Skip to content

H.264, H.265, etc. benchmarks? #65

Closed

Description

@dagoldman

I think it might be helpful to have benchmarks related to the input video compression standard. I believe it would be helpful for me. I am tempted to just use H.265+ but if I knew dvr-scan processed one minute of H.264 a lot faster than one minute of H.265+ I might use H.264 instead.

Most technologies have pros / cons. H.265 compresses a lot more than H.264. Ditto for H.265+ vs H.265. That’s an advantage. But is it possible dvr-scan is much slower to process a minute of H.265 than a minute of H.264? Are there any tests answering the question?

If not already dvr-scan benchmarks comparing H.264 / H.264+ / H.265 / H.265+, I would suggest something like the following: For each standard, make a typical surveillance video file. Tabulate file size, minutes of video, time for dvr-scan to process file, output avi file size.

I understand each input video file is somewhat different. Different frames per second. Different scene. Different activity. So perhaps no “typical” file. And dvr-scan run time depends on the cpu. But I think a comparison might still be useful, because we don’t care so much about a 10% difference or something like that. Instead, we want a much larger performance difference, such as 100%. The benchmark we are looking for, I believe, is: # of seconds dvr-scan takes to process one minute (or other interval) of video

I may be totally off. I am ready to be corrected. But in general, benchmarking is helpful for many purposes. And especially for this kind of long-running application.

Daniel

Activity

  1. Breakthrough commented on Dec 12, 2021

    @Breakthrough
    Owner

    While this is technically possible, the scope of this is pretty big to be honest, and depends quite a bit on your particular system, operating system, exact version of libraries being used, etc. I'm not sure I know enough about all the variables involved to back any claims like that in general.

    I agree such a comparison is very useful, but I think you could perform such a benchmark for your own use case, with your own hardware, and get much more accurate and relevant numbers. DVR-Scan already outputs the average processing speed in FPS, so you could just run it a few times on various types of videos to see if any work particularly better.

    If you or anyone else do end up running such tests please, do post the results here along with your system specs/library versions.

  2. Repository owner locked and limited conversation to collaborators on Dec 12, 2021
  3. converted this issue into a discussion #69 on Dec 12, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions