but that and they were mind them is the in you and for peking university was T
uh is my for your to show some idea how to do the mobile we show such so from the
title we focus the the first is the mobile search is a a a to different or for which research
not the wine is though we focus on how to reduce the bad was cost a means the loop P
three to look
a couple encoding
okay okay of nine
i will give no summarise to have lot of the we search and the give some the channel to the
regarding the mobile we should so should also like you that i'll go of this work and oh a what
i expect that we can do to make some breaks rule
and uh we also do you have a a a a type of the that my which was search you
oh i at based on our experiment the result i will give the uh search you work and the we'll
more detail but be given by the
uh able to look at this make even more a coding
okay
or let's take a look at a speech or uh we try to summarise does this to have a heart
all be sure search by Z is uh a few core
so we have the two part of they have to put a right part for the left part in the
region and a texture
such a as the post
a a local um as something like the landmark uh as that we are looking on
uh is is somehow easy because the object use a rigid and texture
but to for the right parts is a more difficult to a uh uh uh a a days a lot
of the research it computer vision research has to will makes a great have to on the right part
uh uh you you maybe you have no the google that you have that we go go dolls for the
mobile but we should search
i actually a product to hand that have been shifted to the uh have to part of two because the
for the have part is a a easy uh and these possible to do that with
you do what profit uh from using the industrial applications
okay
now is uh we try to come up so uh so row chan in got into the model which are
sort the first these a really do you where is the speech
i i think that maybe be uh is is ease on stand the mobile which are so it means that
you you use does not camera
you step the picture and transmitted to the server but to do some research and retaining the additional information but
i'm used uh all three D pen with you is the use all same very very
uh a be to a some time these uh the the bandwidth is not very stable so if for you
to as a whole in high full told that are we are called around the
uh fifty kilo byte
so uh if you transmit meet a large a amount of data are then the speed that will the B
delay they uh in you each we are have to like to he influence on the following the renting and
the reading in the to be a data set because the the bad with the as a up upstream a
downstream so if we spend a lot of bad with on that
a student the delivery is the and uh at any a multimedia data sent we'll be D for that
okay the the the two kind of the the uh use these new uh step are to to to uh
to high degree of the that solution the first it do you we to a local to
uh a the C hoc is the most of "'em" as a light develop by the
uh us to have a really worth the and the the P C is if that you what you have
a a few years ago by the same you uh a card may don't to was was day
also the other one is the how to of the to do the compression and in level to do the
uh something like uh
uh i agree gay to the but three to what's you hope by the that you goes
okay so uh uh a uh in a kind of research work the i'd uh a very promising work will
use no compress in of what the feature to to me to to the server side but even use a
it's very small and the good the but the is still cost around a
two Q by to uh to it'll by to per uh we to reader
per or uh query image
so is uh a still um suitable for a are stable one is especially for the reality at at limitation
because that use have to be able to do your that we you rate up or
analysis and the retrieval
okay out uh uh as the pipeline though the that of used the mobile about the right he's those but
so we have a two and uh and T use the code a and and is the decoder so with
the code and code of that we transform the lot
uh original noble cadre of E two small mobile cover a C so with this framework a we can get
the uh you fact you by we be to mobile search uh as a as uh the other uh on
other hand is how to you compress the image a signature at the you meet levels
okay oh we we we we try to a brief um the uh step are it's uh in the lead
his uh re in the two year
the first is the transform coding uh of the C all sub three to which what is do you uh
do develop by the
uh stand for a uh a you're at euros roles a group or other that you reckon component of or
into uh image patch the so the line used the most of them are as a actually had the receive
a lot of uh size citation asian and uh also to some work for two worlds and pack a C
D V as a stand
is the C hall
okay this why is a tree structure the i the means how to compress the
uh tree histogram coding to reduce the uh the long non parametric representation how the transmit it and the histogram
with then
a very small uh bit read
and the the knots why he's the compression a look at location me for because the back a world is
not a ways the look at will measure so how to combine the battle word
a features a and the
