I need to grab frames from the webcam of a laptop, transmit them one by one and the receiving side stitch them into a video. I picked ffmpeg-python as wrapper of choice and the example from the docs works right away:

#!/usr/bin/env python

# In this file: reading frames one by one from the webcam.


import ffmpeg

width = 640
height = 480


reader = (
    ffmpeg
    .input('/dev/video0', s='{}x{}'.format(width, height))
    .output('pipe:', format='rawvideo', pix_fmt='yuv420p')
    .run_async(pipe_stdout=True)
)

# This is here only to test the reader.
writer = (
    ffmpeg
    .input('pipe:', format='rawvideo', pix_fmt='yuv420p', s='{}x{}'.format(width, height))
    .output('/tmp/test.mp4', format='h264', pix_fmt='yuv420p')
    .overwrite_output()
    .run_async(pipe_stdin=True)
)


while True:
    chunk = reader.stdout.read(width * height * 1.5)  # yuv
    print(len(chunk))
    writer.stdin.write(chunk)

Now for the compression part.

My reading of the docs is that the input to the reader perhaps needs be rawvideo but nothing else does. I tried replacing rawvideo with h264 in my code but that resulted in empty frames. I'm considering a third invocation looking like this but is that really the correct approach?

encoder = (                                                                     
    ffmpeg                                                                      
    .input('pipe:', format='rawvideo', pix_fmt='yuv420p', s='{}x{}'.format(width, height))
    .output('pipe:', format='h264', pix_fmt='yuv420p')                          
    .run_async(pipe_stdin=True, pipe_stdout=True)