Repository navigation
load-scenes: make faster and clearer #347
Description
Hi!
I've just tried out the new feature, #294
I'd like to suggest a couple of things how it could me made a bit better.
Problem/Use Case
First thing: way I see it, primary use case of "load-scenes" is to save time on detection when a single video must be processed several times. There are other use cases, of course, but for this one it doesn't do much good: detection via the list takes nearly as much time as without it.
Another thing is that currently I can't be certain if the script works with the list at all. At first I made a list without "-s" argument, then I read the document and added the argument, and didn't notice any difference.
Solutions
It would be great to make it so having the .csv allows to skip the detection process entirely (if that's even possible of course).
Also would be nice if the script informed that it currently uses the list instead of a detection from scratch.
Thank you for your amazing work!
Activity
Breakthrough commentedon Aug 22, 2023
We can definitely add some logging to make it more clear when a pre-existing scene list is being used. This should show up when you specify debug logging, but I agree it should always be logged to make it clear from the output what is going on.
Regarding performance, this is the main reason why loading statsfiles was deprecated - with the default settings and downscale factors, processing the video is usually pretty fast. However, there are certainly a lot of ways
load-scenescould probably be improved performance-wise, so happy to look more into that for the next release.Thanks for the feedback!
[cli] Ensure `load-scenes` can only be specified once
[cli] Rework `load-scenes` implementation
Breakthrough commentedon Mar 6, 2024
This is now fixed in
developand will be included in v0.6.3. Theload-scenescommand now skips detection and produces output much faster.[cli] Ensure `load-scenes` can only be specified once
[cli] Rework `load-scenes` implementation