README.md

metadata

title: SPaCial AI OCR Service
emoji: 🔍
colorFrom: blue
colorTo: purple
sdk: docker
sdk_version: 4.21.0
app_file: app.py
pinned: false
license: mit

SPaCial AI OCR Service 🔍

Advanced OCR service using PaddleOCR for dimension detection in technical drawings. Perfect for extracting measurements, tolerances, and text from engineering drawings.

🚀 Features

• High-Accuracy OCR: Uses PaddleOCR with multiple detection modes
• Dimension Detection: Automatically identifies measurements and tolerances
• Vertical Text Support: Handles rotated text (90°, 180°, 270°)
• Interactive Canvas: Draw rectangles or click to detect text
• Real-time Processing: Fast and efficient text extraction
• Multiple Modes: Fast, accurate, and hardcore detection modes

🎯 How to Use

1. Upload an Image

• Click "Choose File" to upload a technical drawing
• Supported formats: JPG, PNG, WebP

2. Auto-Detection

• The system automatically detects all text zones
• Each zone shows confidence percentage and rotation angle

3. Manual Selection

• Double-click: Click on any text to detect it precisely
• Draw Rectangle: Drag to draw a rectangle around text for better detection

4. Zone Management

• Rotate: Click the rotate button to improve vertical text detection
• Fit: Click thumbnail to re-OCR with better accuracy
• Delete: Remove unwanted zones

🔧 API Endpoints

Process Image

POST /ocr/process
- file: Image file
- mode: "fast", "accurate", or "hardcore"
- rotation: 0, 90, 180, 270 degrees

Process Center Point

POST /ocr/process-center
- image: Base64 encoded image
- center_point: {"x": x, "y": y}
- rectangle_bounds: {"x1": x1, "y1": y1, "x2": x2, "y2": y2}
- use_rectangle: true/false
- rotation: rotation angle

🛠️ Technical Details

Detection Modes

• Fast: Quick detection for position finding
• Accurate: Balanced speed and accuracy
• Hardcore: Maximum accuracy with preprocessing

Vertical Text Handling

• Automatic detection of vertical text regions
• 90° rotation for better OCR accuracy
• Proper coordinate mapping back to original image

Text Correction

• Built-in correction rules for common OCR errors
• Tolerance parsing for engineering measurements
• Dimension validation and formatting

📊 Performance

• Processing Time: 1-3 seconds per image
• Accuracy: 95%+ for clear technical drawings
• Supported Languages: English (optimized for numbers and dimensions)
• Image Size: Up to 10MB, auto-resized for processing

🔍 Use Cases

• Engineering Drawings: Extract dimensions and tolerances
• Technical Documentation: Digitize handwritten measurements
• Quality Control: Verify measurements on drawings
• CAD Integration: Convert scanned drawings to digital data

🚨 Troubleshooting

Common Issues

1. Low Confidence: Try "hardcore" mode or draw a rectangle around text
2. Vertical Text: Use the rotate button to improve detection
3. Split Text: Draw a rectangle to capture the full text
4. Small Text: Ensure image resolution is sufficient

Tips for Best Results

• Use high-contrast images
• Ensure text is clearly visible
• For vertical text, use the rotate feature
• Draw rectangles around complex text regions

📝 License

This project is part of the SPaCial AI system for technical drawing analysis.

---

Ready to extract text from your technical drawings? Upload an image and start detecting! 🎯