Merken

The JPEG is dead! Long live the JPEG!

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Beta
Erkannte Entitäten
Sprachtranskript
words In 2010 uh a force to be reckoned with that we will also slow stepped on the stage and announced something very very important for the world and that
was who was getting into the Web performance and the 1 of the most memorable quote memorable quotes that he said at the speech was where something very very high here something like 100 ms is 100 ms are really really important after
he did this who will start announcing things amongst others
the Wikipedia image format and shorthand suicide of the where the mij look at will self-image thanks
where he was trying to be all at once I don't know how familiar you are with the novel image formats but like G 8 PG 24 gives that could carry out in animations or not J. facts that could encode lossy and where P was trying to be all at once so what's in Germany called he II leading
divide the cell so doing and we all and so we won't be try incorporate animation where I did
try to incorporate full of transparency while having 24 bit colors I it's to try to do completely lossless which is something that jet pixels to learn to do and a couple of other exciting features and it's too basically try to solve all the image problems the but the web developer community and you complain about and so you know people should be excited however of course also critiques
of people said well it actually interested in creating anything that's what for the web the just interested in creating something that has more metadata so they can so they can scan images better and you know great better at so that's that you would iteration
but you know and I wasn't there wasn't going to sit there and just take this so so I thought OK let's put this to the test and they're going to refer so I randomly sampled a couple of thousand take
facts from the internet so this is a distribution of 12 thousand randomly sampled j x and by randomly of course mean pseudorandom randomly go because you know true spot so these are just around sampled from things that were accessible on board AT and from web sites that were indexed by the agent in the archives so this is a fair a fair and
J. peg sample but it's of course biased in terms of its publicly accessible and what you can see here is the quality distribution of the data example you see the usual suspects like qualities of 75 85
90 and an optimized shape exit 99 and 100 and this is the size distribution of examples lots of files and out kilobytes and then some small some bigger and bigger and bigger and then we as used to see a less and less with images and then over there so big a bunch again about 100 and now let's run this
12 thousand times and see how well when he actually does in terms of compression this is what
I got out and this is a very very nicely ship Belkin and when I got here is I can show I can show you with the bell curve and p on a random sample like I just created compresses about 33 per cent better than the genetic encoders that were used when the objects were put online so just want peace
sliced 33 per
cent of all of these gen Vijay x criticism very
very good results and of
course uh at that time the will wasn't the only company trying to create new standards school Microsoft had been fighting for new web standards for a long time see you and speedy speedier which is now becoming http to the Microsoft of course really had to announce that own thing because you know the conscious goes from so
when they announced was edge effects are so for
the cane coming from high dynamic images had an image range images J. Microsoft created something in the enrollment of try to create non-standard and I said that there is there was performing much much better the thing the genetics
are trying to solve is that common J. bags only have 24 bits of color and for some technical technological things you want more color granularity 1 final colors better colors you don't want the lossy this object and this is what Jaipur cakes are trying to solve so when you can say the Jeopardy I was trying to be more precise than normal
and the thing is this is a quote from a woman working on the
X 2 6 4 encoder is that you know if you wanna creates the smallest image possible maybe image precision is and what you should be going for maybe what you should be going for is that the image looks good to the human eye and assuming i is not a very good judge in terms of is this color actually the true color or not humanize acquired are skeptical to influence which we're gonna see later so
Microsoft and Google taking on a battle against each other trying to create a new image format in 2010 11 who won the race let's find out yeah know this is how we
how where p is supported right now but again apologize my apologies for that cutting off but I was preparing for 16 benign presentation anyway and so where P is currently supported and everything Google and as you can clearly see on and nothing else so everything that's currently link-based linking over new rendering engine that is on the the order of chromogenic browser that supports where P and
O pronoun to but nothing else and Microsoft
looks even worse Microsoft is supported on the train engine and the new Microsoft around the edge but nothing else so you know you could say at the end of the day
and the problem that both companies face is that the word going open source cool better at this than Microsoft who was quite quick and saying where p is a GPL license I think that Microsoft extra techniques I was called in a lot of battles in which licensed to apply to it and therefore also had an even worse pick up rate so adoption so yeah you could say that open source might have might have solved this problem
if they went forward fully-fledged from day 1 at the end of all this reminds me that explicitly coming about standards and the
company's going on out there prominent for images this terrible reading a new standard in a couple of months later there's new standard and I would have like to plus 1 standards but it's not really so OK at this
point in my talk might say hold it right there where where where am I even listening to no storage space is
cheap bandwidth is
cheap so the question was where it comes to mind why should you
care why should you care about compression all I mean with that with data centers being the cheating on band with getting bigger and bigger why should we care when and I hope the answer for everybody here is because you
