0:00:13i you
0:00:16or with the laboratory relay
0:00:18university board a rest
0:00:20and also i'm of where it from a more realistic from tech
0:00:24or shown somebody
0:00:25so what i'm going to present
0:00:26oh these all approach
0:00:29copes with the particle domain mean namely the animated immediate
0:00:33so the presentation of a line for type i'm going to to the problem state one
0:00:38if is then a pretty state of the art of the to sure
0:00:41proposed proposed approach
0:00:43experimental results and finally conclude the paper
0:00:47use of that action is um or is part of a more general problem which is temporal segmentation
0:00:53oh you don't
0:00:54because temporal segmentation me
0:00:57it's composing the V don't to its fundamental
0:01:00temporal do needs for
0:01:01be do so
0:01:03a be do so
0:01:05a sequence of images which
0:01:07are four db P of a common or
0:01:09and um
0:01:10so basically to to to get of the final find a movie of the final
0:01:15one has to put to get a all of this short
0:01:18which are are Y
0:01:19what we call gradual transitions which are do not
0:01:23the image
0:01:24so basically performing the temporal or segmentation means
0:01:28uh on the for basis to be a the be doctrine
0:01:31so we have two classes of we we foundations of for form
0:01:36it's called sharp transitions or cuts
0:01:38which are the direct concatenation
0:01:40two different roles so here you have the time line
0:01:43you have shown one
0:01:44which is connected to show
0:01:45so here i got a car
0:01:48they are the most frequent
0:01:49for instance
0:01:50a mean of
0:01:51a a bit of for the chip
0:01:53the one cards
0:01:55and the existing approaches i part quite
0:01:59a a highly accurate
0:02:00we got easily and ninety five percent correct detection
0:02:04you can see the the only results of the trick the benchmark mark and compare
0:02:09on the other hand there are the gradual transition which are
0:02:12fourteen time before
0:02:14the most common we
0:02:16natural movies or or be is in general
0:02:21here i have be give a fate in sequence
0:02:24which is a is the progress a partition of one of each
0:02:27starting with a constant image
0:02:29typically that
0:02:30the other
0:02:32a kind of a idea of trying to are the diesel of each arm much more complex because they are
0:02:37the transformation of one in each
0:02:39start image
0:02:39into two but second image which is done
0:02:44compared to cost they are less frequent at least one word or measure last and the existing methods are not
0:02:51a very high reliable
0:02:52that's say we have a average
0:02:54corner detection between seventy and four
0:02:59so white white board for means the temporal segmentation
0:03:02i i'm going to an to to to report results so
0:03:06for the on a like this work was it it the way of understanding
0:03:10the structure of the
0:03:11of the be
0:03:13on the other hand we have but the content description
0:03:16for instance the many summarisation a scheme matters are
0:03:20or based on temporal segmentation or
0:03:22oh there are many approaches which can see are the action
0:03:26relate it was high frequency all
0:03:28for change
0:03:31a to this domain which is the animated movies that use of great you trying to transition has
0:03:37semantic meaning
0:03:41how well i'm going to be then
0:03:43some of the
0:03:44but matters
0:03:46well that's are with that we a definition of this
0:03:48transition to so supposing would have to sequence is to short
0:03:52S one and S two
0:03:54so that is all
0:03:56transition which is
0:03:58obtained by combining the too
0:04:00of duration they can express it
0:04:03a at intensity level
0:04:06i is the linear combination but between the two seconds
0:04:09a sequence sorry
0:04:11oh using a do a linear or more point function F one and F
0:04:16some common functions are
0:04:19such as the one i have presented yes so if on a steep because it decreasing
0:04:23for one as you know why
0:04:25the second
0:04:25function F two is
0:04:27typically increasing so basically what we have here we have a
0:04:31a a doll sequence of the for four
0:04:33which is cool we
0:04:35the fading in E C one of the second show
0:04:39so basically we have
0:04:40a fade out
0:04:41cool be the fading C
0:04:46this kind of time of these of are much more complex to detect compared to the others in one to
0:04:51two face because first of all
0:04:53very hard to
0:04:54to be beat or is a separate
0:04:58uh they they tend to show similar time signature with other channel or or object more
0:05:04based support main evaluation colour X more
0:05:07that that that a and they may have a
0:05:09caught a similar colour is the motion a structure
0:05:12if formation for the whole for the two source of the first one is the can which is a problem