uh a spatial information to to get a some the to do some that your matching the verification
uh uh which can improve the final performance at the server side
okay so uh and let's the summarise the existing work there uh the the big probable and he's done now
days the most the research will has working on the purely content best the compression then be all of our
walk cover image signature based on which on the settings
so uh uh uh all idea is that a a a a a shall we uh the average of the
context you the to simple i'd the design a compression function because the with compact mission we can you reduce
the bit rate of the us english or on the other hand that we can prove that he's been at
the power of the the creature
okay okay we try to summarise all up or in four to the first use the location sensitive compression because
the we stuck on can uh location contact a mobile users
we you can improve the compression rate as we are as as the beach or discriminative
uh power
the signal the this kind of a
because for different region the that different an M are and uh for even a landmark mark that the different
the uh uh we show appearance
so for that i was at K show we do we develop a kind about these feature
means that if you a difficult to recognise we can send more uh the longer to to uh if that
it's very simple we can set the shot the so we can show at they were the scout about uh
we should is bitter
uh the so the wind as a a a symmetry it's because the lower they the for if you want
to get a very promising results a you have to uh get a wrong one hundred thousand to a video
work
an all that is very difficult to maintain a on the mobile side the so uh but you know the
mobile user is with the very useful contact or major so we can make in a symmetry double cat yeah
uh that means we can make the a very small for cadre on that
oh about side and the but to we keep the of a large able cadre at the source side
okay that that's why the to waco
uh a uh so far many research what just a P D were uh uh i uh do you that
when we coding over uh cory to the so but but in all the mobile you a mobile user has
that to make come because you up uh upstream and downstream
so can we use that stream to do some the supervision to smart that each of the mobile use a
how to do that to develop a a a a a a very compact is make you that coding and
uh a a a uh with uh and i a stream uh supervision we can get we a and you
small and uh more robust to upstream stream reader
okay okay that's a result to uh what the weight uh what uh what which E
we have a a the all around it a meeting in in uh and i'm are from the pair yeah
a and the free cost
so uh we they are that to at the first used the highs compression rate
uh for each
query image we have achieved a
paints a B
for a image quality
the uh the other one is the high and
hi to want and idea like to give used on the fee go to you not region
what it was uh what do we the
we also prove that deployed probably the a prototype
uh in five every including in to play area in and you all C D's by a lot of C
and a single point them you can see the at T big i is a large area and the fire
and see is that's small T so we try to
get a some that you this this the to go every area to evaluate a seized
okay okay a yeah is uh search a pipeline
the the for is a when a mobile user enter a given location the first day
so the location base shen we be update
we are informed the also but side
so that's of a side that can do
some the look adaptation and uh transmit to
for uh uh D the uh adaptation the from the source side in the mobile i
and there's the
uh user a based on the adaptive look cadre didn't K to some that coding and a D V but
the code
quarry no to uh a very small
a look read to the
uh uh remote side
that that's a low we are ranking at in about
okay K i stuff
a they are the two part of the a uh uh a a a a hot is a a uh
of side and that that uh bottom part is the mobile i
so uh yeah is the L P V C location do screen the video coding uh it was down um
off line at the sub to do some this special class we combine you we sure and that you a
graphic use stands and the after the come come uh after partition we to L D V coding for uh
the any if the coding for each region
okay so when i you the in to the region and we update to again uh up to the uh
location bayesian a that is such a location to a uh so side
and a does of a the we are sent back the L D V the coding is that's on of
that supervision
uh the the day
with the adaptation the mobile users can at that here and the send a compact three to the so the
search and retrieval we have you happening
at the server side the and find the retaining the the information
uh related to what you want
okay okay
a lower uh let's give some example of the
a some that's that the first but a graph clustering
but that the lying the spatial cues fusion of the uh uh photos
uh from the P in T is is um yeah it's easy to imagine that the photo uh use point
is with uh correspond to life a little with that you tag
right to wine used up partition the partition means a how to but it's a lot C into the all
serious a a a sears all the small region for for each region we extract a L D V C
