Přepis řeči - Trends in Image, Video, and Multidimensional Signal Processing

0:00:13	and i guess
0:00:14	and if you take a P C such as the image we you and multi that processing P C
0:00:18	and that's a T C which just extremely brought
0:00:21	with a way to have a set of interest
0:00:23	so trying to some of
0:00:24	what are the trends in such a broad yeah
0:00:27	wooden in the course of how off uh an extremely challenging problem
0:00:31	that's a of need to find somebody was keep of doing it
0:00:35	so instead what we but to do is try and some i
0:00:39	yeah
0:00:39	oh yeah i
0:00:41	i i
0:00:42	i
0:00:43	i trained but but it and some i some of the specific plans in usually yes
0:00:48	and that about those and hopefully i time for discussion and
0:00:51	but the sitting at that end
0:00:53	okay
0:00:53	so we have to
0:00:55	experts will be speaking on behalf of or specific i you the the D C obviously they cannot are present
0:01:00	everything than the dc but hopefully present option but the D C
0:01:04	that's
0:01:05	somehow like
0:01:06	that is going to be talking on the multi that signal processing side
0:01:10	a a to that is going to talking about
0:01:12	on the image and media processing site
0:01:14	okay
0:01:15	i i'm not of shown my the easy chair so i get to pay to moderate and they are we
0:01:19	here
0:01:20	if we begin i'd like to begin with some global trends that we see around us
0:01:24	i and you planning of a i incident crossing that scene which will really be more
0:01:29	inclusive
0:01:29	that what we can be in the course of a or not
0:01:32	on the application site we see it is a lot more on the web and this will site you but
0:01:37	i do things more more the cloud
0:01:39	that is people like to do all went to reality on the cell phones but using all augmentation of processing
0:01:45	that is
0:01:46	a big use of streaming we of this is already happening be her in the top
0:01:49	the only we yesterday
0:01:51	the video traffic
0:01:52	for the first time has exceeded
0:01:54	the that is that it can be a unit but
0:01:56	so that takes now is
0:01:58	that can what that it is net states
0:02:00	oh and wait
0:02:02	then people sharing
0:02:04	actually but it head on P to be in it was
0:02:08	to and T emitting of becoming more and more common place
0:02:11	that's what
0:02:11	this pushing on the consumer site
0:02:13	and on the cat just like you looking at computed as close to capture imaging
0:02:17	but other than have sense which that i got to the image you interested in you looking at how you
0:02:21	compute
0:02:22	what you interested in
0:02:24	on on the really have buttons techniques side
0:02:26	one of the biggest and which i C is that well
0:02:29	boundaries doing traditional disciplines have become increasingly blurred
0:02:33	you have
0:02:34	the separate disciplines we to think of a as an image processing community quite
0:02:37	separate from a number of other communities
0:02:39	to to the the all these you to have people to the whole boundaries between image processing computer which in
0:02:44	communications psychology machine learning any i
0:02:48	that be very hard to draw all and this is in addition to traditional disciplines do if you already what
0:02:52	so just mathematics and statistic
0:02:54	so with that but introduction
0:02:56	a a give the audio the bad take
0:02:58	we begin with talking about trends and money with signal processing
0:03:14	but that that take uh welcome everybody and things got for that uh and reduction i think um
0:03:19	i would agree with curves comments that uh
0:03:22	many any of the trends that we see a cost not just a
0:03:26	to to C but but the the T C spending the entire uh
0:03:30	signal processing society
0:03:31	then involve kind of convergence of what used to be separate disciplines and also um
0:03:36	pushing computation sort of further for back to the sensor
0:03:39	like a kind of convergence of sensing and computation
0:03:42	um so what i'm gonna focus on uh for the purpose of the next few minute
0:03:46	is the specifically multidimensional aspect of this
0:03:50	um and
0:03:51	so
0:03:52	a kind of a a a line take away from this talk is that
0:03:56	big data is finally here
0:03:58	the compute resources to work on big data are here
0:04:02	and and it's time for us as the signal processing side is to sort of get in the game