create for the you create for users to fight for the user because at the end of the day the but
you can see is that images have a very high correlation to page load time and that is the user experience on web site and everything that public-facing is directly impacted by low-performing images and that's what we're trying to create high-performance images so yes my liquid storage bags uncompressed and servers for the cheap shipping them to users on devices is really really hard Mr. can see this graph from a GP archive on their websites load slowly when they're fun compress badly compressed images so
users hate to away and that's
why we're that's why the image compression communities working so
hard to make things better and
this is a graph that shows you bounds Bennamoun rates from the
Alexa Top 2 thousand websites accumulated over a couple of years you can see that there is a sweet spot around the 1 2nd long time and everything else starts to accumulate abandonments and and about balance and then abandoned by the way the differences and bounds of course should be familiar but abandonment means a user
is not coming back for 6 weeks after having bounds so it is like a lost customer to the website you will not convert ever properly and so as images have a very very high correlation to page load time you can see that would create a high-performance websites and thus from web censorship outperforms images is really important new
experience and which can which
bring us which brings us to the next question so if that that important what is out there that we can use them mean just talked about where Piaget products are so what is out there and I think at this
time this time that I call it because that's what my talk is called the king is
dead long live the king the king is jam-packed but we can still rely on that J peg form from images on web sites
make up up 62 per cent of all the bytes I don't by data that is shipped with a common website and J. packs form that half of the for the make the
lion's share of all the images all there on the web so this is why we should optimize for the 1st step we should do is try to optimize J that because this will bring the biggest result and use the good news J. tags are
supported in so this isn't an experiment tool and Netscape Navigator 2 in 1995 and 1996 respectively and both supporting J. back in that fantastic that's all thanks to this
kind of vandalism because he had put give support in the Mosaic browser and people said well that's kind of cool kinship images of these images you can agree we want something better something with 24 bit color support and this is where j
comes in J. could do full-color there was standardized a
1992 and
was completely open source at the time so when the need for images arose in 1995 1996 but there was a standard available and this is why picture basically so and again focused there was here let's
take a peek into the world of for Japan but I'm so understand what it actually does so we understand how to optimize for it
and what we currently conceived as j acts on the Internet is actually a combination of the typical the full bitstream data and the file format declaration something as Jiffy excess so the j peg standard is actually far more comprehensive than that but you know for us for our purposes J. Peck for is formed by these 2 and the jet is this actually freaking out
free nodes contribute to the jet extending it achieves specifies that an images
stored like this so at the top right see complete image is is the brightness map and then we have to color channels that form the complete image and what do people know is just is
actually very very cool because it's modeled after the human visual system so you know when you see when you look at something from 1992 you might say well that's all we should build build something to something new that sort who Microsoft did but the giant extended was actually conceived by very very smart people and they already incorporated that the human visual system is faulty and can be tricked so In Dungeons and Dragons
terms you human vision might might be plus the 2 for brightness but minus 1 for color so human eyes a very very good in distinguishing between differences in brightness but it's exactly the X F F 0 0 0 0 shade of red or the exons and F C 0 0 0 shade of red human I can't really see that I mean if you if you put the both of us right next to each other the human eye could differentiate but if you you know give give them to different samples you won't see anybody going like
that colors the right column summarizes doesn't see it that way so this is where jury because of strong students it
is because this is where it can lossycompression it can be for giving 1 for getting information on the precision of color this is how joke works well assume also that just slightly altered from the original elephants succeed for so at the end
of all that the compression is done by a discrete cosine transformation type 2 because we just put it in there and this is basically what's critical loss in this with great loss J. Park and this is what looks
a lot in this would look like these are hard times that I think where you could look closely you can see compression artifacts arising while the quality of the images degraded this is a discrete consonants transfer in action and this is what people like you and me and commonly known as the Jaipur qualities OK and emotional
what is the best
qualities setting for data is it's 75 if content
is it 82 85
also more on 90
mind good good good good any other
takers somebody's shouting 194 59 something no have defined this thank you there is no quantities thank you for playing the game the thing is that the quality standard quality metric inject back is not defined in the GOP expanded everybody talks
about quality goes like when I save my dear with safe to web on quality 85 they look fine but the thing is that quality setting is not defined and every every in color something differently under the hood when in
estimating quality when we looked at the Jet example before we solve these nice spikes
at 75 thousand 1885 because this is what's opinionated Jaipur colors at some point said would be the best thing for the so at the
end you can actually say that J encoders something like highly
