBCNN network<br/>[network_fbcnn.py]"]
  node_loss_ssim["SSIM loss<br/>[loss_ssim.py]"]
  node_loss["Other losses<br/>[loss.py]"]
end

subgraph group_shared["Shared utilities"]
  node_image_utils["Image processing<br/>[utils_image.py]"]
  node_model_utils["Inference helpers<br/>[utils_model.py]"]
  node_option_utils["Option parsing<br/>[utils_option.py]"]
  node_logger_utils["Logging<br/>[utils_logger.py]"]
  node_bnorm_utils["Batch norm helpers<br/>[utils_bnorm.py]"]
  node_regularizers["Regularizers"]
end

node_image_input(("JPEG images"))

node_trainer -->|"parses options"| node_option_utils
node_trainer -->|"uses configuration"| node_train_options
node_trainer -->|"creates datasets"| node_dataset_select
node_trainer -->|"creates model"| node_model_select
node_trainer -->|"logs progress"| node_logger_utils
node_dataset_select -->|"selects JPEG"| node_dataset_color
node_dataset_select -->|"selects grayscale"| node_dataset_gray
node_dataset_select -->|"selects double-JPEG"| node_dataset_double
node_dataset_color -->|"loads images"| node_image_utils
node_dataset_gray -->|"loads images"| node_image_utils
node_dataset_double -->|"loads images"| node_image_utils
node_model_select -->|"creates model"| node_model_fbcnn
node_model_fbcnn -->|"inherits"| node_model_base
node_model_fbcnn -->|"builds network"| node_network_select
node_network_select -->|"constructs FBCNN"| node_network
node_model_fbcnn -->|"uses SSIM loss"| node_loss_ssim
node_model_fbcnn -->|"uses regularizers"| node_regularizers
node_model_base -->|"uses helpers"| node_bnorm_utils
node_model_fbcnn -->|"uses model helpers"| node_model_utils
node_model_utils -->|"uses image helpers"| node_image_utils
node_test_color -.->|"reads images"| node_image_input
node_test_gray -.->|"reads images"| node_image_input
node_test_double -->|"reads images"| node_image_input
node_test_real -.->|"reads images"| node_image_input
node_test_color -->|"runs restoration"| node_network
node_test_gray -->|"runs restoration"| node_network
node_test_double -->|"runs restoration"| node_network
node_test_real -->|"runs restoration"| node_network
node_test_double -->|"converts and scores"| node_image_utils
node_test_double -->|"logs results"| node_logger_utils

click node_trainer "https://github.com/jiaxi-jiang/fbcnn/blob/main/main_train_fbcnn.py"
click node_train_options "https://github.com/jiaxi-jiang/fbcnn/tree/main/options"
click node_test_color "https://github.com/jiaxi-jiang/fbcnn/blob/main/main_test_fbcnn_color.py"
click node_test_gray "https://github.com/jiaxi-jiang/fbcnn/blob/main/main_test_fbcnn_gray.py"
click node_test_double "https://github.com/jiaxi-jiang/fbcnn/blob/main/main_test_fbcnn_gray_doublejpeg.py"
click node_test_real "https://github.com/jiaxi-jiang/fbcnn/blob/main/main_test_fbcnn_color_real.py"
click node_dataset_select "https://github.com/jiaxi-jiang/fbcnn/blob/main/data/select_dataset.py"
click node_dataset_color "https://github.com/jiaxi-jiang/fbcnn/blob/main/data/dataset_jpeg.py"
click node_dataset_gray "https://github.com/jiaxi-jiang/fbcnn/blob/main/data/dataset_jpeggray.py"
click node_dataset_double "https://github.com/jiaxi-jiang/fbcnn/blob/main/data/dataset_jpeggraydouble.py"
click node_model_select "https://github.com/jiaxi-jiang/fbcnn/blob/main/models/select_model.py"
click node_model_fbcnn "https://github.com/jiaxi-jiang/fbcnn/blob/main/models/model_fbcnn.py"
click node_model_base "https://github.com/jiaxi-jiang/fbcnn/blob/main/models/model_base.py"
