Speed up `iter_frames` with `resized` problem

[houzeyu2683]

I try to speed up iter_frames in my script, my idea is resize the video to smaller size before iter_frames, but I found it is more slow, here is my code and the running time:

path = 'storage/PLmVlUjKuEG7Qkg8ipBRol7MFmgQ_HK0tl/video/_8tZJnFJ7Sc.mp4'
video = moviepy.VideoFileClip(path)
video = video.resized(1)
print(f"Size: {video.size}")
total = int((video.duration * video.fps))

start = time.time()
iteration = enumerate(video.iter_frames(with_times=True), 1)
for index, (timestep, frame) in iteration:
    print(f"Progress: [{index}/{total}]", end='\r')
    continue
_ = iteration
end = time.time()
print("")
print(end-start)
# Size: (1920, 1080)
# Progress: [18225/18225]
# 195.89423060417175

path = 'storage/PLmVlUjKuEG7Qkg8ipBRol7MFmgQ_HK0tl/video/_8tZJnFJ7Sc.mp4'
video = moviepy.VideoFileClip(path)
video = video.resized(0.5)
print(f"Size: {video.size}")
total = int((video.duration * video.fps))

start = time.time()
iteration = enumerate(video.iter_frames(with_times=True), 1)
for index, (timestep, frame) in iteration:
    print(f"Progress: [{index}/{total}]", end='\r')
    continue
_ = iteration
end = time.time()
print("")
print(end-start)
# Size: (960, 540)
# Progress: [18225/18225]
# 642.015567779541

The first method is using the original video size in the iter_frames, the second one is resized to half in the iter_frames.

[unconv]

When you do video.resized(0.5) it only tells MoviePy to resize the read frame after it was read with FFmpeg. That's why it's slower rather than faster. You can pass a target_resolution to the VideoFileClip instead to have FFmpeg resize it as it reads the video:

video = VideoFileClip("your_video.mp4", target_resolution=(960, 540))