0:05:17so the existing method of equal are divided into several categories of first on it
0:05:22pixel intensity by
0:05:23transform base
0:05:25feature red and there are some other approaches which
0:05:28i don't mixed
0:05:29a fourth one or propose a different solutions so i going to present
0:05:33from each some representative a approach you which are connected to our or
0:05:38oh one of the first approach well you who was using you you in each difference is
0:05:45a was to to accumulate the distance between consecutive frames
0:05:50a should be greater than a of force threshold T one one for
0:05:55the difference for consecutive frames should stay below a second threshold
0:05:59T two which is if you to do you want so basically
0:06:02it the
0:06:03computes the successive difference
0:06:06which are provided by a is all sequence
0:06:08do this work not only for is on but we gradual transmission in general
0:06:14a another approach use the mathematical definition
0:06:18space that mean and variance of pixel intensity show
0:06:21a linear and quadratic
0:06:25so that is find it on on the as a you need we if you are going to compute the
0:06:30of what use of
0:06:31sequence once for a different T
0:06:33or want of time
0:06:34we got a
0:06:36behave or
0:06:37we the F one and F two function
0:06:40so we if you are going to do the mad and replacing the to function
0:06:43we are going to obtain
0:06:45a quadratic behavior
0:06:47according to
0:06:48so here
0:06:49where a a C R three constants
0:06:52which are in time and keeping in depend
0:06:55we can we can uh detect these signature by applying for first or a second or or do but they
0:07:00do but is in order to to do you
0:07:02either a linear
0:07:03decrease or a constant to
0:07:06cost and value of of the of the this fun
0:07:10uh another approach is
0:07:12based on the optical fact
0:07:14i i just my so is a superposition of of fading fade out and in sequence
0:07:20so it detect the amount of fading dean and fading out peaks that which is also the basis for
0:07:27our at forty
0:07:29generally you you based approaches are very reliable
0:07:33similar to to to the but that's for quite detection
0:07:37other approach
0:07:39transform base
0:07:40for instance performing forming the detection on the compressed domain
0:07:44this is my work for a real-time performance but
0:07:49uh the the effect is a quite a visual that we need
0:07:52some kind of visual information not
0:07:56for or frequency domain or
0:07:58something similar so
0:08:00usually lead to increase accuracy a least we have to D compressed was that level of detail
0:08:06second and copy what you are feature rate here and going to present a class of one which is based
0:08:10on contour and edging formations so it's use
0:08:13is the same assumption
0:08:16come to each peak cells from a uh as a starting show are going to disappear
0:08:20why as a can be are from the final four are going to yeah
0:08:25one classic approach used to compute
0:08:28a edge change ratio
0:08:31disappearing feature
0:08:32H for edge
0:08:33excels and appearing in edge peaks that for instance that's here
0:08:36we have a
0:08:37the amount of
0:08:38because of quantum piece cells
0:08:40which is that appeared from image at time K
0:08:43divided by the total number of
0:08:45can two points
0:08:46so called my complete do that too they they should
0:08:49should the provide a high value for a for a dissolve
0:08:53other produce
0:08:54that to use feature points like so or see that it's at the top
0:08:58oh the program we
0:09:00feature in for is very sensitive to motion or visual
0:09:05so we do not know the information that the use most
0:09:07in fact all of the existing a dissolve detection method are
0:09:11design actually designed to cope with natural and be because that that was the target so
0:09:17in this paper we address the particular domain mean which is artistic animated movies are not be
0:09:23we stick by a car to ones
0:09:25there are quite a different
0:09:27and emission mission in the is become a uh
0:09:31that's say an important entertainment in the three
0:09:34from the artistic point of view and also from the entertainment for to there are a lot of it was
0:09:39there are at a or a lot of commercial movie your high i i have used D
0:09:43the of the of the
0:09:45because i state
0:09:47what the law uh cannot up work together and see france for instance
0:09:50the the international house and made at feel more as
0:09:54it's one of the major events in the fields there are
0:09:56a lot of movies competing
0:10:00a it became a a problem to two
0:10:02to process or from or segmentation to this
0:10:06the problem is
0:10:07artistic animated movies are
0:10:09quite different from natural ones
0:10:11in many respects here i'm going to present some of the
0:10:14the most