click node_network_select "https://github.com/jiaxi-jiang/fbcnn/blob/main/models/select_network.py"
click node_network "https://github.com/jiaxi-jiang/fbcnn/blob/main/models/network_fbcnn.py"
click node_loss_ssim "https://github.com/jiaxi-jiang/fbcnn/blob/main/models/loss_ssim.py"
click node_loss "https://github.com/jiaxi-jiang/fbcnn/blob/main/models/loss.py"
click node_image_utils "https://github.com/jiaxi-jiang/fbcnn/blob/main/utils/utils_image.py"
click node_model_utils "https://github.com/jiaxi-jiang/fbcnn/blob/main/utils/utils_model.py"
click node_option_utils "https://github.com/jiaxi-jiang/fbcnn/blob/main/utils/utils_option.py"
click node_logger_utils "https://github.com/jiaxi-jiang/fbcnn/blob/main/utils/utils_logger.py"
click node_bnorm_utils "https://github.com/jiaxi-jiang/fbcnn/blob/main/utils/utils_bnorm.py"
click node_regularizers "https://github.com/jiaxi-jiang/fbcnn/blob/main/utils/utils_regularizers.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_trainer,node_train_options,node_test_color,node_test_gray,node_test_double,node_test_real toneBlue
class node_dataset_select,node_dataset_color,node_dataset_gray,node_dataset_double toneAmber
class node_model_select,node_model_fbcnn,node_model_base,node_network_select,node_network,node_loss_ssim,node_loss toneMint
class node_image_utils,node_model_utils,node_option_utils,node_logger_utils,node_bnorm_utils,node_regularizers toneRose
class node_image_input toneIndigo
```

## Components

### Training and inference

- [Training runner](https://github.com/jiaxi-jiang/fbcnn/blob/main/main_train_fbcnn.py): `main_train_fbcnn.py`
- [Training options](https://github.com/jiaxi-jiang/fbcnn/tree/main/options): `options`
- [Color testing](https://github.com/jiaxi-jiang/fbcnn/blob/main/main_test_fbcnn_color.py): `main_test_fbcnn_color.py`
- [Grayscale testing](https://github.com/jiaxi-jiang/fbcnn/blob/main/main_test_fbcnn_gray.py): `main_test_fbcnn_gray.py`
- [Double-JPEG testing](https://github.com/jiaxi-jiang/fbcnn/blob/main/main_test_fbcnn_gray_doublejpeg.py): `main_test_fbcnn_gray_doublejpeg.py`
- [Real-image testing](https://github.com/jiaxi-jiang/fbcnn/blob/main/main_test_fbcnn_color_real.py): `main_test_fbcnn_color_real.py`

### Degradation data

- [Dataset selector](https://github.com/jiaxi-jiang/fbcnn/blob/main/data/select_dataset.py): `data/select_dataset.py`
- [Color JPEG dataset](https://github.com/jiaxi-jiang/fbcnn/blob/main/data/dataset_jpeg.py): `data/dataset_jpeg.py`
- [Grayscale JPEG dataset](https://github.com/jiaxi-jiang/fbcnn/blob/main/data/dataset_jpeggray.py): `data/dataset_jpeggray.py`
- [Double-JPEG dataset](https://github.com/jiaxi-jiang/fbcnn/blob/main/data/dataset_jpeggraydouble.py): `data/dataset_jpeggraydouble.py`

### FBCNN model

- [Model selector](https://github.com/jiaxi-jiang/fbcnn/blob/main/models/select_model.py): `models/select_model.py`
- [Training model](https://github.com/jiaxi-jiang/fbcnn/blob/main/models/model_fbcnn.py): `models/model_fbcnn.py`
- [Model base](https://github.com/jiaxi-jiang/fbcnn/blob/main/models/model_base.py): `models/model_base.py`
- [Network selector](https://github.com/jiaxi-jiang/fbcnn/blob/main/models/select_network.py): `models/select_network.py`
- [FBCNN network](https://github.com/jiaxi-jiang/fbcnn/blob/main/models/network_fbcnn.py): `models/network_fbcnn.py`
- [SSIM loss](https://github.com/jiaxi-jiang/fbcnn/blob/main/models/loss_ssim.py): `models/loss_ssim.py`
- [Other losses](https://github.com/jiaxi-jiang/fbcnn/blob/main/models/loss.py): `models/loss.py`

