GitLab

Prevents default settings from overriding encoder options input by user

Prevents default settings from overriding encoder options input by user

Prevents default settings from overriding encoder options input by user

Prevents default settings from overriding encoder options input by user

rav1e binary was updated to request user confirmation before overwriting
ivf files. This caused rd_collect.sh to fail when existing ivfs are in
the target directory.

Adds the argument -y when invoking the encoder which causes rav1e to
function as it did prior to the update.

Certain Y4M files are detected by grep as binary which prints an error
message and incorrectly sets FRAMES=1. This patch forces binary files
to be interpreted as text by grep to compute correct number of frames.

Removes the argument --tile-columns=0 from the QSTR for both libaom and
libaom-rt targets as this is the default behavior.

We check that info1.par_n*info2.par_d == info2.par_n*info1.par_d,
 but even if that check passes, it is possible for, e.g.,
 info1.par_n to be info2.par_n*1000 (if the denominator is
 similarly scaled).
But then we were mixing info1.par_n with info2.par_d when computing
 the pixel aspect ratio.
So even when both streams have the same ratio, the result would be
 off by a factor of 1000 (in this example).

Instead, only use the PAR parameters from one stream, now that we
 have checked that they match.

Upgrading it from a warning.

We scale the Gaussian window according to the pixel aspect ratio in
 order to match the shape and size of the window used by the
 fixed-size test images in the original paper.
If the streams do not have the same aspect ratio, then there is no
 correct thing to do.

The Cr plane would not read in the correct data
on high bit depth videos, causing an incorrect result.
This commit fixes the input offset so the correct
values are read.

Tracks user time within rd_collect sub files, which then prints time
in column 4 of the generated out files. Modified bd_rate.sh,
matlab/bd_rate.m, rd_average.sh, and rd_plot.sh to support out files
including time in column 4.

This is done so we can apply different QPs to each of these encoder
 configurations for use with parallelized sub script.

When collecting .out files after running with various QPS, the gather
 step was incorrectly pulling in runs from other codecs.
This fixes a bug that was introduced in b0c158b7.
Signed-off-by: Nathan E. Egge <negge@dgql.org>

This also renamed rd_collect_aom.sh to rd_collect_libaom.sh.
Signed-off-by: Nathan E. Egge <negge@dgql.org>

The aomenc and aomdec binaries only have one codec implementation: av1.
Signed-off-by: Nathan E. Egge <negge@dgql.org>

Signed-off-by: Nathan E. Egge <negge@dgql.org>

Signed-off-by: Nathan E. Egge <negge@dgql.org>

Signed-off-by: Nathan E. Egge <negge@dgql.org>

Signed-off-by: Nathan E. Egge <negge@dgql.org>

The existing $AOMENC line had --cpu-used=0 in the $QSTR which came after
$EXTRA_OPTS and had the effect of canceling any user specified value.
Signed-off-by: Nathan E. Egge <negge@dgql.org>

Signed-off-by: Nathan E. Egge <negge@dgql.org>

Signed-off-by: Nathan E. Egge <negge@dgql.org>

Regression from commit 5eb3f15c "Add missing file rate.c"

Empty input video files will not be processed in 'dump_video' and
an error message will be shown instead.

Added extra check in the function 'daala_decode_img_out'
to make sure that the passed pointers are not NULL.

rd_collect will ignore CIEDE2000 when the python packages are missing.
An error message is shown to the user indicating which dependencies are
required.