0:04:06	okay that's the story
0:04:07	a a and you can see that in the background of this interest slide is
0:04:11	a some representation of big data that all um
0:04:14	talk about a minute
0:04:16	um
0:04:17	so here's a here's a fun exercise uh to think about while i
0:04:22	a you a little bit about what i'm gonna say
0:04:24	a a a a a lot of
0:04:26	we or analogies and metaphors and the news and nine and recent times
0:04:31	and uh i've put a few of them on the slide
0:04:34	um so the question i would want close for you is
0:04:37	when it comes to a vast quantities of data
0:04:40	and our job a signal processors to extract from this vast quality of data
0:04:45	the relevant salient information
0:04:47	um i we swimming or we'd drowning
0:04:49	uh are we drinking or are we sinking
0:04:52	right
0:04:52	so
0:04:54	a can match these quotes swimming and sensors and rounding and data
0:04:58	a a that was a quote from a uh recently retired us air force general
0:05:02	it has basic complaint was that
0:05:04	every time another you a be flies with more cameras on it
0:05:09	we get more data but we don't necessarily get more information
0:05:13	a a a a global should of information was referenced by yesterday morning morning's point speaker
0:05:18	and uh the last quote of course is easy it's so it's a famous line from the
0:05:22	english class rhyme of the ancient mariner
0:05:25	and that we have to ask ourselves to really is this a case of what are are everyone already dropped
0:05:30	to drink
0:05:31	in other words how do we take is massive quantities of data
0:05:34	and exploit them and a useful way for various signal processing purposes
0:05:39	okay
0:05:39	so the other thing also just as might well it word about this
0:05:43	but cloud what is the cloud i meant to put in a a a graphic of a cloud but
0:05:47	you know is just sort of water vapour anyway or something right
0:05:50	a what what is the story so
0:05:53	but a it is kind of a a a convenient shorthand
0:05:56	for a data intensive computing
0:05:59	right
0:06:00	and a what that is to say is
0:06:02	how do we think about scaling up traditional signal processing algorithms
0:06:06	a single core or or given a set of of of of course a ranged into a cluster
0:06:11	to really touch data that's faster passing the terabytes K
0:06:15	okay
0:06:16	so you do need to know much more about cloud
0:06:18	but if you want to that just for a real uh earlier this week you will have seen someone from
0:06:22	phase book
0:06:23	uh telling you precisely how some of this exciting work is being done
0:06:28	so
0:06:28	that leads me to my next point which is evidence for travel
0:06:32	so it's fine for me to stand up here and talk about this but
0:06:35	like where is the real evidence
0:06:37	of this is an emerging trend
0:06:39	so if you take a look at icassp last year
0:06:42	we had a a
0:06:43	uh one tutorial on user dynamics of social networks we had actually another entire special issue devoted to social networking
0:06:50	and a special session on signal processing for graph
0:06:54	i
0:06:54	a me this is one manifestation of this kind of mad the trend of that i put on the top
0:06:59	of the slide here
0:07:00	which just say that as the data rates increase
0:07:03	and how to pace are ability to make sense of them
0:07:06	the the the the piece of at extracting low dimensional structure from high dimensional data becomes even more important
0:07:12	so whether you want to you know everything fits under this rubric back from compressed sensing
0:07:17	all the way back to a principal components analysis
0:07:20	just kind of nice because as car of mentioned
0:07:23	i this a fun is increasingly touching on psychology a
0:07:26	was the psychologist to came up with latent factor analysis in the first place right
0:07:30	so this year's icassp with see even more of the same
0:07:33	um several of the tutorials touched on big data
0:07:38	that we had yesterday planner you we have another platter E and bayesian nonparametric
0:07:42	a those are uh a a very nice and scalable class of techniques for treating very large got data corpora
0:07:49	um
0:07:50	and we have a special session this you a low dimensional structure and high dimensional data
0:07:54	um so my question for you is
