📖 Autoscanlation - Automatic Comic Book Translation

Autoscanlation is an AI-powered tool that automatically detects, upscales, translates, and replaces text in scanned comic book pages.

🖼️ How it Works:

1. OCR Detection – Detects speech bubbles & text using Florence-2.
2. Text Region Upscaling – Uses ESRGAN to enhance text clarity.
3. Translation – Translates detected text into a target language.
4. Text Replacement – Overlays translated text back onto the image.

🎨 Perfect for:

• Manga & comic book scanlation (fan translations).
• Restoring low-resolution scans with enhanced readability.
• Automating multilingual comic translations.

🔧 Installation

1. Clone the repository:

   git clone https://github.com/yourusername/Autoscanlation.git
   cd Autoscanlation

2. Create a virtual environment:

   python -m venv venv
   source venv/bin/activate  # macOS/Linux
   venv\Scripts\activate     # Windows

3. Install dependencies:

   pip install -r requirements.txt

4. Download ESRGAN Pretrained Model To upscale text regions, you need to download the ESRGAN model file RRDB_ESRGAN_x4.pth.

Download the Model Download RRDB_ESRGAN_x4.pth or visit ESRGAN Pretrained Models. ] Move the Model File to the Correct Directory Once downloaded, place it inside ESRGAN/models/:

Autoscanlation/
│── ESRGAN/
│   ├── models/
│   │   ├── RRDB_ESRGAN_x4.pth  ✅ (Put the model here)
│   ├── test.py
│   ├── RRDBNet_arch.py
│   ├── ...

Verify Installation Run:

ls ESRGAN/models/

You should see:

RRDB_ESRGAN_x4.pth

Now ESRGAN will work correctly! ✅

🛠️ Usage

To run the full OCR & ESRGAN pipeline:

python flocr.py input_images output_images --source_lang fr --target_lang en

input_images/ → Folder containing scanned comic book pages. output_images/ → Translated pages with replaced text.

🖼️ Example

Before:

After:

📝 License

MIT License