Skip to content

Splitting Videos

This how-to guide showcases various ways to split a video using fmdt, using demo.mp4 as our sample video

Split at arbitrary intervals

We can split a video at any arbitrary interval using the function fmdt.split_video_at_intervals() which has the following signature:

def split_video_at_intervals(
        video_filename: str,
        start_end: list[tuple[int, int]],
        nframes_before=3,
        nframes_after=3,
        overwrite=False,
        verbose: bool = False,
        condense: bool = True,
        out_dir = None 
    ) -> None:

Let's use split_video_at_intervals to make two frame-perfect cuts of demo.mp4 using the intervals (5, 20) and (100, 200)

import fmdt

vid = "demo.mp4"
start_end = [(5, 20), (100, 200)]

fmdt.split_video_at_intervals(vid, start_end, 0, 0, True)

This will create a new folder demo with the following structure:

./demo
├── f005-020.mp4
└── f100-200.mp4

Split video at meteors

We can call fmdt-detect and then split a video where meteors are detected.

import fmdt

fmdt.detect(vid_in_path="demo.mp4", trk_path="tracks.txt").split()

Which uses ffmpeg to approximately (to the nearest millisecond) trim a video around non-overlapping meteor detections.

There is also a more verbose alternative which accomplishes the same split:

fmdt.detect(vid_in_path="demo.mp4", trk_path="tracks.txt")
fmdt.split_video_at_meteors("demo.mp4", "tracks.txt", overwrite=True)

Split and Visualize

We can produce more informative videos if we apply fmdt-visu before our splitting operation:

vid = "demo.mp4"
trk = "tracks.txt"
log = "detect_log"
trk2roi = "trk2roi.txt"

dres = fmdt.detect(vid_in_path=vid, 
                   trk_path=trk,
                   log_path=log, 
                   trk_roi_path=trk2roi)

dres.log_parser().visu().split()

Real Examples

In this section we are going to show a real example using fmdt.split_video_at_intervals. 

intervals = [(102, 149), (156, 189), (194, 204), (223, 231), (249, 256)]

fmdt.split_video_at_intervals(video_filename="demo.mp4", start_end=intervals, nframes_before=-15, nframes_after=50)

These are the intervals of interest. In this specific case the frames aren't precise so when we call split_video_at_intervals we add a -15 frame buffer "before" our intended start plus a 50 frame buffer after our intended stop frame.

Danger

TODO: This is buggy for the last split, indeed 256 + 50 = 306 frames is out of the full video range (full video is 256 frames). We should managed this better.

If you have initialized your config and have all of the correct windows videos then we can go ahead and load up the first one:

import fmdt
v = fmdt.load_window()[0]

#============ Real interactive session ==================#
>>> v 
window_3_sony_0400-0405UTC.mp4

>>> v.full_path()
'/home/ejovo/Videos/Window/window_3_sony_0400-0405UTC.mp4'

We then split the video with one line:

intervals = [(785, 790), (1222, 1250), (1426, 1439), (2288, 2323), (2836, 2850),
             (2810, 2888), (2928, 2933), (3426, 3434), (3857, 3862), (4155, 4179), 
             (4262, 4268), (4447, 4460), (5323, 5330), (6790, 6811), (7199, 7207)]

fmdt.split_video_at_intervals(v.full_path(), intervals, nframes_before=-15, nframes_after=50)