0:07:56	what's twenty twelve and be on going to look like
0:07:59	are we gonna sink or swim or or are out it's up to us the data are already here the
0:08:03	computational resources are coming on line
0:08:06	and we have to ask ourselves how do we look at these trends that are taking place across disciplines
0:08:13	and
0:08:14	for some common signal processing framework around a
0:08:17	but that's my challenge to you
0:08:19	i
0:08:20	so
0:08:21	i thought about this problem uh for a long time and i watched
0:08:25	and try to take part of some of these changes have happened
0:08:29	and this is what i think is happening here's my assessment
0:08:32	a a a a a converging around
0:08:34	graphs graph representations as a kind of common framework
0:08:38	and there's to key points
0:08:40	the first is that a graph representations presentations are of various sort of a handy way to think about
0:08:46	data data that are very high dimensional but simultaneously very sparse
0:08:50	okay
0:08:50	so
0:08:51	covariance structures are or correlation amongst a protein expressions engine general
0:08:57	right
0:08:58	there were co occurrence uh
0:09:00	a and document
0:09:01	a
0:09:02	or
0:09:03	i i could and examples from the other technical committee speech processing for example et cetera
0:09:09	um
0:09:10	other kind of key point is that they give us a common framework into which we can hack
0:09:16	oh
0:09:16	sort of traditional structured data
0:09:19	signals images speech text on it so forth
0:09:22	and kind of unstructured structured things like documents
0:09:26	or are collections of
0:09:27	stop
0:09:28	like collections of multimedia data
0:09:31	um
0:09:31	all of this can kind of we put under this common framework so what i've shown at the bottom
0:09:36	is
0:09:36	then the box
0:09:38	the basic kind of duality between a writing down a picture of a graph and what you have no and
0:09:43	you have edges
0:09:44	and turning that into uh
0:09:46	a matrix variate structure
0:09:48	right so the nodes become rows and columns in a matrix and
0:09:51	the edges become a entry is nonzero entries and the adjacency structure
0:09:56	so
0:09:56	there is very little a new really new under the sun right
0:09:59	and and if you back up to the left and the right of this diagram you'll see this famous example
0:10:04	of on the rolling the swiss roll if you're remember this is a
0:10:07	and i can maps or or a
0:10:10	nonlinear pca example
0:10:12	then a which
0:10:13	this
0:10:13	no a graph representation plays a fundamental role
0:10:16	okay
0:10:17	so
0:10:19	taking a high dimensional nonlinear structure
0:10:22	like a plan that's been embedded in three dimensional space and on wrapping it
0:10:26	to something that's for a
0:10:29	this case requires building a graph sparsifying the graph on the set of points
0:10:34	computing shortest path across the graph
0:10:36	and then under wrapping this
0:10:38	up or all and to something flat
0:10:40	right
0:10:41	so
0:10:41	you can start with structured data and up but structured data but you may well ask their graphs and the
0:10:46	middle
0:10:47	or or you can talk about pure and structured data like collections of text
0:10:51	and documents and looking for co occurrences of words and other patterns
0:10:55	right
0:10:56	so
0:10:57	i
0:10:58	a strong evidence that
0:11:00	a key point of this framework is the it its ability to bring together structured and unstructured data with a
0:11:05	common framework
0:11:06	that's what makes this sort of very hand
0:11:08	once i turn "'em" my underlying objects and to graph representations
0:11:13	then i think compute with them according to various rules a linear algebra
0:11:17	and uh and the example that i shown here
0:11:20	or are dealing with a
0:11:22	with the images images from flicker curve
0:11:24	and a
0:11:25	face recognition apps are already being built
0:11:28	that leverage the graph structure and to these images of co occurrence and photographs right
0:11:34	is predictive of
0:11:35	that or uh social network and trees and some like face book
0:11:39	okay
0:11:40	and uh uh in it's it's clear i think to see how how extensible these things are
0:11:44	so for the reason cited by the various communities all the various communities many of which car mentioned
0:11:49	a a kind of quiet coalescing around this type of representation