0:10:14importance so first of one that are many only make animation taking
0:10:19you got paper drawing
0:10:20three D
0:10:22and an object animation blast
0:10:24C modeling so it's
0:10:26the content is very in very different
0:10:29the motion and not
0:10:31always want you know that you to the animation techniques there are a lot of movies which are made by
0:10:36stop motion
0:10:38take or which are made frame by frame
0:10:41also each movie tend to have a a different colour but i here you have a
0:10:45i i one each or or or two images from a one and with still
0:10:50so they they tend to have
0:10:52a specific colour well that
0:10:54uh that the knees
0:10:56fiction or or a highly abstract
0:10:58you have a lot of visual F X job i
0:11:01strange and also there on of physical so we you we can we cannot to
0:11:06unlike the
0:11:08uh the events from the class
0:11:10point to we so
0:11:11basically you can have anything
0:11:14objects appear disappear
0:11:16any kind of visual F X so that is no
0:11:19oh that is there is no
0:11:20continuous flow
0:11:23the problem them at the we propose is quite simple but
0:11:26a yet efficient
0:11:28what we do we use only intensity information
0:11:32and for each
0:11:33frame we are going to compute
0:11:35what we call
0:11:36fading excel
0:11:37it the simple racial with
0:11:39the amount of fading out its cells
0:11:42the amount of training in excel
0:11:44which is normalized a is back to one of this is a in this size
0:11:48so basically we if we if we are going to a a like this
0:11:52measured you at time shown
0:11:54use old
0:11:57uh you isolated peaks
0:11:58the problem is how to make the difference between these all star nation and are
0:12:03changes which are due to motion or visual X
0:12:06so for that we use but between thresholding approach which i shall describe in the form
0:12:12first of all
0:12:13in order to overcome for all this one you need you we are going to analyse the fading he's than
0:12:18in of very restrained
0:12:19time don't of only three for
0:12:22that is a localisation using that winters for so we have to situation we have a
0:12:29that is all which are
0:12:31clearly not which provide a than not a number of fading use so which is quite fight
0:12:35so when whether we have
0:12:37the number of fading be solved
0:12:38a greater than a than a certain threshold
0:12:42a these value when there is a a lot of i thing we can declare a dissolve in the in
0:12:46there but
0:12:48between i
0:12:50and last
0:12:51how to max
0:12:52on the on the on the both sides where T mess is the that's say
0:12:57an average is all
0:13:01so that the the most simple situation we got
0:13:03oh that is on but there are some other the also which show
0:13:07a lower
0:13:08level of fighting be a and which are cool with all which are put to
0:13:13in other transition like motion
0:13:15or a visual X so we use
0:13:17we use a second trash for which is a
0:13:19quite a lower
0:13:20is lower than the first one we call it the tolerance threshold
0:13:23when are the F B is greater than the second verse what we may have a dissolve transition
0:13:29in fact
0:13:30uh the the frame you made
0:13:32maybe a dissolve middle frame
0:13:34so two
0:13:35to find it is easy is all
0:13:38what we are we are looking for in
0:13:40oh um you know a decreasing in on both sides
0:13:43all this is that
0:13:44so basically having been
0:13:46an mac
0:13:47but what we do here i have think that i
0:13:50uh a that a P function for a
0:13:53a a segment of of of a we we have a
0:13:56to that as a clear is old here
0:13:58and we have the to search for the sort and search for that for that one
0:14:02what but has some other on which are what we
0:14:05some other for change still
0:14:06what we do
0:14:07well we you detected a peak a greater on the second as well
0:14:11we are going
0:14:13time ones where
0:14:15a a if i'm function start increasing the again that on the right and on the left
0:14:20once we got the those times more ones
0:14:22what we we are going to a to assess
0:14:25the and it would be to and the B and those
0:14:27to values which are denoted
0:14:29you left and you die
0:14:33transition these value shall be at is on each
0:14:37the to that is are great and then hop
0:14:40the size of the
0:14:42be that
0:14:43the F B I
0:14:45so we are going to be clear
0:14:46that is all
0:14:49uh we have tested our uh our approach on of
0:14:53five hundred and S to D all that's several on a midi sequence is for each i have a peak
0:14:58at the
0:14:59a label according to that is the and if you could is that we have a high it difficult content
0:15:04we shall see at the end some examples to
0:15:07as see how how to
0:15:08how bizarre
0:15:09a contents are and average difficulty
0:15:12so to was this perform a we use the class
0:15:14or you don't cold the racial so precision is about false detection
