visiongraph-ndi 0.2.0

NDI support for the visiongraph library.

Visiongraph NDI

Visiongraph NDI is an extension of the visiongraph library, adding support for real-time video streaming over the network via NewTek's NDI® protocol. This library provides both sending and receiving capabilities, making it easy to integrate NDI into computer vision workflows.

Installation

Install visiongraph NDI with pip:

pip install visiongraph-ndi

Requirements

Ensure you have the latest version of visiongraph installed.

Usage Examples

Sending Video Streams

The NDIVideoOutput class allows you to broadcast images in real-time. This example shows how to generate a simple time-based pattern and send it as a video stream.

from visiongraph_ndi.NDIVideoOutput import NDIVideoOutput

# Set up NDI output with a stream name
with NDIVideoOutput("Demo") as ndi:
    image_bgr = ... # load or get an image in BGR format
    ndi.send(image_bgr)  # Send the generated image

The NDIVideoOutput class initializes an NDI output stream with the given name. Use ndi.send(frame) to transmit frames, which can be in BGR or BGRA format.

Receiving Video Streams

The NDIVideoInput class supports finding available NDI sources and streaming from them in real-time. Below are examples of how to find NDI sources and receive streams.

Finding Sources

Use NDIVideoInput.find_sources() to list all available NDI sources on the network. It is possible to set the timeout to find sources (default: 1.0 seconds).

from visiongraph_ndi.NDIVideoInput import NDIVideoInput

# Discover NDI sources for 5 seconds
for source in NDIVideoInput.find_sources(timeout=5.0):
    print(source.name)

Receiving Video Streams

The following example demonstrates how to receive a video stream from a the first NDI source available and display it using OpenCV.

import cv2
from visiongraph_ndi.NDIVideoInput import NDIVideoInput

# Connect to an NDI source
with NDIVideoInput(stream_name="Demo") as ndi:
    while ndi.is_connected:
        ts, frame = ndi.read()  # Receive timestamped frame

        if frame is None:
            continue  # Skip if no frame is received

        # Display the frame
        cv2.imshow(f"NDI {ndi.source.stream_name}", frame)
        cv2.waitKey(1)

The NDIVideoInput class automatically connects to the first available NDI source or allows for specific source selection if desired (using the stream_name and/or host_name parameter in the constructor). By calling ndi.read(), you retrieve frames as numpy arrays, ready for processing or display.

About

Visiongraph NDI is powered by the NDI Python wrapper cyndilib (MIT license). Special thanks to nocarryr for making this integration possible.