StreamGrid ⚡

Ultra-fast multi-stream video display, Run real-time object detection across multiple video feeds with real-time speed. Whether you're on CPU or GPU, StreamGrid handles the load like a champ.

From smart surveillance to AI-powered dashboards, StreamGrid makes it effortless to display and analyze multiple video streams side-by-side, with Ultralytics YOLO-based detection running on all of them in real time.

Installation

pip install streamgrid

Quick Start

Python

from ultralytics import YOLO
from streamgrid import StreamGrid

# Load model
model = YOLO("yolo11n.pt")
# model = YOLO("yolo11n_openvino_model")  ~2-3x faster
StreamGrid(model=model)  

# Inference on GPU
StreamGrid(
    sources=[
        "path/to/video1.mp4", "path/to/video2.mp4", 
        "path/to/video3.mp4", "path/to/video4.mp4"
    ], 
    device="cuda")

# Store stream results in CSV file
StreamGrid(sources=sources, analytics=True)

CLI (Command Line Interface)

streamgrid model=yolo11n.pt

# Run inference with GPU device
streamgrid model=yolo11n.pt device=0  

# save the output
streamgrid model=yolo11n.pt device=cpu save=True

# Pass source
streamgrid model=yolo11n.pt sources=["video1.mp4", "video2.mp4"]

# Store Stream results in CSV file
streamgrid model=yolo11n.pt analytics=True

Performance (Beta, final benchmarks will be released soon)

StreamGrid automatically optimizes performance based on the number of streams:

• 1-2 streams: 640×360 resolution, up to 15 FPS per stream
• 3-4 streams: 480×270 resolution, up to 10 FPS total (CPU processing)
• 5-9 streams: 320×180 resolution, up to 5 FPS per stream
• 10+ streams: 240×135 resolution, up to 3 FPS per stream

Note: Performance benchmarks are based on CPU processing. GPU acceleration can significantly improve throughput.

Contributing

We welcome contributions! Please feel free to submit a Pull Request or open an issue for discussion.