### Shared utilities

- [Image processing](https://github.com/jiaxi-jiang/fbcnn/blob/main/utils/utils_image.py): `utils/utils_image.py`
- [Inference helpers](https://github.com/jiaxi-jiang/fbcnn/blob/main/utils/utils_model.py): `utils/utils_model.py`
- [Option parsing](https://github.com/jiaxi-jiang/fbcnn/blob/main/utils/utils_option.py): `utils/utils_option.py`
- [Logging](https://github.com/jiaxi-jiang/fbcnn/blob/main/utils/utils_logger.py): `utils/utils_logger.py`
- [Batch norm helpers](https://github.com/jiaxi-jiang/fbcnn/blob/main/utils/utils_bnorm.py): `utils/utils_bnorm.py`
- [Regularizers](https://github.com/jiaxi-jiang/fbcnn/blob/main/utils/utils_regularizers.py): `utils/utils_regularizers.py`

### Other components

- **JPEG images**

## Connections

- Training runner → Option parsing: parses options (evidence: `main_train_fbcnn.py`)
- Training runner → Training options: uses configuration (evidence: `README.md`)
- Training runner → Dataset selector: creates datasets (evidence: `main_train_fbcnn.py`)
- Training runner → Model selector: creates model (evidence: `main_train_fbcnn.py`)
- Training runner → Logging: logs progress (evidence: `main_train_fbcnn.py`)
- Dataset selector → Color JPEG dataset: selects JPEG (evidence: `data/select_dataset.py`)
- Dataset selector → Grayscale JPEG dataset: selects grayscale (evidence: `data/select_dataset.py`)
- Dataset selector → Double-JPEG dataset: selects double-JPEG (evidence: `data/select_dataset.py`)
- Color JPEG dataset → Image processing: loads images (evidence: `data/dataset_jpeg.py`)
- Grayscale JPEG dataset → Image processing: loads images (evidence: `data/dataset_jpeggray.py`)
- Double-JPEG dataset → Image processing: loads images (evidence: `data/dataset_jpeggraydouble.py`)
- Model selector → Training model: creates model (evidence: `models/select_model.py`)
- Training model → Model base: inherits (evidence: `models/model_fbcnn.py`)
- Training model → Network selector: builds network (evidence: `models/model_fbcnn.py`)
- Network selector → FBCNN network: constructs FBCNN (evidence: `models/select_network.py`)
- Training model → SSIM loss: uses SSIM loss (evidence: `models/model_fbcnn.py`)
- Training model → Regularizers: uses regularizers (evidence: `models/model_fbcnn.py`)
- Model base → Batch norm helpers: uses helpers (evidence: `models/model_base.py`)
- Training model → Inference helpers: uses model helpers (evidence: `models/model_fbcnn.py`)
- Inference helpers → Image processing: uses image helpers (evidence: `utils/utils_model.py`)
- Color testing → JPEG images: reads images (evidence: `README.md`)
- Grayscale testing → JPEG images: reads images (evidence: `README.md`)
- Double-JPEG testing → JPEG images: reads images (evidence: `main_test_fbcnn_gray_doublejpeg.py`)
- Real-image testing → JPEG images: reads images (evidence: `README.md`)
- Color testing → FBCNN network: runs restoration (evidence: `main_test_fbcnn_color.py`)
- Grayscale testing → FBCNN network: runs restoration (evidence: `main_test_fbcnn_gray.py`)
- Double-JPEG testing → FBCNN network: runs restoration (evidence: `main_test_fbcnn_gray_doublejpeg.py`)
- Real-image testing → FBCNN network: runs restoration (evidence: `main_test_fbcnn_color_real.py`)
- Double-JPEG testing → Image processing: converts and scores (evidence: `main_test_fbcnn_gray_doublejpeg.py`)
- Double-JPEG testing → Logging: logs results (evidence: `main_test_fbcnn_gray_doublejpeg.py`)
