[FFmpeg-devel] [GSOC 2016] Ideas on FFV1 P frame project

[Michael Niedermayer]

Hi

On Sun, May 15, 2016 at 07:44:11PM +0300, Станислав Долганов wrote:
> Hello,
> I'm currently working on GSOC 2016 project for upgrading FFV1 codec with an
> ability to use P frames. Maybe some of you have any suggestions or
> recommendations.
> 
> While working on qualification task I implemented a simple modification of
> codec to encode P frames without motion compensation, so now the base for
> trying several MC approaches is ready.
> 
> My plan of upgrade is the next:
> 1) I want to try current realization of OBMC algorithm from snow*.c
> 2) Then try to take some ideas from HEVC
>   2.1) In paper "Overview of the High Efficiency Video Coding (HEVC)
> Standard" (can be found here
> http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=6316136) motion
> compensation process is described. It includes quarter pixel interpolation
> technic, intra- and inter-frame predicting strategies, quadtree of frame
> blocks, several types of predicted units, so there is a lot of stuff to try
> next.

good ideas

it also probably makes sense to try 1/8 pel and smaller blocks at some
later point if time remains


>   2.2) In "PARALLEL AMVP CANDIDATE LIST CONSTRUCTION FOR HEVC" (
> https://www.researchgate.net/profile/Liang_Zhao57/publication/261497818_Parallel_AMVP_candidate_list_construction_for_HEVC/links/54117bd10cf264cee28b3964.pdf)
> can be found algorithm to
>   make MC process faster.

It probably makes sense to first identify which parts take most time
and cause parallelization bottlenecks before optimizing


>   2.3) Authors of "HEVC Lossless Coding and Improvements" (
> http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.352.3447&rep=rep1&type=pdf)
> proposed Sample-Based Angular Intra Prediction algrorithm special for
> lossless compression, which can be tried too.

yes


> 3) Try to use filtration just like bilateral in "SEPARABLE BILATERAL
> FILTERING FOR FAST VIDEO PREPROCESSING" (
> http://elynxsdk.free.fr/ext-docs/Bilateral/ICME2005_TPLV.pdf) as
> pre-processing part, it can reduce compression size due to encoding noise
> values only from a current frame.

> 4) Maybe try some non common ideas like not use motion vectors as in
> "Lossless Video Sequence Compression Using Adaptive Prediction" (
> https://www.researchgate.net/profile/K_Sayood/publication/51386836_Lossless_Video_Sequence_Compression_Using_Adaptive_Prediction/links/5489cf110cf225bf669c7487.pdf
> ).

this sounds very interresting

I think the general plan layed out by your mail above looks good!

Thanks

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

Many that live deserve death. And some that die deserve life. Can you give
it to them? Then do not be too eager to deal out death in judgement. For
even the very wise cannot see all ends. -- Gandalf
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 181 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20160516/3a3b35e2/attachment.sig>