0:11:53	right
0:11:53	so i'll leave you with some very basic challenge problems
0:11:57	a i think these are these are if not the big three than than three of the biggest certain my
0:12:03	so the first one is
0:12:05	the physics the phenomenology
0:12:08	we don't really understand very well yeah
0:12:11	but a massive graphs
0:12:13	how massive of behave what they look like
0:12:16	whether other social media data
0:12:18	or graphs derived from large corpora images are so on and so forth
0:12:22	we need to understand the physics
0:12:24	to think about it and signal processing language right
0:12:26	you can't do detection and estimation theory for radar less you understand
0:12:30	the physics involved that a radar system
0:12:32	then not argue that you can do
0:12:35	yeah a graph based signal processing and less you understand the driving phenomenology
0:12:39	right
0:12:40	so that leads the second point which is
0:12:42	so
0:12:43	basically we don't have
0:12:45	a so everything that you can answer
0:12:47	by looking in an undergraduate statistics or or in your level signal processing text
0:12:53	about
0:12:53	and matched filtering analysis of variance uh so on and so forth least squares
0:12:59	we don't know these things that program
0:13:01	i
0:13:02	there's not the same natural vector space structure as there is when you talk about
0:13:06	simple signals and images and euclidean space and that the euclidean space
0:13:10	and
0:13:10	so we need we need
0:13:12	we need the fundamental theories i mean
0:13:15	know when you look back at what the theory of information look like before shannon came along and unified it
0:13:21	that was sort of
0:13:23	spread out and do couple
0:13:25	i we need a sort of unifying theoretical under framework to understand
0:13:29	how these things uh a one of the fundamental them
0:13:33	of of of graph but based signal processing detection and estimation
0:13:37	um then than the last one is
0:13:40	if you step back and look
0:13:41	what we're really doing is were just adding this
0:13:44	correlates of context or the structural context
0:13:48	on top of
0:13:49	traditional signal processing cards that we are ready to large extent don't how to deal
0:13:54	and are getting together context and content i think is another
0:13:57	another another challenge for us
0:13:59	um
0:14:00	and
0:14:01	if you ask yourself well
0:14:03	but got a general the theory of signal processing for graphs and that even look like
0:14:07	the answers are really not clear but
0:14:09	i have left you with just a couple examples of the bottom
0:14:12	oh of the way is an a which
0:14:14	new were newly discovered mathematics can often kind of look behind the scenes on and and for quite a long
0:14:19	time and that suddenly it pops up and becomes L relevant
0:14:22	um um a couple of past examples boolean algebra
0:14:25	a a weighted a quite a long time before the advent of does a logic
0:14:29	that are and the matrix E which is uh another subject of a tutorial than the site
0:14:34	i i i have a huge impact and wireless communications of the late nineties and early two thousands
0:14:38	so it's an open question as to what mathematical advances
0:14:42	are going to drive signal processing frameworks for network data
0:14:46	and uh i heard you to a consider um taking up the challenge
0:14:50	uh that's it thanks very much and uh uh now i'm gonna
0:14:54	turn it over to car
0:14:57	i Q patrick
0:14:58	i
0:15:03	one we do questions would be to questions on to the end of the way would be people have a
0:15:07	in questions at this point
0:15:09	i could have some for back click now
0:15:11	and depending on how long things school we can decide where to cut think sure
0:15:15	so
0:15:16	a they
0:15:16	question
0:15:17	that people would like to ask
0:15:20	so maybe i'll begin with one
0:15:22	what
0:15:23	what you see in terms of education
0:15:25	and mentioned she didn't you can have been very
0:15:29	and just on the base eight
0:15:32	and
0:15:33	i
0:15:34	what is your view in terms of how do we teach
0:15:37	and educate
0:15:39	the next generation of these edges and students in the C
0:15:42	yeah
0:15:43	also so that that's a very good point uh
0:15:45	so the good news as as as that the data are Q here and the compute infrastructure is also coming
