David, thanks for the reply.

On Thu, March 17, 2011 9:36:24 PM David Schleef wrote:
> Low delay works here, with both 4:2:2 and 4:4:4, using GStreamer:

Unfortunately GStreamer (from git) isn't close to configuring and
building under Cygwin...at least on my system.

Other than the flashing noise component (see previous email for sample)
ffmpeg 4:2:2 produces near-valid data.  PSNR measurments indicate that
4:4:4 luma is fine, but chroma is a smeared mess.  This may be the
"comet" artifact you mention.

> Possible reasons it's not working for you:  Orc is generating bad code
> (sometimes happens, although pretty infrequently these days).  You
> can test this by running the testsuite (make check), and/or running
> your code using ORC_CODE=backup in the environment.

make check output says "All 8 tests passed" however these two failures
were recorded:

cogorc_resample_horiz_1tap:
    backup function  :   PASSED
    compiled function:   COMPILE FAILED
cogorc_resample_horiz_2tap:
    backup function  :   PASSED
    compiled function:   COMPILE FAILED

Rerunning the ffmpeg tests with ORC_CODE=backup produces identical
output (i.e. the noise artifact is still present).

> Another possible reason, if you're seeing diagonal "comets", is that
> the transform-depth is changed away from the default.  This is a bad
> idea.

I have seen the "comets" effect appear at certain compression ratios
then not be there at slightly higher and lower ratios.  I haven't made
any explicit attempt to modify transform-depth.  The minimal changes
I've made are in libavcodec/libschroedingerenc.c libschroedinger_encode_init():

schro_encoder_setting_set_double(p_schro_params->encoder,
    "gop_structure", SCHRO_ENCODER_GOP_INTRA_ONLY);
schro_encoder_setting_set_double(p_schro_params->encoder,
    "enable_noarith", 1);
schro_encoder_setting_set_double(p_schro_params->encoder,
    "rate_control", SCHRO_ENCODER_RATE_CONTROL_LOW_DELAY);
schro_encoder_setting_set_double(p_schro_params->encoder,
    "bitrate", avccontext->bit_rate);

(I'm assuming the first two are redundant.)

> As a side note, low delay profile is very sensitive to the values of
> horiz-slices and vert-slices.  These should be factors of (width/16)
> and (height/8), which pretty much means 120 and 135.

The sample footage linked in my earlier email is 1920x1080.

Regards,

Stacey


      

------------------------------------------------------------------------------
Colocation vs. Managed Hosting
A question and answer guide to determining the best fit
for your organization - today and in the future.
http://p.sf.net/sfu/internap-sfd2d
_______________________________________________
Schrodinger-devel mailing list
Schrodinger-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/schrodinger-devel

Reply via email to