[FFmpeg-devel] [PATCH] avfilter: add vmafmotion filter

[Ronald S. Bultje]

Hi,

On Fri, Sep 29, 2017 at 9:39 AM, Ronald S. Bultje <rsbultje at gmail.com>
wrote:

> Hi,
>
> On Fri, Sep 15, 2017 at 4:47 PM, Ashish Pratap Singh <ashk43712 at gmail.com>
> wrote:
>
>> From: Ashish Singh <ashk43712 at gmail.com>
>>
>> Hi, this patch addresses the previous issues and changes it to a single
>> input filter.
>>
>> Signed-off-by: Ashish Singh <ashk43712 at gmail.com>
>> ---
>>  Changelog                   |   1 +
>>  doc/filters.texi            |  14 ++
>>  libavfilter/Makefile        |   1 +
>>  libavfilter/allfilters.c    |   1 +
>>  libavfilter/vf_vmafmotion.c | 325 ++++++++++++++++++++++++++++++
>> ++++++++++++++
>>  libavfilter/vmaf_motion.h   |  58 ++++++++
>>  6 files changed, 400 insertions(+)
>>  create mode 100644 libavfilter/vf_vmafmotion.c
>>  create mode 100644 libavfilter/vmaf_motion.h
>
>
> I made tiny changes to this patch (see attached patch). Basically, it adds
> an option (similar to psnr/ssim filters) to log the per-frame scores to a
> file, which can be convenient for debugging the per-frame scores, or using
> them externally using CLI. Secondly, it adjusts the filter/sad so that the
> calculations use a slightly higher internal depth for the fixed-point
> integer calculations. This makes the score slightly closer to the
> floating-point equivalent in the reference code. Lastly, it normalizes the
> 10-bit score back to 8-bit, which is what the reference code does also. It
> also makes use of the vmafdsp.sad() function pointer instead of calling
> image_sad() directly, so SIMD will be used when added later on.
>
> This should address all review comments (the rest were basically things
> that will be used in subsequent patches, so whether we address it or not
> doesn't really affect the final product), so if there's no further
> comments, I'll squash the two patches together and push it.
>

And ... pushed.

Ronald