IndoxMiner

IndoxMiner is a powerful Python library that leverages Large Language Models (LLMs) for data extraction, advanced object detection, and image classification. It combines schema-based data extraction from unstructured data sources such as text, PDFs, and images, with state-of-the-art object detection and image classification models. IndoxMiner enables seamless automation for document processing, visual recognition, and classification tasks.

🚀 Key Features

• Data Extraction: Extract structured data from text, PDFs, and images using schema-based extraction and LLMs.
• Object Detection: Leverage pre-trained object detection models for high-accuracy real-time image recognition.
• Image Classification: Utilize advanced classification models for identifying objects or features in images.
• OCR Integration: Extract text from scanned documents or images with integrated OCR engines.
• Schema-Based Extraction: Define custom schemas for data extraction with validation and type-safety.
• Multi-Model Support: Supports a wide range of models for detection and classification.
• Async Support: Built for scalability with asynchronous processing capabilities.
• Flexible Outputs: Export results to JSON, pandas DataFrames, or custom formats.

📦 Installation

You may also install required object detection dependencies like Detectron2 or YOLOv8 using: Install IndoxMiner with:

pip install indoxminer

You may also install additional dependencies for object detection and classification, such as YOLOv8 or Detectron2.

📝 Quick Start

1. Data Extraction

IndoxMiner integrates seamlessly with OpenAI models for schema-based extraction from text, PDFs, and images. Here's how you can extract structured data from a document:

Basic Text Extraction

from indoxminer import ExtractorSchema, Field, FieldType, ValidationRule, Extractor, OpenAi

# Initialize OpenAI extractor
llm_extractor = OpenAi(api_key="your-api-key", model="gpt-4-mini")

# Define extraction schema
schema = ExtractorSchema(
    fields=[
        Field(name="product_name", field_type=FieldType.STRING, rules=ValidationRule(min_length=2)),
        Field(name="price", field_type=FieldType.FLOAT, rules=ValidationRule(min_value=0))
    ]
)

# Create extractor and process text
extractor = Extractor(llm=llm_extractor, schema=schema)
text = """
MacBook Pro 16-inch with M2 chip
Price: $2,399.99
In stock: Yes
"""

result = await extractor.extract(text)
df = extractor.to_dataframe(result)

2. Object Detection

IndoxMiner provides powerful object detection capabilities with support for a variety of models, such as YOLO, Detectron2, and DETR.

Supported Models for Object Detection

Example: Object Detection with YOLOv5

from indoxminer.detection import YOLOv5

# Initialize YOLOv5 model
detector = YOLOv5()

# Detect objects in an image
image_path = "dog-cat-under-sheet.jpg"
outputs = await detector.detect_objects(image_path)


You can also switch to other models by specifying the model name, e.g., `detectron2`, `detr`, `yolov8`, etc.

```python
detector = YOLOv8()  # For YOLOv8

3. Image Classification

IndoxMiner now supports advanced image classification with models like SigCLIP, ViT, MetaCLIP, MobileCLIP, BioCLIP, AltCLIP, and RemoteCLIP.

Supported Models for Classification

Example: Classification with RemoteCLIP

from indoxminer.classification import RemoteCLIPClassifier
from PIL import Image

# Initialize RemoteCLIP
classifier = RemoteCLIPClassifier(
    model_name="ViT-L-14",
    checkpoint_path="/path/to/RemoteCLIP-ViT-L-14.pt"
)

# Classify an image
image = Image.open("/path/to/airport.jpg")
labels = ["An airport", "A university", "A stadium"]
classifier.classify(image, labels, top=3)

Example: Classification with SigCLIP

from indoxminer.classification import SigCLIPClassifier
from PIL import Image

# Initialize SigCLIP
classifier = SigCLIPClassifier()

# Classify an image with default labels
image = Image.open("/path/to/image.jpg")
classifier.classify(image)

🔧 Core Components

Classification Models

• Flexible Input: Supports single or batch image classification.
• Custom Labels: Define your own labels for classification tasks.
• Visualization: Generates bar plots of predicted probabilities.

🔍 Error Handling

IndoxMiner provides detailed error reporting for both data extraction, object detection, and classification tasks.

try:
    results = await extractor.extract(documents)
except Exception as e:
    print(f"An error occurred: {e}")

🤝 Contributing

We welcome contributions! Please follow the standard workflow:

1. Fork the repository.
2. Create a feature branch.
3. Commit your changes.
4. Push to the branch.
5. Open a Pull Request.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🆘 Support

• Documentation: Full documentation
• Issues: GitHub Issues
• Discussions: GitHub Discussions

🌟 Star History