ML Vision Lab 👁️

Modular computer vision implementations - A collection of production-grade vision systems spanning multiple domains.

Featured Projects • Installation • Quick Start • Contributing

📑 Table of Contents

• Project Organization
• Core Features
• Prerequisites
• Tech Stack
• Installation
• Quick Start
• Project Matrix
• Development Standards
• Contributing
• Documentation
• Benchmarks
• Versioning
• Authors
• Citation
• License
• Acknowledgments

🗂️ Project Organization

graph TD
    A[ML Vision Lab] --> B[projects]
    A --> C[core]
    A --> D[docs]
    B --> E[food-classification]
    B --> F[object-detection]
    B --> G[medical-imaging]
    B --> H[satellite-analysis]
    C --> I[utils]
    C --> J[models]
    C --> K[pipelines]
    D --> L[api]
    D --> M[guides]
    D --> N[architecture]

Loading

ml-vision-lab/
├── projects/               # Individual vision projects
│   ├── food-classification/  # Food analysis system
│   ├── object-detection/      # Real-time detection
│   ├── medical-imaging/       # DICOM processing
│   └── satellite-analysis/    # Geospatial vision
├── core/                   # Shared vision components
│   ├── utils/              # Common utilities
│   ├── models/             # Base model architectures
│   └── pipelines/          # Processing workflows
└── docs/                   # Project documentation

✨ Core Features

mindmap
  root((ML Vision Lab))
    Cross-Project
      Modular architecture
      Shared pipelines
      Hardware optimization
      Standardized metrics
    Project Types
      Classification
      Detection
      Medical
      Satellite
    Optimization
      GPU acceleration
      TensorRT
      Memory efficiency
    Development
      MLflow tracking
      DVC versioning
      CI/CD pipelines

Loading

Cross-Project Capabilities

• Modular project architecture
• Shared preprocessing pipelines
• Hardware-optimized inference
• Standardized evaluation metrics
• GPU-accelerated processing
• Production deployment examples
• Memory-efficient inference
• TensorRT integration

Project Types

• Image Classification
• Object Detection & Tracking
• Medical Imaging Analysis
• Satellite Imagery Processing
• Industrial Quality Inspection

🔧 Prerequisites

• Python 3.11+
• CUDA 12.2+
• OpenCV 5.0+
• PyTorch 2.3+
• TensorFlow 2.15+
• NVIDIA GPU (Compute Capability 6.0+)

🛠️ Tech Stack

graph TD
    A[Tech Stack] --> B[Core Libraries]
    A --> C[Project Libraries]
    B --> D[PyTorch]
    B --> E[TensorFlow]
    B --> F[OpenCV]
    B --> G[CUDA]
    C --> H[MONAI]
    C --> I[RasterIO]
    C --> J[DeepSORT]
    C --> K[MLflow]

Loading

Core Libraries

• PyTorch - Deep learning framework
• TensorFlow - Machine learning platform
• OpenCV - Computer vision operations
• CUDA - GPU acceleration
• TensorRT - Inference optimization
• NumPy - Numerical computing
• Pandas - Data manipulation
• Scikit-learn - Machine learning utilities
• Matplotlib - Visualization
• Plotly - Visualization
• Pillow - Image processing

Project-Specific Libraries

• MONAI - Medical imaging
• RasterIO - Geospatial analysis
• DeepSORT - Object tracking
• Albumentations - Image augmentation
• MLflow - Experiment tracking
• DVC - Data version control

📦 Installation

# Clone repository
git clone https://github.com/BjornMelin/ml-vision-lab.git
cd ml-vision-lab

# Create virtual environment
python -m venv .venv
source .venv/bin/activate  # Linux/MacOS
# or
.venv\Scripts\activate  # Windows

# Install core requirements
pip install -r requirements.txt

# Install project-specific requirements (optional)
pip install -r projects/food-classification/requirements.txt

🚀 Quick Start

Food Classification

from projects.food_classification import predict

result = predict("pizza.jpg")
print(f"Identified: {result.label} ({result.confidence:.1%})")

Object Detection

from projects.object_detection import VideoAnalyzer

analyzer = VideoAnalyzer(model="yolov9")
analyzer.process_stream("input.mp4", output="results.mp4")

📊 Project Matrix

Project	Task	Models	Input Types
Food Classification	Image Classification	EfficientNetV2, ViT	JPEG/PNG
Object Detection	Real-time Tracking	YOLOv9, DeepSORT	Video Streams
Medical Imaging	DICOM Analysis	UNet3+, MONAI	CT/MRI Scans
Satellite Analysis	Geospatial ML	ResNet50-ADE20K	GeoTIFF

🔧 Development Standards

flowchart TD
    A[Development] --> B[Code Quality]
    A --> C[Testing]
    A --> D[Documentation]
    B --> E[Black]
    B --> F[MyPy]
    C --> G[PyTest]
    C --> H[Coverage]
    D --> I[Docstrings]
    D --> J[Examples]

Loading

Code Quality

# Format all projects
black projects/

# Type checking
mypy projects/

# Run tests
pytest projects/ --cov

Project Structure Template

projects/new-project/
├── app/          # Application interface
├── engine/       # Core logic
├── models/       # Trained weights
├── tests/        # Unit tests
├── README.md     # Project docs
└── requirements.txt # Local dependencies

🤝 Contributing

Adding New Projects

1. Create project folder in projects/
2. Follow structure template
3. Add cross-links to:
   • Core utilities (avoid duplication)
   • Related projects
4. Submit PR with:
   • Black-formatted code
   • Google-style docstrings
   • Unit tests (≥80% coverage)

See CONTRIBUTING.md for full guidelines.

📚 Documentation

Pipeline Optimization

graph LR
    A[Input] --> B[Preprocessing]
    B --> C[Inference]
    C --> D[Postprocessing]
    B --> E[GPU Pipeline]
    C --> F[TensorRT]
    D --> G[Batch Processing]

Loading

• GPU-accelerated preprocessing
• Batch processing optimization
• Memory-efficient inference
• TensorRT integration
• Multi-GPU support
• Mixed precision training

Models

Model	Task	Performance	Speed (FPS)
YOLOv8	Detection	mAP: 52.3	120
Mask R-CNN	Segmentation	mAP: 47.8	45
DeepSORT	Tracking	MOTA: 76.5	80

📊 Benchmarks

Performance on standard datasets:

Task	Dataset	Model	GPU	FPS	Accuracy
Detection	COCO	YOLOv8	A100	120	mAP: 52.3
Segmentation	COCO	Mask R-CNN	V100	45	mAP: 47.8
Tracking	MOT17	DeepSORT	3090	80	MOTA: 76.5

📌 Versioning

We use SemVer for versioning. For available versions, see the tags on this repository.

✍️ Authors

Bjorn Melin

• GitHub: @BjornMelin
• LinkedIn: Bjorn Melin

📝 Citation

@misc{melin2024mlvisionlab,
  author = {Melin, Bjorn},
  title = {ML Vision Lab: Production Computer Vision Implementations},
  year = {2024},
  publisher = {GitHub},
  url = {https://github.com/BjornMelin/ml-vision-lab}
}

📄 License

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

🙏 Acknowledgments

• OpenCV community
• YOLO authors and contributors
• Deep SORT implementation team
• Medical imaging community (MONAI)
• Satellite imagery processing teams
• TensorFlow and PyTorch teams
• NVIDIA for CUDA and TensorRT support

---

Made with 👁️ and ❤️ by Bjorn Melin