Hi Yushin
On 20/10/15 18:47, Yushin Cho wrote:
Hi Thomas,
I have some questions.
1) In subsection "3.2. Motion estimation process",
"Since F does not exist the motion estimation process consists of
matching blocks B+mv0 in R0 with blocks B+mv1 in R1."
Does 'B' there probably mean each block in frame F? It was not defined anywhere
Yes.
.
2) Once interpolated reference frame F is obtained, which I understand will
spend some bits
unless all 16x16 blocks can be skipped, a regular motion estimation on top of
frame F
with target original input frame is performed?
There are no bits spent on the interpolated reference frame. Yes, a
regular ME is done after that. In our code the two MEs are different,
but really the conventional ME should make use of candidates derived
from the interpolation.
As implied by "The aim is to create a reference frame that is temporally
co-located with
the current frame being predicted," in "1. Introduction".
3) Lastly, is the interpolated reference frame F only used for B-frame,
since frame F itself is anyway motion compensated bi-directionally,
i.e. past and future frames (or the other way), R0 and R1,
though the current frame can be coded as a P-frame style once F is provided?
Yes, I've tried it and it doesn't give gains. I think this is because
the interpolation is "pull" from references rather than "push" + infill.
"Pull" has the advantage that there is no infilling to do , and for the
bidirectional case it's usually not desirable as good references can be
found on both sides of a motion discontinuity. But with 1-sided
interpolation i.e. extrapolation, at motion discontinuities there is no
information for revealed areas, so some kind of infilling is needed.
Hence moving edges tend to be truncated with this method. Perhaps some
of the Daala inpainting work could be applied in this context.
best regards
Thomas
Thanks,
Yushin
On Tue, Oct 20, 2015 at 12:46 AM, Thomas Davies (thdavies)
<[email protected] <mailto:[email protected]>> wrote:
I submitted the following draft which describes a method of
synthesising new reference frames for video coding, and how that
is done in Thor.
Regards
Thomas
Begin forwarded message:
*From:* <[email protected] <mailto:[email protected]>>
*Date:* 19 October 2015 17:12:42 BST
*To:* Thomas Davies <[email protected]
<mailto:[email protected]>>, Thomas Davies <[email protected]
<mailto:[email protected]>>
*Subject:* *New Version Notification for
draft-davies-netvc-irfvc-00.txt*
A new version of I-D, draft-davies-netvc-irfvc-00.txt
has been successfully submitted by Thomas Davies and posted to the
IETF repository.
Name: draft-davies-netvc-irfvc
Revision: 00
Title: Interpolated reference frames for video coding
Document date: 2015-10-19
Group: Individual Submission
Pages: 8
URL:
https://www.ietf.org/internet-drafts/draft-davies-netvc-irfvc-00.txt
Status: https://datatracker.ietf.org/doc/draft-davies-netvc-irfvc/
Htmlized: https://tools.ietf.org/html/draft-davies-netvc-irfvc-00
Abstract:
This document describes the use of interpolated reference frames in
video coding in general, and in the Thor video codec in particular.
Please note that it may take a couple of minutes from the time of
submission
until the htmlized version and diff are available at
tools.ietf.org <http://tools.ietf.org>.
The IETF Secretariat
_______________________________________________
video-codec mailing list
[email protected] <mailto:[email protected]>
https://www.ietf.org/mailman/listinfo/video-codec
--
Thanks!
Yushin
_______________________________________________
video-codec mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/video-codec
