Extending ClipToolkit.py

[5559175]

I have a few requests to further how FFMPEG can be manipulated to alter the output clips:

1. I would prefer to always utilise "-c copy" so as to avoid an extra transcoding step. I have altered the block at line 510 in videowriter.py myself to this:

ff_cmd: list[str] = [ config.ffmpeg_path, '-ss', f"{K:.3f}", '-i', input_path, '-ss', f"{(S - K):.3f}", '-t', f"{(E - S):.3f}", '-c', 'copy', '-y', video_path ]
and the block at 517 to this:

ff_cmd = [ config.ffmpeg_path, '-i', tmpf, '-ss', f"{start_ts}", '-t', f"{duration:.3f}", '-i', src, '-map 0', '-c', 'copy', '-an', '-y', video_path ]

...but this is probably not elegant and overrides any options file - for me I don't care as I will only ever want to copy the stream, never transcode/choose another codec.

2. Can a "padding" option be added to add a few seconds to either side of the detection window? Using the test video when I ran MetDetPy and passed the resulting detection JSON through ClipToolkit, the clip stopped with the meteor trail still in view - I would really like to be able to capture around the entire event including eventual disappearance of the trail. I don't know how to do this myself but I'm sure an alteration of the code for this should be straightforward?

3. Can the file extension be changed? I couldn't see how to do it from the documentation - I would prefer ".mp4" not ".avi".

Lastly, in my proposed automation workflow, my long 8-9hour nighttime 4K footage is first transcoded down to 1080p (or 720p - I have yet to experiment fully with which works best) which I will feed to MetDetPy, but I want to keep the original highest quality, then for ClipToolkit to create its clips from the 4K file, not the one which MetDetPy used...is this possible?

Whatever I do when I have a detections JSON file to feed to it, I cannot get it to also accept a DIFFERENT video filename/path - the only way is to change the name of the file in the JSON from what I could see - otherwise it always errors (saying "Json must be a list") if you try to supply a video filename in the command...

[Designerspr]

• Stream copy option: I agree this is a useful addition. Currently, the settings don't support "copy" as a codec, but we will add support for it. Please keep in mind that using -c copy can sometimes lead to inaccurate start and end points if they are far from keyframe positions.

• "Padding" option: A great suggestion. We will implement a padding feature soon.

• Changing file extension: You can already achieve this by passing the --suffix mp4 option. This will change the output file extension as expected.

• Detecting at 720p, clipping from 4K: As you mentioned, manually editing the video path under basic_info in the JSON file is the current workaround. Since ClipToolkit already supports multiple input methods, adding another video-specific argument might overcomplicate the interface. We will consider how to support this more elegantly in the future, but we don't plan to change this immediately.

P.S: Regarding the "Json must be a list" error: this occurs because when input a json file after a video filename, the json file is expected to be JSON array (or JSON array file). This is kept for compatibility (see Ordinary Mode).