0:00:13yeah
0:00:14thank
0:00:23uh morning
0:00:24uh to they are like to send a paper
0:00:27a stretching
0:00:28an you're in this party
0:00:30compensated prediction
0:00:32and he's for multi view video coding
0:00:35so um
0:00:39he's the content
0:00:40or
0:00:40uh my presentation
0:00:42uh uh make a joint action of the disparity compensated in section
0:00:46especially that's the techniques used in the
0:00:48curve
0:00:49uh multi we the coding standard
0:00:52and
0:00:52you propose a lost yeah
0:00:55i forty for S E S
0:00:57for shot
0:00:58for the S yes six technique
0:01:00and the related uh consideration
0:01:02and some experimental sound would be given
0:01:05and
0:01:06the conclusion
0:01:07joint final
0:01:08so um
0:01:10P is the introduction
0:01:12for the current uh a multi view video coding
0:01:15a a standard
0:01:16that at to
0:01:18it is then that they exploit
0:01:20correlation between the real
0:01:22okay
0:01:22as a this fight
0:01:24previous uh
0:01:25H
0:01:26so um
0:01:28it is a a a a a a a extension of the explore
0:01:32as one a V C
0:01:33so uh it is
0:01:35is this
0:01:36uh
0:01:38in and
0:01:40so so
0:01:41the that this party
0:01:43uh information
0:01:44recall or that by the disparity compensated prediction
0:01:47because uh
0:01:49this technique
0:01:50station
0:01:50uh exposed to
0:01:52what you see they
0:01:53don't
0:01:53same coding instead
0:01:55all
0:01:55spot to the cool
0:01:56so
0:01:57um
0:01:59also so
0:02:00the the the coding to is just the same as they used in the
0:02:04and C P
0:02:06so um
0:02:08so uh if they use the same set of coding tools
0:02:11a they
0:02:13the only sport the pure translation uh um most a back to this party act
0:02:19so it is a broad based um
0:02:22um
0:02:23motion compensated prediction technique
0:02:27so that this party back to K
0:02:29send the
0:02:30the should all the blocks
0:02:33where
0:02:34uh but i
0:02:35not indicate the act
0:02:37uh
0:02:39a to that
0:02:45so um
0:02:46i'll met what is to put holes a a a technique
0:02:49to improve the this
0:02:51particle can see text
0:02:52for the and B C
0:02:54ah
0:02:55and the effect that um
0:02:57oh we want to cheap
0:02:58is to um
0:03:00buy by a a a a a new technique
0:03:02a proposed the course of a sub sample the block matching
0:03:05it does not imply a uh
0:03:07additional memory for
0:03:09storing the test
0:03:10additional
0:03:11we're frames
0:03:12and
0:03:13it
0:03:14you know
0:03:14need to perform a this to though uh and the polish
0:03:19so um all propose at any will be uh
0:03:22able to integrate
0:03:24to the actual to exploit
0:03:25is it
0:03:27so uh the S
0:03:28talk out the um a the
0:03:31this party fact
0:03:33uh first
0:03:34for four we propose
0:03:36that
0:03:37so
0:03:40a a uh i here we show a a a a a real world example or all a a a
0:03:44this party
0:03:45that
0:03:46it is a a real picture taken from the
0:03:48i
0:03:49a still can
0:03:50two can
0:03:52so it is obvious that um
0:03:54the the difference
0:03:55between the frames
0:03:57a are not just
0:03:58translation
0:03:59okay that's that
0:04:00uh
0:04:00to what you close to
0:04:03so for the uh
0:04:05oh example there is a wall
0:04:07uh uh on the on the right side
0:04:09and in many find that we find that uh
0:04:12for the last one
0:04:13but
0:04:14oh we use this one as the reference
0:04:16you can see that a on the wall the plate on the war
0:04:19it's up here to be a can
0:04:21press
0:04:22in the right view
0:04:24but
0:04:25so we find that a a there is a
0:04:27a compression in that
0:04:29if you a um not turning around
0:04:31you will be a stretching that