opinionated report but which they do really really weird stuff under the hood and they have an opinion about it so what they do is they go like well it's optimizes Huffman table 1 on donuts optimize that have table twice or 9 times and let's contrast optimizational it's not neutral optimization that's the corn or sharpening were known you know every encoder something completely different of of course and only for example says go to hell with J peg
where we've been created something completely different that is why when you use Photoshop instead of combining things you don't get quality gray scale from 0 to 99 or 100 but from what is at 9 0 2 9 something because they completely with a completely different libraries and this leads to a very very all things he the story of the little
blue Red Riding Hood the he has you can see the images are completely identical but for the color of the writing of and rats riding hood is big and the
question is human will system yeah we have rods and
cones analyse and these rods and cones are brightness and color and we have different and the columns for different kinds of wavelength of color as long medium and short cones and they are distributed
like this separation of this that long with wave cones are much much more abundant short wave that means warm colors I actually more likely to be seen in that
makes us may go beyond we we have a radical once we have an ivory perceptible for red because it tells us that something is edible something is delicious and tasty and wants to be and so red is good
in online and this is where many
encoders are more like all his radical I'm going to be more conservative when compressing this because helped Japan encoders went like if I'm going to mess with red instead of blue and maybe the human the human eye might detect difference because there's no constant for so
speaking of that of that loss in this that is therefore the blue and is not therefore red
there's also loss of lossless operations for generic wealth and and 1 of the most
well known is the effect France and then there's like the ugliness stepsister J peg can which is basically a brute-force script picture and what they do is
they optimize Huffman tables chairs they try and try to optimize Huffman tables 1 so it runs ones tries to find an optimized Huffman table that name goes like this this might be better or not and typically scandals like now on the run you like 100 times and let's find let's see which of the 100 Huffman tables like come up with this the best way to kind of cool so all of that saved it and because the government tables just change the DCT
coefficients this is actually a almost lossless our optimization that's pretty cool
but enough of that mind now let's go back to the original story
and so we form of where P and J. techniques are Microsoft and Google and of course you know that's not the end of the story because there was a 2 thousand 10 and 11 so by the end of 2011 start of 2012 looks a lot
got into the battle with Google and Microsoft and the what they did is there ran studies showing that
neither working genetics are all performed traditional J. it with JPEG is encoded with the modern encoder Another importance 15 years old and there are no effort study that was done renounced and the like and once later in a 2nd study that was taken up much more seriously and after that the guy who created the studies have
also announced the creation of a new tool by I so that that's called Monster monster that is the most modern jet encoder we have on our hands right now it was announced in March 2014 and is now being backed by by companies like Facebook because Facebook has switched to and away from and so now large companies are funding money into the open source development of J. pigments ship is completely open
source and get up get up and it's also so what kind of stupid do it optimizes
Huffman tables and that many times so basically it makes terrific trying typically scanned redundant actually not J. packers just replaced jeopardy drawn in the famous image optimum by a canal in sky who views using a matter OK those knows j images image opt in this the nice to American dragon drop images on it just doesn't
magic in every image is small OK a couple things you think you suggested optimism of famous most famous open-source tools for image optimization and whatever image Optim does a lot of other tools try to imitate because image optimism defect of standard for good who and Gallagher said monster has just become the tool of choice under the load for image often because it's just better by now it's just better than any other region encoded we also it can now should custom
quantization tables this was not possible because of patenting issues onto a short while ago but not all the patent for this has expired and will stick because already picked that up also does something that no other object and governance stress optimization so and velocity part of gender compression it tries to find ways to make the DCT coefficients even smaller the resulting in smaller about size that's great and it doesn't optimise progressive including so progress of J. packs are like a layer cake way when you download the minute you get a very low resolution image and and you know it starts to build up it's almost like the modern times a whole so what went straight it does better than the other encoder That is the layers in which the images
shipped actually are arranged in a way that is they have minimum size which is also very very clear and and all of that is done with minimal
impact on the essence who have been
to you that the essence let's take a holiday for democracy and and let's compress
it and when a compressor and look at the pixels that have changed during lossy compression this is what I can get out this is net of pixels being touched by a JPEG encoder when lossy compressing crazy persons for
and this is something we can
express the number and then make use of injecting encoders and the tool of choice is right now the the similarity structural similarity index the structure similarity index has replaced the peak signal-to-noise ratio and now which was number we measured quality of image by earlier but the structure similarity index is actually much much closer to the human eye already which is cool it's still not optimal and and there are papers being written at universities like this 1 where a new