0:15:51	on and so probably
0:15:52	many of us a educational institutions have access to pretty good compute resources at this point
0:15:58	so i think the positives are that
0:16:00	we can get a our students access to these kind of data example are for really in the course of
0:16:06	their undergraduate education
0:16:08	um
0:16:09	the flip side and this is something that i didn't talk about all but is very interesting area to get
0:16:13	into
0:16:14	i issues about privacy and security
0:16:17	right so i mentioned things like i mentioned face but i could've mentioned to or so on and so forth
0:16:22	some of the most interesting social media data sets
0:16:25	oh also the same ones that
0:16:28	we need to be very careful about in terms of privacy concerns
0:16:32	and at the policy level and the united states at least the national science foundation has been very involved right
0:16:38	now in trying to figure out how to handle that
0:16:41	other words a the right a research grant application
0:16:44	to do signal processing education on some kind of corpus of face book data
0:16:49	no know is that all right should the and S even be funding that what are the human subjects requirements
0:16:55	to experiment with such data
0:16:57	and so on and so forth
0:16:58	for the moment all we've managed to do
0:17:01	as collected data
0:17:03	when people sort of have you have and their can then know that there
0:17:07	a a being collected on so for instance
0:17:10	uh a student a mighty business school
0:17:13	agree too
0:17:14	you know you cellular phones with the understanding that
0:17:17	uh
0:17:18	their proximity to other people on the study group would be measured and these data are actually available from mit
0:17:24	T
0:17:24	uh uh you can go on and use them they're called the reality mining data
0:17:28	right
0:17:28	so this is a very interesting but
0:17:31	my a us understand about better the phenomenology of the subset of people who choose to go to mit mighty
0:17:36	in the school but it's hard to know all right if that's the same thing as as the more general
0:17:40	population
0:17:42	so i think that's that's one of the big challenges is of is
0:17:45	there is a a real opportunity to get people exposed to these kind of data early and their education but
0:17:50	we have to work through some of these issues about
0:17:53	uh data privacy and uh you can read about this in the paper almost every day
0:17:58	thanks
0:17:59	thanks to
0:17:59	yeah you the question
0:18:01	for be
0:18:02	in
0:18:03	this one
0:18:04	oh right so then lena you
0:18:06	okay have something that here "'cause" i want to look at some okay paper thanks yeah
0:18:11	i i go go
0:18:13	going to talk about a three D you'll process
0:18:16	yeah also a long as then and of the back in explosion
0:18:20	um
0:18:21	so
0:18:22	so a of an application didn't and that i was to in the nineteen thirties black and white be and
0:18:28	then we and the to come section two
0:18:30	somewhat poor resolution to
0:18:32	it should be a solution from you each T for two for two D
0:18:37	and then also be used in
0:18:39	we we look at how experience
0:18:43	that we had a be used to have temporal spatial we have a lot of below
0:18:48	a a a a a a a a a a you know there is an had growing um
0:18:54	um
0:18:55	small was so that you know small the
0:19:00	that gives them what to you and mobility and dark you know
0:19:06	still have what can we do about that
0:19:08	i i'm still a typical but to two hours by there and with my
0:19:14	capture them out
0:19:17	T V oh i
0:19:20	and
0:19:21	i
0:19:24	would like to to me the and my and they're
0:19:27	um
0:19:28	and so and and the for user
0:19:32	can
0:19:34	sir
0:19:35	um
0:19:36	can have a problem because instrument
0:19:43	but can't have the van and a and and then so they have to a good cop
0:19:49	to give you my experience we have to capture a you know it's and have to step and can come
0:19:56	a of time
0:19:57	and the because you have some mismatch
0:20:01	and in my view
0:20:04	and
0:20:06	and
0:20:07	i
0:20:09	and
0:20:10	captures
0:20:12	a a a a a a a and that you a process and a good then
0:20:17	and compute and to go to do that we can start the main area
0:20:21	for example i would like to maybe can capture
0:20:27	that
0:20:27	and and from back
0:20:29	to to the the type of using them