0:04:33so i here use the a vertical option
0:04:36for and and example
0:04:38uh
0:04:39for example in the C
0:04:41where some some horizontal objects
0:04:44on a if you look at two
0:04:46let me
0:04:47you can of fun this
0:04:49the texture on the
0:04:50see
0:04:51and
0:04:51you a that a a on the right real the the texture it uh appear to be
0:04:56uh a two a a i
0:04:59um um we said to have it is a a a a a a
0:05:02a year
0:05:04fact
0:05:04although though um
0:05:05you may not be a a a a um
0:05:07oh perfectly uh a processing uh more model that but
0:05:11we believe that um
0:05:13uh using the sheer in if can cross to me this that
0:05:17in this case
0:05:18so
0:05:20um
0:05:21so he he is the the uh uh uh S A six if that are we we want to a
0:05:25sprawl
0:05:26so it should be noted that a a a you the camera
0:05:30set
0:05:30this up and down
0:05:32is that the uh uh that right
0:05:34the the S S six if should be horse so yeah for
0:05:38should be a vertical instead or something
0:05:42so
0:05:43oh how do we provide a um a the S E X fig
0:05:46if fact
0:05:47and the uh uh this partly compensated prediction
0:05:50what in into what you
0:05:51i i that
0:05:53we can prove i
0:05:54a some uh uh are transformed the frame
0:05:57vol
0:05:58for the the the
0:06:00this part to text
0:06:04a a for example we can provide a some uh a she had that
0:06:07a a frame
0:06:08and also uh
0:06:10a strange uh a stretch and compress the frame
0:06:14and we
0:06:14can make use of these and frame to at all different uh uh i is sick
0:06:19fact
0:06:20to to improve the that
0:06:22the this on the
0:06:24text
0:06:26but
0:06:26ah
0:06:28uh this
0:06:29uh a what if that we we'll have a a a a a very serious uh
0:06:33but not for
0:06:34it major problem
0:06:36uh
0:06:37for for practical use
0:06:40because uh and this frame based on require a some um memory space
0:06:45uh a in the and code
0:06:47because you you need to provide a
0:06:50a a number all
0:06:51reference frame
0:06:53a a for you can you can
0:06:55for for the problem actually has one
0:06:58so
0:06:59um the number uh the
0:07:01then we we we'll be role
0:07:03a a a a a with the number all
0:07:05it's is actual
0:07:06uh can
0:07:07provide
0:07:08and also
0:07:09uh the the computing line and for providing this uh S A six frames
0:07:14is also a complex because you be two
0:07:17uh a a you the poll at of the reference frame that
0:07:21performing transformation
0:07:23for for the S is section five
0:07:25the frame based approach is not
0:07:27right
0:07:29so
0:07:30in this research we will
0:07:32uh a post match what
0:07:34uh
0:07:35to achieve the S A six
0:07:37this C P
0:07:38i i a new technology at a new to a uh and i call
0:07:42a a sample the block matching
0:07:44that's i set it to you know it does not require additional memory and
0:07:49additional look
0:07:50the
0:07:52so um
0:07:55uh i here is all on
0:07:58the S A six this if you
0:08:01a a in the extra two six for U V C
0:08:04are we have a a what a pixel uh a of us to use it
0:08:09and
0:08:10a in a typical uh echo that
0:08:12we will have a uh some four times in the for it to that and frame
0:08:17for for as to um
0:08:19form the problem matching
0:08:21and and uh the
0:08:22and the uh motion compensated prediction works
0:08:27so um
0:08:30in the in the and C P and this if you process
0:08:33um
0:08:35uh the echo the we sub same the the frame
0:08:39to taking the uh uh
0:08:42problem
0:08:44in the original scale
0:08:45because you you you
0:08:47in in with that
0:08:48in four times
0:08:49and subset wave by what times
