Metadata-Version: 2.2
Name: binaexperts
Version: 0.5.4
Summary: Binaexperts SDK, a comprehensive computer vision toolkit
Author: Nastaran Dab, Mahdi Tajdari
Author-email: n.dab@binaexperts.com, m.tajdari@binaexperts.com
License: MIT
Classifier: Programming Language :: Python :: 3
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Requires-Python: >=3.9
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: PyYAML==6.0.2
Requires-Dist: setuptools==68.2.0
Requires-Dist: requests
Requires-Dist: jsonschema==4.23.0
Requires-Dist: opencv-python==4.10.0.84
Requires-Dist: numpy==2.1.1
Provides-Extra: inference
Requires-Dist: torch>=2.0.0; extra == "inference"
Requires-Dist: torchvision; extra == "inference"
Requires-Dist: gitpython>=3.1.30; extra == "inference"
Requires-Dist: matplotlib>=3.8.0; extra == "inference"
Requires-Dist: pandas>=2.1.0; extra == "inference"
Requires-Dist: tqdm>=4.66.1; extra == "inference"
Requires-Dist: seaborn>=0.13.2; extra == "inference"
Requires-Dist: scipy>=1.12.0; extra == "inference"
Dynamic: author
Dynamic: author-email
Dynamic: classifier
Dynamic: description
Dynamic: description-content-type
Dynamic: license
Dynamic: provides-extra
Dynamic: requires-dist
Dynamic: requires-python
Dynamic: summary

# BinaExperts SDK

**BinaExperts SDK** is a comprehensive toolkit for computer vision, designed to empower developers and researchers in building innovative visual solutions. Focused on scalability and versatility, it provides a robust foundation for creating state-of-the-art computer vision applications. Its architecture supports seamless integration with various platforms and is continuously evolving to address emerging challenges and opportunities in the field.


## Installation

Install the BinaExperts SDK directly from PyPI:

```bash
pip install binaexperts
```
If you need additional dependencies for inference, install with:

```bash
pip install binaexperts[inference]
```

This will install additional packages required for running inference tasks, such as PyTorch and torchvision.
Ensure that you have a compatible CUDA setup if using GPU acceleration.

## Usage

Once installed, you can start leveraging the SDK’s two core functionalities:

- **Dataset Conversion**: Easily convert datasets between various formats using a unified API. The conversion module supports both file-based and in-memory operations and currently supports the following formats:
  - **COCO**
  - **YOLO**
  - **BinaExperts**
- **Inference**: Perform predictions on static images or live video streams. Inference currently supports YOLOv5 and YOLOv7 models for local and live applications.


## 1. Dataset Conversion

The `Convertor` class simplifies the process of converting datasets between formats such as COCO, YOLO, and BinaExperts.

```python
import binaexperts.convertors

# Initialize the Convertor
convertor = binaexperts.convertors.Convertor()

# Convert from a source dataset to a target format
converted_output = convertor.convert(
    target_format='yolo',
    source='path/to/source_dataset.zip',  # Use 'source' instead of 'source_path' to match method signature
    destination='path/to/target_dataset.zip'  # Optional; if omitted, returns a file-like object.
)

# Save the output if it's returned as an in-memory object
if converted_output:
    with open("output_dataset.zip", "wb") as f:
        f.write(converted_output.read())

print("Conversion completed successfully!")