0:20:32	um
0:20:32	but
0:20:34	can
0:20:35	and am so we had to go to is something with that it to plot we present position
0:20:41	but you for dimensions an email either few was or whatever information need
0:20:47	and um
0:20:48	hmmm
0:20:51	efficient solutions but the to the function solutions
0:20:54	for
0:20:57	sure
0:21:00	and
0:21:01	so
0:21:03	and i
0:21:06	we happen on a point
0:21:08	um
0:21:09	a the for example if is not good
0:21:13	a bit of the a right you know
0:21:15	um
0:21:16	and i
0:21:19	and if uh_huh channel
0:21:24	and
0:21:27	in the
0:21:29	hmmm
0:21:30	um
0:21:32	a
0:21:35	that
0:21:37	section for for example
0:21:39	the projection
0:21:41	and you have to
0:21:45	and the left you can be too
0:21:48	in in the head
0:21:50	but be it is score
0:21:53	and
0:21:55	uh_huh
0:21:55	to options that what i i've been to use
0:22:00	to
0:22:02	that
0:22:03	some some information
0:22:05	as in
0:22:06	is not
0:22:08	i
0:22:08	um
0:22:10	a similar to
0:22:15	i have a to show and
0:22:18	a i and a conventional two
0:22:20	how
0:22:21	the the of missed a and
0:22:24	do some special with
0:22:25	um but to be
0:22:27	that would we have to pair
0:22:28	and a few
0:22:29	so
0:22:30	since since pulp
0:22:34	and
0:22:35	to
0:22:41	time
0:22:42	to cook missed and but to five not sure
0:22:46	to and a factor so can we need a chance with limited resources and that a lot the band like
0:22:53	an X not doing so they want to the three D and not be do not to be
0:22:57	so some of the things that we need to on D to three time
0:23:03	test
0:23:04	and
0:23:06	uh_huh
0:23:06	if i say
0:23:08	so and so as if you like you
0:23:13	i i two to come for mentioned
0:23:16	when to the U
0:23:19	um
0:23:20	i
0:23:21	the complaints
0:23:24	i
0:23:25	the
0:23:26	content can you know
0:23:28	system passed of the content not of this in
0:23:34	um
0:23:35	if we look at that at the slides in
0:23:41	from which we can to that information of the three
0:23:47	imagine
0:23:48	and that we can look at to you are be captured for
0:23:56	oh
0:23:58	so you using uh collapse
0:23:59	and i'm i'm i'm not sure if you can see the the present
0:24:05	if you we have like three
0:24:08	and an structure from motion so try to that section of by motion for me D J like that look
0:24:14	like a shame
0:24:16	i
0:24:20	that
0:24:22	um occlusion Q
0:24:24	i yeah i don't or that's to the medic but you
0:24:31	and
0:24:33	just
0:24:35	and and the agenda the will come back to the question but
0:24:39	but not
0:24:40	i
0:24:41	but
0:24:43	i in to D two D to three D conversion
0:24:46	i that information and to be able
0:24:50	to to construct a a bunch of come here to i'm
0:24:56	i mentioned that if we have a you know a couple of view
0:25:01	to measure that and i had to men them and a global method
0:25:08	i could and but uh
0:25:10	i i'm and the globe in the back up to my
0:25:13	function of the was to compute the disparity you
0:25:16	yeah the local ones they more local but
0:25:20	i i you know are know that he
0:25:23	to to on the accuracy of
0:25:27	i exploit quite clear image
0:25:31	that's not
0:25:33	i
0:25:36	good
0:25:38	yeah
0:25:39	though
0:25:40	i the optimized in fact
0:25:42	however
0:25:43	i the trying to improve you the the either or the i Q
0:25:47	one i'm going to one by i go back to an from the G
0:25:52	to to the cup
0:25:54	the
0:25:55	the man
0:25:56	one at the cost at the
0:25:59	that's
0:25:59	construct a but in the and so do not computation
0:26:03	um
0:26:04	that used in the project
0:26:06	topic and i i mention the that we would like to do
0:26:11	and may be used
0:26:13	i i can you how we have that image but most of didn't that we use the information
0:26:21	to construction of the
0:26:23	that that that do you
0:26:25	a and that i'd that maybe other techniques one that using some image work
0:26:30	based
0:26:32	i a chance
0:26:34	um
0:26:34	we we have to the project that
0:26:37	just
0:26:38	how wear pink a function that we used to generate the views
0:26:42	um
0:26:44	come to stand i mean a simple not exploit to view
0:26:51	and we would like to the notion an efficient back in me to do that
0:26:55	and include include in the extension of the you statistics for a standard
0:27:00	could you could the to code our web the three D you have to form that we have to make