0:08:51so that the scale
0:08:53change
0:08:54but
0:08:54you can a the location
0:08:56the a pixel location
0:08:58so
0:08:59we define a
0:09:00oh we first define this uh uh is
0:09:03a a a subsampling sampling rate five
0:09:05question
0:09:06okay
0:09:07it should be
0:09:08for simple
0:09:10um
0:09:10to achieve the the S A six
0:09:12uh effect
0:09:13we can modify this um
0:09:16a subsampling sampling re
0:09:18for um
0:09:20for for for different probably test
0:09:22to achieve the if that you like to
0:09:24sport
0:09:26so
0:09:27yeah use the um
0:09:29um
0:09:29um stress
0:09:31and compression
0:09:32um
0:09:34uh
0:09:35if X that we we decide
0:09:38so that we use
0:09:39okay we use
0:09:40the the a that's same we we define by this
0:09:43creation
0:09:45but that and you can see that and the
0:09:47um top left corner
0:09:49all of the the the matrix data
0:09:52um
0:09:53this is and compression
0:09:55a a problem
0:09:57we pacing the original
0:09:59for
0:10:00part to
0:10:01i in the corner
0:10:03so i we will achieve a a a
0:10:06strange and compression horizontal horizontal
0:10:08for example we do uh we we this up to do the that the S C problem at a with
0:10:14three
0:10:14we you have a
0:10:15the
0:10:16the subsampling grid shown in
0:10:18uh the left of figure
0:10:20you can see that um this
0:10:22some and one where you all the horizontal september we all the
0:10:26all
0:10:27i the reference frame
0:10:28three
0:10:28we go
0:10:30you can do
0:10:33this uh pixels
0:10:35for example if we
0:10:37we we are subject to you of the S C problem to i five
0:10:41right you have a is trash problem
0:10:45so here is the
0:10:47the
0:10:47stress and press
0:10:50so for this we
0:10:52we um
0:10:53uh you use this
0:10:55uh are also send bring re
0:10:57well
0:10:57by a S problem to
0:11:00so
0:11:01a by such doing by one or a we will have a
0:11:05a a a a a a right view that
0:11:07uh problem
0:11:09for
0:11:10to
0:11:12the row by the the you if at the left
0:11:14that that if that we
0:11:16does
0:11:16such a use some on
0:11:18on minus one
0:11:20i
0:11:21this is uh
0:11:22example of the search
0:11:25of course a to achieve a more a stretch and compression
0:11:30it she hearing um
0:11:31if
0:11:32we can be more uh meet a in the
0:11:35S C A S
0:11:36from
0:11:40so so there's a how we uh cheap that S E is six five
0:11:44the uh C P
0:11:46oh uh you can uh we like to all would post some my experimental result
0:11:51i here is the that
0:11:53a general condition that
0:11:55can pick a the the
0:11:57uh
0:11:58in the be people in the
0:12:00G M Z
0:12:01seventeen
0:12:02encoder
0:12:03so
0:12:03there a high profile
0:12:05we use
0:12:06some a video sequence that is quite a used a
0:12:10paul room X that what's a right now
0:12:13so uh a a the
0:12:14prime time
0:12:15thus does and for your information
0:12:19so for the um
0:12:20for the
0:12:22prime we can we use the for our uh and also
0:12:26we use
0:12:26uh X
0:12:28where we use
0:12:28H
0:12:29it and
0:12:31session
0:12:31okay
0:12:32let
0:12:33we the horse on those us a ten or in way from one
0:12:36to not
0:12:38and
0:12:38X
0:12:39as she ring levels will be used at
0:12:41but minus four to four
0:12:44i i should of that uh our for is currently only have high
0:12:47P
0:12:49well for for the uh
0:12:53also so that uh um and a is on the uh uh coding efficiency
0:12:58we we also sell to set
0:13:00the and
0:13:01one is um with all you frames
0:13:03uh
0:13:04i in this set um
0:13:06the let
0:13:09and that sequence
0:13:10the base will will be a i frame
0:13:12and the right real is all P frame
0:13:15the P frame only um
