README.md

metadata

title: Comic Panel Extractor
emoji: 📚
colorFrom: indigo
colorTo: purple
sdk: docker
pinned: false

https://jebin2-comic-panel-extractor.hf.space/

📚 Comic Panel Extractor

Automatically extract panels from comic pages using YOLO segmentation and image processing.

Note: Currently using pretrained model from mosesb/best-comic-panel-detection until custom training is complete. The model downloads automatically on first run.

Installation

git clone https://github.com/jebin2/comic-panel-extractor.git
cd comic-panel-extractor
pip install -e .

Usage

Extract Panels

Web Interface:

serve-comic-panel-extractor
# Visit http://localhost:7860

CLI:

comic-panel-extractor path/to/comic.jpg

Python:

from comic_panel_extractor.main import ComicPanelExtractor
from comic_panel_extractor.config import Config, load_config

config = load_config()
config.input_path = "comic.jpg"
extractor = ComicPanelExtractor(config)
panels, data, paths = extractor.extract_panels_from_comic()

Annotate Data

Visit http://localhost:7860/annotate

Shortcuts:

• Click/drag for boxes
• S = Save
• D = Auto-detect
• Delete = Remove

Train Model

1. Setup Dataset:

# Add to .env
SOURCE_PATH=/path/to/images

# Create dataset (80/10/10 split)
python -m comic_panel_extractor.create_dataset

2. Configure Training (.env):

EPOCH=200
YOLO_BASE_MODEL_NAME=yolo11s-seg
YOLO_MODEL_NAME=comic_panel

3. Train:

python -m comic_panel_extractor.train

Run Inference

python -m comic_panel_extractor.inference

from comic_panel_extractor.yolo_manager import YOLOManager

with YOLOManager() as yolo:
    yolo.load_model('weights.pt')
    yolo.annotate_images(['image.jpg'], 'output')

Configuration

config = load_config()
config.min_width_ratio = 0.15    # Min panel width (% of image)
config.min_height_ratio = 0.15   # Min panel height (% of image)
config.min_area_ratio = 0.05     # Min panel area (% of image)

Docker

docker build -t comic-panel-extractor .
docker run -p 7860:7860 comic-panel-extractor

API Endpoints

Extract:

• POST /api/extract/convert - Upload & extract panels

Annotate:

• GET /api/annotate/images - List images
• GET /api/annotate/annotations/{image} - Get annotations
• GET /api/annotate/detect_annotations/{image} - Auto-detect
• POST /api/annotate/annotations - Save annotations

Author

Jebin Einstein E - jebineinstein@gmail.com