ColabRTC

ColabRTC contains early experiments to stream media from your machine to a Google Colab instance using Web Real-Time Communication (WebRTC). It is built on top of aiortc.

The motivation behind this project was to learn about WebRTC and because an efficient implementation of this kind of communication could be useful to prototype Deep Learning models without a local GPU.

How to use ColabRTC?

The API allows one to create a ColabCall, which consists of a Python peer (running on Colab) and a Javascript peer (started in a Colab cell HTML context, which runs on the local machine). The signaling medium is the Colab filesystem (and Google Drive, optionally).

An example:

from call import ColabCall
import cv2

# Define a frame processing routine
def process_frame(frame, frame_idx):
  # To speed up, apply processing every two frames
  if frame_idx % 2 == 1:
    edges = cv2.Canny(frame, 100, 200)
    return cv2.cvtColor(edges, cv2.COLOR_GRAY2BGR)
  
call = ColabCall()
# Create the Python peer
call.create(frame_transformer=process_frame)
# Create the Javascript peer and display the UI
call.join()