0:15:18while you call is a well-known detection
0:15:22what are the results so
0:15:24or or one we got
0:15:25a precision of
0:15:27ninety four percent white thirty four is close to eighty percent that
0:15:31you can i sixty good detection and only twenty three for detection
0:15:37but at the sequence level
0:15:39precision and recall racial a range of four
0:15:42at T C to one hundred and the record
0:15:45step one P two one hundred so
0:15:47we have
0:15:48certain second for which we detect all
0:15:51all the mission
0:15:52and there are some for which we we
0:15:54we which you to the
0:15:56very complex and we got to a little or detection issue
0:16:02we we we have a to compare our of what which is quite simple
0:16:05to the existing approaches
0:16:08we have to choose
0:16:09three of them the variance of pixel intensities the one i have presented in the introduction
0:16:15and the edge
0:16:17range that they should be um a which is based on two hundred so here we have an example for
0:16:22one movie which is for mister part
0:16:24so we have a
0:16:25trace the
0:16:26the variance of be in T D here we have a
0:16:29D that is on problem to reach he's marked with vertical but lines
0:16:33we can see that there is no problem shape which is stated by the definition we
0:16:36we can now we can we cannot use it
0:16:39if you are tracing the the exchange ratio
0:16:42we see it it it's very
0:16:44a highly sensitive to visual F X and noise
0:16:48unusable usable and if you are a things that the proposed measure
0:16:52we can see whether there are some of duration that is also a quite
0:16:56oh oh that limited
0:16:58and not an example of a we which is
0:17:01complex as to buy the which show very discontinuous content
0:17:04we got the
0:17:05very which is
0:17:07which show a particular
0:17:08signature what what is not a part shape
0:17:11for green for is not reliable because we don't have a lot of in the movie
0:17:15while that's a classic approach example for our with we get
0:17:18very good
0:17:20detections so
0:17:21basic we we were unable to compare the precision and recall or for four
0:17:27approach approach to because we couldn't
0:17:30make them board
0:17:31uh uh i'm going to show you a few examples of
0:17:34a all which were successfully detect a and also to see
0:17:38the difficulty of the of here
0:17:41i'm going to show it on a typical dissolve transmission
0:17:44it's quite strange so that's a classic animated movies a
0:17:48but is similar to a fate by is quite a a a quite a diesel
0:17:52if fact
0:17:53here we got a dissolve transition which
0:17:56in which both
0:17:57what will are short are very similar from the point of view of the structure and also the colour
0:18:02and you it it's a tough
0:18:04the the use of trying to which is
0:18:07called with a a lot of motion and a very a lot of intensity variation which
0:18:12is also successfully detect
0:18:16we have proposed an intensity based approach is it's a simple matter is quite a of fast an efficient method
0:18:21to our to or corpus
0:18:23what are the main limitation so
0:18:26forced to one is the choice of of several threshold
0:18:29we had an able to detect of that this all model as you can
0:18:33a channel and
0:18:34we have some probably some of the phase which is reduced or
0:18:38a sometimes as
0:18:40this is you to the the pixels
0:18:44thank you for a for hour
0:18:50i think about them
0:18:51any questions have time for one
0:18:55to do you mine do but just getting the
0:19:12a i i just does of forty four
0:19:15so for them matter which is this a to information as we use the canny edge detector to know that
0:19:27another one
0:19:29how how to compute
0:19:30you are a precision and recall
0:19:32um are you considering a and to in which the detection of a describe right or is
0:19:37one single frame
0:19:38but that's that's that would that the good which and so for we are are a menu labelling as the
0:19:43sequence of simple so we are basically detecting by hand uh well that is all are yeah and uh i'm
0:19:48considering reporting detection you find look at
0:19:52yeah i'm support supporting support already the use of it
0:19:55but at least they want people
0:19:58we we are not image to detect that is this
0:20:00yeah you you like to so we are so you have to suburb and yeah in in which
0:20:04detection detection it can see does have a someone to what them because can in fact we are not able
0:20:08to detect the one but so we can what to
0:20:11but we can
0:20:13not problem it was
0:20:15we can uh a statistic
0:20:17we can do that is probably
0:20:19the average is or land for each domain
0:20:22for animated movies
0:20:24was segment three second
0:20:25what for natural reasons maybe twice
0:20:29that depends on a on a on this domain
0:20:32thank you very much for example