---
license: mit
language:
  - en
tags:
  - image-restoration
---

<table>
  <tr>
    <td align="center">Input (B&W)</td>
    <td align="center">Colorized Output</td>
  </tr>
  <tr>
    <td><img src="DeOldify/test_images/1910Finland.jpg" alt="Input Black and White Image" height=500/></td>
    <td><img src="DeOldify/result_images/1910Finland.jpg" alt="Colorized Output Image" height=500/></td>
  </tr>
</table>

# Image Colorization Application

This application provides a web-based interface for colorizing black and white images using DeOldify technology. It consists of a FastAPI backend for processing images and a Streamlit frontend for a user-friendly interface.

## Features

- Colorize black and white images using DeOldify
- Choose between Artistic (more vibrant) and Stable (more realistic) models
- Adjust render factor to control quality and processing time
- Generate multiple renders with different factors to find the optimal result
- Download colorized images

## Installation

### Prerequisites

- Python 3.10
- CUDA-capable GPU (recommended for faster processing)
- Git, Git-Lfs

### Setup Instructions

1. **Clone the DeOldify repository inside your project directory:**

```bash
git clone https://huggingface.co/sayed99/Image-Colorizer
```

2. **Set up the environment:**

You can set up the environment using either Conda (recommended) or pip.

- **Using Conda (Recommended):**

```bash
conda env create -f environment.yml
conda activate deoldify
```

- **Using Pip:**

Make sure you have a compatible Python environment (3.10).

```bash
# Create a virtual environment
python -m venv .venv

# Activate the virtual environment
# On Windows:
.venv\Scripts\activate
# On macOS/Linux:
source .venv/bin/activate

# Install the required dependencies
pip install -r requirements.txt
```

This installs both the dependencies for our application and most of the dependencies needed for DeOldify.

## Directory Structure

- `DeOldify/`: Cloned repository containing the core DeOldify library.
- `models/`: Contains the downloaded `.pth` model weights.
- `input_images/`: Default location for uploaded images (can be configured).
- `output_images/` or `result_images/`: Default location for single colorized images.
- `multiple_renders/`: Default location for images generated using the multiple render factors feature.
- `app.py`: FastAPI backend server.
- `streamlit_app.py`: Streamlit frontend application.
- `requirements.txt`: Pip dependencies.
- `environment.yml`: Conda environment specification.

## Running the Application

1. **Start the FastAPI server:**

```bash
python app.py
```

This will start the API server on http://localhost:8000

2. **In a new terminal window, start the Streamlit web application:**

```bash
streamlit run streamlit_app.py
```

This will open the web interface in your browser, typically at http://localhost:8501

## API Documentation

Once the FastAPI server is running, you can access the API documentation at:

- http://localhost:8000/docs (Swagger UI)
- http://localhost:8000/redoc (ReDoc)

## API Endpoints

- `POST /colorize`: Colorize a single image with specified parameters
- `POST /colorize_multiple`: Generate multiple colorizations with different render factors
- `GET /image/{image_path}`: Retrieve a colorized image by its path

## Streamlit Web App

The web application provides an intuitive interface for:

- Uploading images
- Selecting colorization model (Artistic or Stable)
- Adjusting render factor
- Generating and comparing multiple renders
- Downloading colorized images

## Technical Details

- The application uses DeOldify's image colorization models
- Default render factor is 10, but can be adjusted between 5-50
- The artistic model is used by default, but can be switched to stable
- For multiple renders, you can specify the minimum and maximum render factors and step size

## Performance Tips

- If you encounter CUDA out-of-memory errors, try lowering the render factor
- For best results with the artistic model, a render factor of 10-15 is recommended
- For the stable model, a render factor of 20-25 typically works well
- increasing the render factor consume more gpu memory.

## License

This project uses DeOldify, which is licensed under MIT License.