pythonopencvvideo-processingpython-imageio

How to seek a frame in video using `imageio`?


I have a video and I would like to extract only specific frames from the video.

Currently what I do is:

index = [1,2,3,4,5,6,7,8]
img_list = []
for i, frame in enumerate(iio.imiter("imageio:cockatoo.mp4")):
    if i in index:
        img_list.append(frame)

img_array = np.asarray(img_list)

Is there a way to 'seek' to only the frames I want like it is done in opencv as shown here?


Solution

  • You can get specific frames by using the index kwarg of imread:

    import imageio.v3 as iio
    import numpy as np
    
    index = [1,2,3,4,5,6,7,8]
    img_list = []
    for idx in index:
        img_list.append(iio.imread("imageio:cockatoo.mp4", index=idx))
    
    img_array = np.stack(img_list)
    

    If you want more performance, you could use pyav as backend instead of the default imageio-ffmpeg.

    import imageio.v3 as iio
    import numpy as np
    
    index = [1,2,3,4,5,6,7,8]
    img_list = []
    for idx in index:
        img_list.append(iio.imread("imageio:cockatoo.mp4", index=idx, plugin="pyav"))
    
    img_array = np.stack(img_list)
    

    Timings (for 10 repeats each):

    >>> timeit.timeit('[iio.imread("imageio:cockatoo.mp4", index=idx) for idx in index]', setup="from __main__ import index, iio", number=10) 
    9.865169799999876
    >>> timeit.timeit('[iio.imread("imageio:cockatoo.mp4", index=idx, plugin="pyav") for idx in index]', setup="from __main__ import index, iio", number=10) 
    2.250104900000224
    

    This, ofc, has the drawback of re-opening the file each time, which you can avoid using imopen:

    import imageio.v3 as iio
    import numpy as np
    
    index = [1,2,3,4,5,6,7,8]
    
    with iio.imopen("imageio:cockatoo.mp4", "r") as img_file:
        img_list = [img_file.read(index=idx) for idx in index]
    
    img_array = np.stack(img_list)
    

    Unfortunately, the imopen route currently doesn't work for pyav, because of a bug that I missed while writing the plugin. However, since I am aware of it now, I should be able to write a fix in the coming days :)

    Edit: The bug is now fixed. You can now use

    import imageio.v3 as iio
    import numpy as np
    
    index = [1,2,3,4,5,6,7,8]
    
    with iio.imopen("imageio:cockatoo.mp4", "r", plugin="pyav") as img_file:
        img_list = [img_file.read(index=idx) for idx in index]
    
    img_array = np.stack(img_list)