[FFmpeg-devel] FFv1.3: Final testing - How to proceed?

[Michael Niedermayer]

On Wed, Oct 17, 2012 at 05:01:00PM +0200, Peter B. wrote:
> On 10/17/2012 05:06 AM, Peter B. wrote:
> > On 10/17/2012 03:16 AM, Michael Niedermayer wrote:
> >> On Tue, Oct 16, 2012 at 08:52:58PM +0000, Carl Eugen Hoyos wrote:
> >>> Peter B. <pb <at> das-werkstatt.com> writes:
> >>>
> >>>> I'm running FFv1.3 tests on the SGI image files (RGB48) from the SVT
> >>>> collection [1], and as FFv1 doesn't support rgb48be at the moment, I've
> >>>> tried converting it to e.g. rgb0.
> >>>>
> >>>> When converting the SGI image sequence from SVT to FFv1.3, only the left
> >>>> half of the image is actually displayed when watching the video using
> >>>> ffplay. The size of the shown video depends on how many slices are used.
> >>> The problem is reproducible with the following command:
> >>> $ ./ffmpeg -i tests/lena.pnm -pix_fmt rgb0 -vcodec ffv1 -level 3 
> >>> -strict -2 test.avi
> >> fixed
> >>
> > Confirmed for bgr0 files.
> > Running regression tests on SVT's 576i25 material now.
> Testsuite finished for SVT 576i25 SGI-Material converted to different
> colorspaces.
> All framemd5 check results are good:
> 
> //------------------------
> '1_CrowdRun_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_bgr0': 112 files
> checked
> '1_CrowdRun_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv420p': 112
> files checked
> '1_CrowdRun_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv422p16': 56
> files checked
> '1_CrowdRun_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv422p': 112
> files checked
> '1_CrowdRun_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv444p16': 56
> files checked
> '1_CrowdRun_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv444p': 112
> files checked
> '2_ParkJoy_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_bgr0': 112 files
> checked
> '2_ParkJoy_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv420p': 112
> files checked
> '2_ParkJoy_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv422p16': 56
> files checked
> '2_ParkJoy_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv422p': 112
> files checked
> '2_ParkJoy_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv444p16': 56
> files checked
> '2_ParkJoy_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv444p': 112
> files checked
> '3_DucksTakeOff_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_bgr0': 112
> files checked
> '3_DucksTakeOff_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv420p':
> 112 files checked
> '3_DucksTakeOff_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv422p16':
> 56 files checked
> '3_DucksTakeOff_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv422p':
> 112 files checked
> '3_DucksTakeOff_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv444p16':
> 56 files checked
> '3_DucksTakeOff_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv444p':
> 112 files checked
> '4_InToTree_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_bgr0': 112 files
> checked
> '4_InToTree_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv420p': 112
> files checked
> '4_InToTree_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv422p16': 56
> files checked
> '4_InToTree_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv422p': 112
> files checked
> '4_InToTree_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv444p16': 56
> files checked
> '4_InToTree_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv444p': 112
> files checked
> '5_OldTownCross_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_bgr0': 112
> files checked
> '5_OldTownCross_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv420p':
> 112 files checked
> '5_OldTownCross_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv422p16':
> 56 files checked
> '5_OldTownCross_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv422p':
> 112 files checked
> '5_OldTownCross_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv444p16':
> 56 files checked
> '5_OldTownCross_576i25_CgrLevels_SINC_FILTER_SVTdec05_-ffv1_yuv444p':
> 112 files checked
> //------------------------
> 
> Which RGB colorspaces (except bgr0) are supported by FFv1.3 and should
> be checked?

the list of supported pixel formats is in ffv1*.c
grep for PIX_FMT


> For testing pix_fmts with alpha-channel, I hope to be able to generate
> input-test content using FFmpeg, applying some non-empty but bogus alpha
> mask.
> 
> After these RGB and >8 bit YUV tests, the bitstream itself seems
> reliable to be be used for these colorspaces tested.
>

> @Michael:
> Except for the multipass tests, which tests are still open from your
> point of view, before marking FFv1.3 as non-experimental?


I need some statistics about the effects of various things. GOP,
multipass, slice counts, ...
I need them to decide on defaults and to know if ffv1.3 can/need/should
be changed radically or if the current design is fine.
one example of how this should look is for example:

                    v1  v3 4slices v3 16slices v3 context1  2pass context=1 ... average worst best
thisrgbvideo        +1% +2% +0.2%  -0.5% ...                                    +1.2%
thisyuv444video      ...
thisyuv420video
thishighmotionvideo
thislowmotionvideo
average
worst
best

the values in the table cells would here then be the filesize
difference of gop=1 vs gop=12 for example

based on this we can then see what effect the gop size has, if there
are some unexpected behaviors, ...

thanks

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

If you think the mosad wants you dead since a long time then you are either
wrong or dead since a long time.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20121022/88791c81/attachment.asc>