[FFmpeg-devel] evaluating the experimental status of ffv1 version 3

[Peter B.]

I've now rewritten my test-script from scratch to exclude some
performance and bit-error-resilience tests, I've had included before.

Dave Rice and I have decided to use all the testvideos from "Delf's
Collection" at xiph.org [1]. This first series of tests now includes
only YUV, but in several resolutions (from QCIF to beyond HD - up to 4k)
and the following subsampling:
444, 422, 420

The tests are run with nested loops, encoding FFv1.3 using the following
parameters in all combinations to each other:

//------------
Contexts: 0 1
Coders: 0 1
Slices: 4 6 9 12 16 24 30
Slice CRCs: 0 1
GOP sizes: 1 10 25 50
//------------

Each input video has 224 corresponding output videos, for every one of
which a framemd5 video checksum file is created.
The videos are deleted right after the checksums were generated.
Their filesize and parameters are logged to a CSV file. One for each
input video.


The next tests will be with RGB input material. Using a sequence from
the Open Movie "Sintel" [2] and the SVT test sequence [3].
Additionally, multipass encoding will be tested separately, but I'm not
sure if it makes sense to do this for all videos, but rather with longer
ones from different input sources to get a more real-life feeling for
the results?
Michael, what is your opinion about that?

Results and script-code will be posted as soon as the first run with
<=SD material is finished.

Yours,
Pb

== References:
[1] http://media.xiph.org/video/derf/
[2] http://media.xiph.org/sintel/
[3] http://media.xiph.org/svt/