🚀 Solovision

Solovision is a state-of-the-art real-time object detection and tracking system that seamlessly integrates with ReID (Re-Identification) architecture. Built on top of YOLO object detection, it provides robust multi-object tracking capabilities with advanced features for identity preservation across frames.

✨ Key Features

• 🎯 High-Performance Tracking: Implements ByteTrack algorithm for reliable multi-object tracking
• 🔄 ReID Integration: Advanced re-identification capabilities for maintaining object identity
• 🚀 Real-time Processing: Optimized for real-time applications with efficient processing
• 📊 Multiple Detection Backends: Support for YOLOv8, YOLOv9, YOLOv11 and all other previous YOLO variants
• 💪 Robust Motion Prediction: Kalman filtering for smooth trajectory estimation
• 🎨 Flexible Visualization: Customizable visualization options for tracking results
• 🔧 Easy-to-use CLI: Simple command-line interface for quick deployment

🛠️ Installation

Install the solovision package in a Python>=3.9 environment.

pip install solovision

Install from source:

git clone https://github.com/AIEngineersDev/solovision.git
cd solovision
pip install .

Install in Dev

pip install poetry
poetry install
poetry shell

🚀 Quick Start

Basic Usage

from solovision import ByteTracker
from ultralytics import YOLO
import cv2

# Initialize tracker
tracker = ByteTracker(
    reid_weights="path/to/reid/weights",
    device="cuda",
    half=True
)

# Process video
cap = cv2.VideoCapture(0)
while True:
    ret, frame = cap.read()
    if not ret:
        break
        
    # Get detections from yolo
    model = YOLO('yolov8m.pt')
    detections = model.predict(frame)
    
    # Update tracker
    tracks = tracker.update(detections, frame)
    
    # Process tracking results
    for track in tracks:
        bbox = track[:4]
        track_id = track[4]
        # Draw or process tracking results

Command Line Interface

# Detect objects across videos or streams
solovision detect --source video_path --conf 0.25 --iou 0.45  

# Track objects using unique id with custom settings
solovision track --source video_path --yolo-model yolov8n.pt --reid-model osnet_x1_0_msmt17.pt --show --save --half \
                --show-trajectories --save-txt --save-crops --per-class \
                --classes 0 2 --device 0 --imgsz 640

# Runs Interactive Web Application to perform real-time inference
solovision run_app 

# View all available CLI args
solovision --help

🎯 ReID Models Support

Solovision supports various state-of-the-art ReID architectures:

• OSNet (x0.25, x0.5, x0.75, x1.0)
• OSNet-AIN
• OSNet-IBN
• ResNet (50, 101)
• CLIP-ReID

Check out the Model Zoo for pre-trained weights and performance benchmarks.

🔧 Advanced Features

• Tracking Analytics: Line graphs and timestamp plotting for track id's
• Separate Merged Tracks: Save separate videos of persistant tracks from multiple video sources
• Per-Class Tracking: Enable separate tracking for different object classes
• Feature History: Maintain temporal appearance features for robust tracking
• Camera Motion Compensation: Automatic adjustment for camera movement
• Multi-Camera Support: Persist Tracker across multiple cameras/source

📊 Performance

• Runs at 30+ FPS on modern GPUs with YOLOv8n
• Support for half-precision (FP16) inference
• Optimized for both accuracy and speed
• Scalable for multi-camera deployments

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.

📝 License

This project is licensed under the GNU Affero General Public License v3.0 - see the LICENSE file for details.

📚 Citation

@software{solovision2024,
  author = {Diddi, Dhruv and Mohammed, Zeeshaan},
  title = {Solovision: State-of-the-art Real-Time Object Tracking System},
  year = {2024},
  publisher = {GitHub},
  organization = {AIEngineersDev},
  url = {https://github.com/AIEngineersDev/solovision}
}

🙏 Acknowledgments

• ByteTrack algorithm implementation
• Ultralytics YOLO
• OSNet for ReID features
• BOXMOT
• FastReID

---

Made with ❤️ by Solo