code
okay that a uh the post and the lenny the then you process is a with dating L T V
C by of samples the posting uh i read them to the origin or old had already
uh key i used uh the the the target is the uh a good codebook book is expected to minimize
the ranking lost
uh a a i X use the ranking position function and uh that the X is the uh uh is
the uh uh i D F distance
i is very popular in uh a in uh traditional you for making a table now we want to a
been my that cost uh
to achieve that but that to figure out of the who a good codebook book
okay uh for now uh just now high mesh we have that to do that we have done the uh
reading partition so for each region
we specify the leading
so uh because the we have to take into account at both the a small sea creature
as we all as the uh to reach people not
so that left uh so the
the right hard the fourth item is of medium my that cold side
that that the part is minimize the retrieval loss
perform the an so if you used that small
uh look at re
uh but the the but of a question used a how we had does the training data come from
so uh now we we we make assumption
if we do not do some the uh
a small uh we keep the original large vocabulary then up of was you be good i mean the origin
back of world
oh oh oh we can sample as us that of the you need from each region
we assume
the original back a world
yeah the a good performance
so after uh feature action after we get uh using that small cold able a we assume a we we
from the should be not very our away from the original problem
so is a sort of a pseudo parry
so for this uh now uh for these three region the we have in some holes
uh image queries
so we are we want tools
we all want to minimize the
uh mean my the class
a yeah uh here is the us the i mean of that sample a lost to post uh we stop
was mess or
this part is ever weighting and this is part the ranking function
is that a is uh explore a plane show function
oh is like the decoder sickly creatures
so we want to minimize the sum of the law
the uh as is is uh for each region we have the in sample cory so we give the summary
to and two to at you find all overall cost
of the net be table and do
okay the fun we we fight we trying to find out uh find out the best cold the that in
practice we used the a greedy search a greedy algorithm
to to uh
uh to select the best model was that by uh one step by step
yeah yeah is still waiting uh are a waiting a update
so uh uh here is the compression function is uh so row minus step
of the proposed the uh the uh a hard to a time in a to the posting we pretty fine
a stressful
so what the weighting error is the is nest then
T we was no uh to the postings of it we can get of the optimize the uh we'll cover
okay i is the uh overall of the all algorithm
so the input is or in a battle what the features and the output the output is in the transformation
matrix and L D V C code the we are be uh we have we got based on the these
the uh posted it opposed us mobile cat so the yeah is a Y or loop is the posting process
so the lost the estimation ca war the action and error waiting and of to do some the transform training
so
uh that's the uh uh uh also nine of the all agree and we see called a low wrecking sensible
country T
for a D V to construct chip
okay
yeah is a result
now uh we use comparison with a out very popular as that a of research
vol
this one this one is origin original bad what a bag of what so is very large
and it this one
uh is that a the a a after that three three if used to quick coding and this is that
you that you goals while for a hearing a uh we actually we used does a a a uh X
uh a from their website to do in uh
and uh to to do the testing a data set
the red curve is our results
so uh and uh look at it is uh read out the uh what call and that G sorry i
mean mean average precision
the horizontal until is the option be so from this read that we can find a you we star
hence speech that we have a you have a by prime using mlp precision now uh we do experiment of
of five uh and uh all but a meeting data sets
uh this without come from the our around a less thousand five hundred a chlorine it's that we we draw
a we use a uh we collect
uh a five uh five us these three hundred car is a se at the so we C to this
so we have in total what's other the five hundred a court
a is the some the uh not switch of the results
so the last a that the column is the core you meet the right column use the result of that
different the a
uh a is each it out the the top
the top row or is a without out uh a using all post in the interval cadre and the center
or is that we using the origin of a a bag of word
oh from these figure we can find that you've in with the very small smaller
uh what kind of we still achieve better for missing uh every out that we see compatible to that we
use in the very a large original bag or or the features
okay uh dot now a a it's a this is uh the uh example uh you not face into to
indicate to a is that cold a word the where is the law of batch of a what a look
at
so the left the column is the query image or we can see that coke or you meet is overlap
with that red circles
the rest of the actual actually is that a cold or a is that a L D V C code
