nxva 1.3.3

tools for video analysis

NXVA - Nexuni Video Analysis

A comprehensive computer vision toolkit for video analysis, object detection, tracking, and pose estimation

Overview

NXVA (Nexuni Video Analysis) is a powerful and versatile Python package designed for advanced video analysis tasks. It provides a unified interface for object detection, multi-object tracking, pose estimation, and real-time streaming capabilities. Built with modularity and ease-of-use in mind, NXVA supports multiple deep learning frameworks and model formats.

✨ Key Features

• Multi-Model Object Detection: Support for YOLOv5, YOLOv11 with ONNX, PyTorch, and TensorRT formats
• Advanced Object Tracking: SimpleTracker and NexuniSort algorithms with feature-based tracking
• Pose Estimation: MMPose integration for human pose detection and analysis
• Multi-Camera Streaming: Real-time streaming with automatic reconnection and GStreamer support
• Flexible Configuration: YAML-based configuration system for easy setup and deployment
• GPU Acceleration: Full CUDA and TensorRT support for high-performance inference
• Model Conversion: Built-in tools for model format conversion and optimization

🏗️ Architecture

nxva/
├── v5/          # YOLOv5 detection, classification, pose estimation
├── v11/         # YOLOv11 detection, classification, pose estimation  
├── sort/        # Object tracking algorithms (SimpleTracker, NexuniSort)
├── pose/        # MMPose integration for pose estimation
├── utilities/   # Utility functions and tools
├── streaming/   # Multi-camera streaming capabilities
└── va/          # Video analysis server components

🚀 Quick Start

Installation

pip install nxva

Basic Usage

Each module provides detailed usage instructions and examples:

• Object Detection: See YOLOv11 README and YOLOv5 README for detection setup and usage
• Object Tracking: Refer to Sort README for SimpleTracker and NexuniSort usage
• Pose Estimation: Check Pose README for MMPose integration guide
• Multi-Camera Streaming: See Main README for streaming configuration
• Complete Examples: Explore tutorials/ for Jupyter notebook examples

📋 Requirements

Core Dependencies

• Python 3.6 or higher
• OpenCV 4.6.0+
• PyTorch 1.8.0+ (with CUDA support)
• NumPy 1.23.0+
• PyYAML 5.3.1+

Optional Dependencies

• For ONNX models: ONNX Runtime
• For TensorRT: TensorRT 7.0.0+ (not 10.1.0)
• For Pose Estimation: MMPose, MMDetection, MMEngine, MMCV
• For Advanced Features: ultralytics, torchvision

📚 Documentation & Examples

The package includes comprehensive tutorials and examples:

• Jupyter Notebooks: Step-by-step tutorials in tutorials/
• Configuration Examples: Ready-to-use configs in example/
• Specific Use Cases: Detection, tracking, pose estimation examples
• Module Documentation: Detailed README files for each component

Tutorial Topics

• YOLOv11 Training and Inference
• Multi-Object Tracking with NexuniSort
• Real-time Streaming Setup
• Pose Estimation with MMPose
• Model Conversion and Optimization

🎯 Use Cases

• Security & Surveillance: Real-time monitoring with object detection and tracking
• Sports Analysis: Pose estimation and movement analysis
• Industrial Automation: Quality control and process monitoring
• Retail Analytics: Customer behavior analysis and people counting
• Research & Development: Computer vision prototyping and experimentation

📄 License

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