0:13:17a predicted by
0:13:18uh this party it
0:13:21and i set is a a hierarchical be frame with uh
0:13:24save and B frames that is by company
0:13:27in
0:13:28in uh a new
0:13:30techniques
0:13:32uh this is
0:13:33used to uh explore the effect
0:13:36all the
0:13:37all our proposed idea
0:13:40in
0:13:40some practical group
0:13:43so for the
0:13:45at first
0:13:45set all the uh um
0:13:48uh encoding we sell
0:13:50um
0:13:51we find that uh we have uh improvement in all case
0:13:55and uh uh F H B that's and it's all one point six nine
0:14:01send
0:14:02so uh
0:14:04this is
0:14:05yeah that this one point
0:14:08no
0:14:09so
0:14:10vol
0:14:11the the the coding structure we've a seven hierarchical B thing
0:14:15we
0:14:17we found that uh we still have improvement in
0:14:19in
0:14:20all this case
0:14:21but um
0:14:22but the improvement is uh uh a little bits
0:14:25a
0:14:26i it is possible that a the improvement
0:14:29i do
0:14:29by was beeping because we did up i
0:14:32uh
0:14:33no one B frame
0:14:34but the F H B and
0:14:37is you are out of the
0:14:39so you want to two
0:14:40two one to five
0:14:42then
0:14:44so ah a
0:14:45a he is
0:14:47close and this
0:14:48oh
0:14:49take
0:14:50so that we have
0:14:51uh a proposed post that in that one two
0:14:54all
0:14:54to that that the the S A S six back
0:14:57for the disparity compensated prediction
0:15:00we use the subset for the wrong match
0:15:03technique
0:15:05this is a um that's yes S brought i'll taking the from the interpolated frame
0:15:10so that we don the two
0:15:12two
0:15:14and and the memory and
0:15:16notation full
0:15:18uh taking in
0:15:19uh
0:15:20blocks
0:15:21and somehow T
0:15:22spot
0:15:24so a all for future work
0:15:27we find more um
0:15:29a and we we
0:15:30for
0:15:31both uh D C P A and C P
0:15:33for for or five to with the the proof
0:15:36the um
0:15:37formant
0:15:38a a a a a general video you to
0:15:41and
0:15:42you also find some possible that you know were from
0:15:45subsampled sample you
0:15:48oh
0:15:49this is
0:15:50one one
0:15:50is that they
0:15:51think the very much
0:15:59we
0:16:00maybe one or two
0:16:01which
0:16:08yeah why
0:16:09but that that is the most for things you already an estimation
0:16:14on
0:16:15as you a in our analysis is found that the p-frame frame use of the P right
0:16:20we
0:16:21and use only a uh in the in the and
0:16:24at the and we C and a moment
0:16:26the P frame is used to be uh predicted by
0:16:29the uh i i i I detection from the same view
0:16:33it is usually a out there are very uh well to
0:16:36so so the room for whole
0:16:38using the in the in the we detection may not be very very good
0:16:43and is not use that and i do you
0:16:45yeah use the don't frames
0:16:47of course not
0:16:48that i frame is
0:16:50in
0:16:50i intra prediction
0:16:52because you produce
0:16:54even though it terms of motion
0:16:56still do in did you
0:16:58all
0:17:02the intra frame
0:17:03we not use of
0:17:05that's the ah
0:17:06and i'm not that from but the standard itself
0:17:08but
0:17:10five a okay i guess it's a but i think is just a meeting a homography
0:17:17i as opposed to sharing
0:17:19yeah estimation
0:17:20actually is a a a and that what is
0:17:23try to provide some
0:17:24uh approximation all this uh
0:17:27a so called more to try formation by a very simple way
0:17:32so so
0:17:34that the if at all in the the transmission
0:17:37may not be uh uh
0:17:40or a to the uh
0:17:42uh uh generated
0:17:44but
0:17:44would be that that uh this is a a quite good
0:17:47approximation
0:17:49and questions
0:17:52i