a right to like it every ten read out that so the the top retrieval results so we uh if
we you you you look at a carefully you can find that a uh are also very uh a a
a um any uh small called walk are still overlap on the retrieval images that them actually image
no match a cold work are a come from the L D B C code called or
okay okay S uh another exact in that's station the full
so that that uh that the part of the left part it L D V T called uh is that
a serious a local patch
and the the of that the use of chlorine the right to why i actually used uh a a works
to the for the L D B T sat
so uh i'll i'll main contribution is a how to do
uh to uh to my the training uh of that that the parts the uh L T V if called
a set from that
uh
uh from each region and another way used the how will take advantage of the to a coding that to
downloading loading L L P to be say it's and a based on the download L T V say a
and the we a though it is a uh a very small bit a very small uh
six each chip based on the L T V he set as is L T V C set you region
wise is we that every region
okay K i guess some the right uh uh demonstration most original be a as you we have the set
top the online webs that this this station to evaluate our algorithm
well the
for so that is uh this is that query image and the based on core you need you we attend
the result
is a wise tough
uh
a we use the google map
a look at this a picture uh we calculate a and beat
so for it uh of a it is a museum chair that we can get of the option it is
that a fifty two a row of for it as it's in each car we only
a a to the fifty two beat
okay and an example for singapore able is tough fifty B
okay
okay
uh this one is tough and come come come data set that we only need a a eighteen beach
per or for this image
okay so some is that we develop a back to be to compare just get
uh the main contribution is a how to the a location discriminative what kind of coding
and uh is uh at these uh it's east uh and T V C code it's it can make that
discriminative the
uh i in feature content all uh the uh the uh most important way used that we make the signature
very compact extremely compact only ten speech
image a
a also he's virus kind of a because a we can stand of the most important called work
and uh but that is a power to destroy a a way the uh that's important a
okay okay the uh is says in in in session a set to a coding
oh i do not know the fine and than ask question is so what's the mobile search is so a
our argument is the mobile so is quite different to make uh we show search so the scenes a how
to combine the be sure such and the side information then we can give V were very per missing and
a successful for application the mobile we shows
such so no of well now now we just use a location and information on the side of measure actually
you use are i've id D and the some that a graph attacking there are also a a a a
another that type of the values for side information
so that's uh all
a point thing to so much
yes
sorry
yeah on the uh right it the let there right at the centre left
so that the home is it at i to do with the uh where
oh you means uh you mean that is part of written out
it that when you
those are the return the match
yeah yeah that top reading result he's the best the result of but uh
uh the last to server why maybe is not a very of exact Y
yeah i do we can would take a a couple of M I T and that T means that
hi to retain in most are relate in the war uh a related to image
in that top position but the we do not the cap but we have a
whether the each position retaining the correct the result so image
is not no we didn't use it
because now they that we just to you the to verify with that can is K up achieve maybe that
in the future work are you can applied your of medication it with that result saying is the regardless of
the you image
well as you did you could you image you consistency is well
is it possible to get some or seekers speakers six trip for your were
no we didn't do that to so far
i think
thank you
i know
i was interested in how many bits you need for for
training or would only menu into
i
mean
you
a you see that
of in but maybe of maybe it's don't you know
yeah that would question
uh now uh a now i'll walk is for of focusing on the landmark search actually the for H or
ct we to some job right partition and a for each partition we sample some that uh some that mean
than a mark of you
then we sample around the sub hundred uh uh me to to do that the training phase that is that
a case the result compact called that is better especially very suitable for uh
search of them and an M are use the as you of the i don't know very sparse you in
the area then the problems a one not to be good
a bit rate you need to know how you mean the upstream or downstream
but the don't okay that downstream stream i is a some sort of index
around the uh wine kilo be
one kid beat
and the up stream is uh a power that pens but ten speech a typical T because a it's at
it
per that mark image but if you want to do the uh we sure such for product or or some
there are less texture object to then the P read we are go higher
yeah
thank you
if