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

Peter B. pb at das-werkstatt.com
Tue Sep 25 10:50:25 CEST 2012

Quoting Michael Niedermayer <michaelni at gmx.at>:

> On Tue, Sep 25, 2012 at 02:26:58AM +0200, Peter B. wrote:
>> On 09/24/2012 05:37 AM, Michael Niedermayer wrote:
>> > should be fixed with the patchset i just posted
>> It actually is:
>> I've ran the tests with all sub-SD material from Delf's collection - and
>> every single one of them is correct now. :)
> ok, ill apply the patchset in a moment
> thanks

Thank YOU! :)

My TODO list regarding further FFV1.3 tests:

1) === Reliability tests:
These tests are to check if things work as they should. The tests done  
now were exactly those.

I think they should be done for:
   a) YUV: 411?, 420, 422, 444 - 8,9,10,16 bits
   b) RGB: Up to the bit-depth currently supported by FFv1.3 (Georg  
Lippitsch added >8bit RGB support, right? However. I don't know if it  
works up to 16bits already)

The sub-SD material from Derf's collection only contained the pix_fmts:
gray, yuv420p, yuv422p

2) === Compression tests:
These tests are to get numbers of the compression performance of  
FFV1.3 with different material and parameters.

   a) GOP size
   b) Slices
   c) Multipass: stats from same video / stats from other video
   d) Coder/Context
   e) CRC on/off

3) === Benchmarking tests:
These tests should give information about the performance of FFv1.3  
(in fps) with different material and parameters.

Benchmarking is tricky by nature ;)
Factors like CPU, disks used (SATA, SSD), etc. have a great influence  
and therefore have to be (a) chosen wisely and (b) be documented  

Why benchmarking?

For me personally, the main reason for doing benchmarking on FFv1 is  
to get an idea of "what is possible" in realtime, as we're using it to  
capture PAL yuv422p 8bit live, but e.g. would want to do >8bits in the  
near future, too.

Another reason would be to be able to playback FFv1 material fluently  
- even HD and beyond.

I have downloaded *all* videos from Derf's collection (SD and above),  
but HD-tests will (of course) take incredibly longer, compared to the  
sub-SD tests before.
So I wanted to ask your feedback about which tests to run and in which  
order, so we can do first things first.

I'll start the reliability testsuite on the HD-YUV material now,  
because it will have to be done anyway.

Thank you very much,

More information about the ffmpeg-devel mailing list