0:27:07	any compatible for my
0:27:10	to do this to do you think sequence of a black for like by that left for the that thing
0:27:16	you would just one sequence
0:27:18	that
0:27:20	or could use the frame compatible for way that you can put in a frame
0:27:26	i could but then
0:27:28	yeah
0:27:29	but the problem is to have to down sampling and not the solution and then when one a display not
0:27:33	you know a two point eight
0:27:35	and interpolation method
0:27:36	we can
0:27:37	and also
0:27:40	right to use the can in for the baseline we could
0:27:45	you used to take come a compact to perform
0:27:49	is just keep it
0:27:50	to get back to the original resolution interest in at six three curve
0:27:58	and
0:28:00	i
0:28:01	compensate for it
0:28:04	much
0:28:07	you
0:28:08	and to a good quality
0:28:10	can
0:28:11	to you to to
0:28:15	interesting
0:28:16	a and and also i that from you the two D yeah and these we can
0:28:20	the quite and um is it or or or you know be created about what to do and then a
0:28:28	form of you just a just cup
0:28:30	because case
0:28:32	and each you you meet condition
0:28:34	and you have you a quite understood to since the future can you you have used
0:28:40	use
0:28:40	is that that you might to
0:28:42	so
0:28:43	yeah information will now able to check the that had just
0:28:49	the
0:28:49	come
0:28:51	and
0:28:53	okay
0:28:56	uh_huh agenda
0:28:59	but
0:29:00	if
0:29:02	the content and
0:29:05	to theory
0:29:07	can i got to the house
0:29:09	and to be the quality content
0:29:13	and
0:29:14	i
0:29:15	the objective quality net fixed how to start with subset that the experiments to stand the major of three D
0:29:21	video quality and a combat
0:29:24	so we look at an option not
0:29:27	but but
0:29:30	i it out before we move into objective quite X
0:29:36	and and you can do if you are
0:29:39	come come
0:29:41	and
0:29:41	and
0:29:43	i in the past
0:29:45	should ensure on my team that's not going to be some to get on
0:29:50	i
0:29:51	but but you have to look for it
0:29:53	i
0:29:54	a sure to have a good subject score
0:30:01	our objective metrics we the base of that
0:30:05	known from find look okay
0:30:07	you know just
0:30:08	i back to the
0:30:10	the the not got that D was not good be able to fix
0:30:14	yeah have to be a given is to be a shame oh might not go to
0:30:18	i i mean either one and a
0:30:20	and had but for that i
0:30:23	and
0:30:24	email
0:30:26	kind show that was
0:30:29	to put the pen from an issue of can you
0:30:32	and also still the post but
0:30:35	and
0:30:36	this i to this that you should just be proceedings of the ieee
0:30:41	that's a in three display
0:30:43	but but two thousand and
0:30:45	if you wanna do some for the reading
0:30:46	huh thank you
0:30:48	thank you later
0:30:55	i sort of
0:30:57	spend a little time but we could still allow for
0:31:01	one one two interesting questions from the audience so the anybody
0:31:04	has questions please
0:31:07	yeah one of the microphones
0:31:14	oh
0:31:14	i
0:31:15	should
0:31:17	oh um yeah is only if you speech to
0:31:19	i guess and of the
0:31:21	more the the the the uh using other types of modalities and
0:31:25	to create three D like for example you know the box connector
0:31:28	uh those types of things that
0:31:30	using
0:31:31	a lose or something like that to
0:31:33	the fuse it's create make three dimensional
0:31:35	uh
0:31:36	representations
0:31:40	the question is about to
0:31:42	but
0:31:43	for can into them read representations such as the experts can eh
0:31:47	which to be uses yeah
0:31:49	i
0:31:51	a what about you
0:31:54	and used to
0:31:56	oh
0:31:58	and and and to have no
0:32:00	um
0:32:03	uh_huh
0:32:04	i to the clean that you have a
0:32:08	compute
0:32:10	um the hmmm the
0:32:14	and my
0:32:16	i
0:32:18	yeah
0:32:19	and the of time to could
0:32:22	contents
0:32:23	um
0:32:24	and that comfortable and
0:32:28	that's
0:32:30	and
0:32:33	i the question
0:32:35	a question i should see
0:32:37	like in that case
0:32:38	in the
0:32:39	interest of letting everyone had the lunch break
0:32:41	that's thing the
0:32:42	for this because the again

Trends in Image, Video, and Multidimensional Signal Processing

Expert Sessions

Přednášející: Gaurav Sharma, Patrick J. Wolfe, Lina Karam, Autoři: Gaurav Sharma, Patrick J. Wolfe, Lina Karam