```
### In-Memory IO Conversion Example

In this example, we demonstrate how to convert a dataset directly from an in-memory BytesIO object and obtain the result in memory—no disk I/O required.
```python
import io
import binaexperts

# Simulate an in-memory YOLO dataset as a BytesIO object
yolo_dataset_buffer = io.BytesIO()
# Assuming yolo_dataset_zip_content is the binary content of a YOLO zip file
yolo_dataset_zip_content = b"binary content of YOLO zip file"  # Placeholder for actual binary content
yolo_dataset_buffer.write(yolo_dataset_zip_content)
yolo_dataset_buffer.seek(0)  # Reset to the beginning

# Initialize the Convertor
convertor = binaexperts.Convertor()

# Perform the conversion from YOLO to COCO format using in-memory IO
output_coco_buffer = convertor.convert(
    target_format='coco',
    source=yolo_dataset_buffer,  # Input as BytesIO
)

# Access the converted COCO dataset in memory
coco_data = output_coco_buffer.getvalue()

# Further processing or sending `coco_data` over a network...
print("In-memory conversion completed successfully!")
```
**Summary**: Use the conversion module to seamlessly switch between dataset formats with minimal configuration, whether working with files on disk or in-memory streams.

---

## 2. Inference

The SDK now includes inference capabilities that allow you to run predictions on images and video streams.

### Static Image Inference

Use the `image_inference` function to perform predictions on a single image. Choose between **YOLOv5** or **YOLOv7** as your model type.

```python
from binaexperts.app import Inference

# Initialize image inference (select 'yolov5' or 'yolov7')
inf = Inference(model_type='yolov5',
                device='cuda',
                model_path="path/to/model.pt",
                iou=0.5,
                confidence_thres=0.5)

inf.live_inference(source=0)
```


### Live Inference

For real-time applications, use the `LiveInference` class to process video streams or camera inputs.

```python
from binaexperts.app import Inference

# Initialize live inference (choose 'yolov5' or 'yolov7')
inf = Inference(model_type='yolov5',
                device='cuda',
                model_path="path/to/model.pt",
                iou=0.5,
                confidence_thres=0.5)

result = inf.local_inference(image_path="path/to/image",
                             destination="path/to.destination"
                             , output_format="png")
```


**Summary**: The inference module extends the SDK’s capabilities, enabling both static and live predictions. It supports YOLOv5 and YOLOv7 models, allowing you to integrate computer vision inference into various applications seamlessly.

---


## Features

- **Dataset Conversion**: Convert datasets effortlessly between various formats.
- **Inference**: Run predictions on static images and live streams using YOLOv5/YOLOv7.
- **Modular Design**: Easily extendable for future formats, training pipelines, and additional inference features.
- **Flexible IO**: Supports both file-based and in-memory operations for versatile deployment scenarios.

---

## Future Roadmap

### Data Preparation
- Enhanced conversion tools and additional format support.
- Automated dataset validation to ensure data integrity.

### Training
- Auto-training workflows with model selection, hyperparameter tuning, and comprehensive training pipelines.

### Inference Enhancements
- Further improvements for both local and live inference.
- Expanded support for additional model architectures.

### Community Suggestions
- We welcome your ideas and contributions to further enhance the SDK.

---

## Project Structure

```
binaexperts/
│
├── bina/
│   ├── app/
│   ├── services/
│   ├── __init__.py
│   
│
├── common/
│   ├── __init__.py
│   ├── loadhelpers.py
│   ├── logger.py
│   ├── setup_utils.py
│   ├── utils.py
│   ├── yolo_utils.py
│
├── convertors/
│   ├── schema/
│   │   ├── __init__.py
│   │   ├── base.py
│   │   ├── const.py
│   │   ├── convertor.py
│
├── SDKs/
│   ├── YOLO/
│   │   ├── yolov5/
│   │   ├── yolov7/
│   │   ├── __init__.py
│
├── __init__.py
```

---

## Contributing

The **BinaExperts SDK** was designed and developed by the technical team at **BinaExperts**, led by **Nastaran Dab** and **Mahdi Tajdari**. Contributions, bug reports, documentation improvements, and feature suggestions are welcome. Please reach out to the project team for contribution guidelines.

---

## Acknowledgments

Special thanks to **Nastaran Dab** and **Mahdi Tajdari** for their leadership and contributions in developing and maintaining the **BinaExperts SDK**.

---

## License

This project is licensed under the **MIT License**.
