Extract images from a video as numpy

[srikanthsridar]

Hello kkroening,
Is it possible to use ffmpeg extracted frames in jpg to numpy using ffmpeg-python?

[depau]

If you run ffmpeg manually with subprocess you can read (write) from (to) any pipe of the process.

Check this code from the tests.

Note that a .compile() operator has been added, which returns the command line arguments ready for subprocess, i.e.:

p = subprocess.Popen(out.compile(), stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
output = p.stdout.read()

You may also want to check this out.

[srikanthsridar]

Sure thank you.Will check this and get back to you.

[srikanthsridar]

Between is it possible to mention image format as "jpg" or "png" in ffmpeg in extraction step?

[srikanthsridar]

I am able to pipe the images but the image size(file size) in RAM is more than the jpg extraction.I will be reading thousands of files so I need in compressed format like jpg.Any thoughts on this?

[depau]

You should check the ffmpeg documentation. I'm pretty sure there's an encoder for jpeg but I don't remember how it works off the top of my head.
See here if you can find anything: https://ffmpeg.org/ffmpeg-all.html

[kkroening]

Here's an example in Jupyter notebook:

I'll post the notebook file in the examples soon.

[kkroening]

If you want to read individual frames as jpg:

def read_frame_as_jpg(in_filename, frame_num):
    out, err = (
        ffmpeg
        .input(in_filename)
        .filter_('select', 'gte(n,{})'.format(frame_num))
        .output('pipe:', vframes=1, format='image2', vcodec='mjpeg')
        .run(capture_stdout=True)
    )
    return out