Layout Analysis in OmniDocs
Layout analysis is the process of detecting and classifying regions (text, tables, images, etc.) in documents or images. OmniDocs provides a unified interface to several state-of-the-art layout detection backends, making it easy to experiment, compare, and integrate them into your workflows.
Layout Analysis?
Layout analysis breaks a document page into its logical components—like paragraphs, tables, figures, and headers—by predicting bounding boxes and labels for each region. This is a crucial first step for downstream tasks like OCR, table extraction, and document understanding.
🧩 Supported Layout Detectors
OmniDocs supports multiple layout detection engines, each with its own strengths:
| Detector | Backend/Model | Highlights |
|---|---|---|
| Paddle | PaddleOCR Layout | Fast, robust, easy to use, good for scanned docs |
| RTDETR | RT-DETR | Real-time, transformer-based, accurate |
| Surya | Surya Layout | Modern, high-accuracy, Indian docs friendly |
| YOLO | YOLOv8/YOLOv5 | Fast, customizable, works on many layouts |
| Florence | Florence Layout | (If available) Large foundation model, generalizes well |
Tip: You can easily switch between detectors by changing a single import/class name.
📝 How to Use
All layout detectors follow the same API pattern:
from omnidocs.tasks.layout_analysis.extractors.paddle import PaddleLayoutDetector
detector = PaddleLayoutDetector(device='cpu')
image_path = "path/to/your/document.png"
annotated_image, layout_output = detector.detect(image_path)
# Visualize results
detector.visualize((annotated_image, layout_output), "output.png")
- Change
PaddleLayoutDetectorto any other detector (e.g.,RTDETRLayoutDetector,SuryaLayoutDetector,YOLOLayoutDetector) to use a different backend. - All detectors return both the annotated image and a structured output with bounding boxes and labels.
📒 Example Notebooks
See the tutorial notebooks for hands-on examples: - Paddle Layout Analysis - YOLO Layout Analysis - RTDETR Layout Analysis - Surya Layout Analysis
Each notebook demonstrates: - How to initialize and use the detector - How to visualize results - How to interpret the output
🛠️ Advanced Tips
- Device Selection: Most detectors support
device='cpu'ordevice='cuda'for GPU acceleration. - Custom Models: For YOLO and Surya, you can plug in your own trained weights.
- Batch Processing: Use Python loops or scripts to process folders of images.
- Output Structure: All detectors return bounding boxes, class labels, and (optionally) confidence scores.
🔗 Next Steps
- Try out the notebooks with your own documents.
- Read the API Reference for advanced usage and customization.
- Explore downstream tasks like OCR and table extraction using the detected layouts.
OmniDocs makes layout analysis accessible, reproducible, and extensible, no matter which backend you choose.