[FFmpeg-devel] GSoC 2018
thilo.borgmann at mail.de
Fri Jan 26 19:27:22 EET 2018
Am 26.01.18 um 18:15 schrieb Pedro Arthur:
> 2018-01-26 14:26 GMT-02:00 Rostislav Pehlivanov <atomnuker at gmail.com>:
>> I think actually writing and reading papers for the filter is pointless -
>> RAVU already exists in shader form, its well known in the community and
>> we'll essentially get it for free once the Vulkan backend is complete.
>> What the task should be would be to make a CPU-only version of the shaders.
> I didn't know about these filters, I did a quick search and found a github
> repo "mvp-prefilters" but there isn't any documentation explaining the
> internals of it.
> Can you point me some resources where I can find for example which models
> are they using, motivation and maybe some benchmark with other models?
> My initial idea was that the student would evaluate a few SR models, make a
> balance based on quality/performance and pick the best fit.
> If it's already done there is only the porting task, but I guess maybe
> there is newer and unevaluated models or we could improve something.
> Also the qualification task you just put in is already done - there's
>> already a filter to convolve 2 images.
> Should I revert the wiki? or you have an idea for a qualification task?
No need to revert it. Just update the qual task one a better one is found.
More information about the ffmpeg-devel