variants of the structure similarity index of being published which are even closer to the human eye don't you remember all the acronyms like C Y S and whatever you can just cool estimate find papers from university where with all 1st try to find the visual stress structurally algorithms that underlies closer to the human visual system what I'm using right now is again Cornelissen's these dissimilarity index so to show you the similarity of the picture that you
just see it as a dissimilarity of 0 . 1 3 per cent so 0 . 1 3 per cent of the image had been time have been touched during encoding and this a number I can measure and put into tools I might ask
yourself what percentage of differentiation during a lossy encoding might be OK in which 1 is better so we know this if the image change by 50 per cent with the human I see that and the answer is yes but what eyes can't see probably a million always edge cases are differences of 1 . 5 per cent between 2 images that you give it to the test
group images with a slight differences of 1 . 5 per cent in just to new encoding differences the human eye the average human eye and it was a significant amount of time cannot differentiate if you if 1 is quite different from the other nodes landed like those is like search images like find the some differences can be those with usability test so when we have
similarity index you can actually automated J. pick quality including that's pretty cool so we can use something like a
tool that I created it's called cj peg said and what it does is it get it gets rid of V 8 the notion
that you need to define 1 certain J. product quality for your encoded like we just talked about go lower you go economic and compress all my images with 85 475 or whatever maybe that's not a good idea maybe certain images would be it would benefit from higher compression ratio 1 the images would be better with with low compression ratio and so on procedure that the actually uses the similarity index compresses the image finds out if the difference to the output image is acceptable to the to the human eye and if yes yes if not infrequently compresses with a different setting trying to find an optimal sweet spot of 1 . 5 per cent of different but funny enough water when I ran this through my for my data set I found out that I
often can reuse much much lower Jaipur qualities than I was expecting so for example 59 came up a lot with my with my tool which is funny and then expect that but there are so many J example that have just smooth gradients or are blurred because of depth of field photography and they benefit a lot from high compression ratios you can compress them strongly while creating artifacts so this where to like this comes in handy because you can automate the process while applying and speaking of automated deployment there's
another idea if we are
automating gender already my not adapted to the user's current situation if I am on a train writing through you with that metal tube writings through the seek book and I might have a very very bad data collection and like ends up so I
have like 8 100 ms of round-trip time that's really really bad they might try to open a website and that much that managing overload so what we could do is we could find out that the user currently has a very very high latency connections say edge and we can compress images strongly for that specific session and when the user has high-performance connection like G or is even Wi-Fi establish on that that have that whatever and we can ship ideal resolution images with with weak compression and again this would look like that the
high-resolution picture the high resolution image words as the low resolution image which saved by uh well shipping it's really cool and it's called a AIC and the technique is out on several and vendors by now and you can also with the HTML 5 navigation timing API
create this yourself on side so either you can buy this Singapore is a product we can build yourself just reading all the round-trip time performance of the recorded user sessions and a call and adapting gender-equality accordingly thereby creating a good user experience for all users and people converge would do whatever you want on the website and in this
notion I also thought well it's kind of cool to be able to find out and automatically find out that 1 J peg setting for each image but why it is that possible to adapt the agenda quality within an image would that be cool to if I had a photo of sky and then some people in front of I could compress that
sky far more strongly than the people but if I just use something like as and when the entire image I would get compression artifacts around people if I used it used a high compression ratio so I was thinking that we must be a way around so again let's take the holiday photo of
crazy people and let's find
out where would incur a compression art and this is a all directional Sobel edge detection algorithm on the on the phone and you can see that all the edges have been highlighted and these edges are the most likely places where compression artifacts would occur if I compress this very very strongly this because you know that J peg but subject works so I thought if I could just use a use encoder of that would iterate over the
image from say left top to bottom right because its object encoder works it works by a by a blocks going left of the button right now I could just
find the gonna look on a corner economist of uh sky sky sky sky sky and the next problem of the and I could just add them gender-equality per tire I could save me even more bytes that's what I was trying to get to using
adapt which is nothing but a 300 aligned bash scripts that which I would love to get the just demo that this is possible and after
creating this a demo was taken with the Sobel edge detection of my so might not be the best idea i'm because in within the corner it wasn't something that has corners and might also details that I want to preserve so the next logical
step for me was to find out How can I highlight entire salient regions of image salient regions means regions of an image that are interesting to the human eye there's also algorithms to find to discover this because you know we need to attract quality trying to teach computers how to see so we tried and so we have to find and relevance to see um make a computer see which areas of the image are interesting do this by combination of sharpness color vibrancy cetera cetera et cetera so what I did I looked through lots of the saliency algorithms and try to find out the 1 that it would be
beneficial to my calls and when I found out is the next maximum that symmetric surround saliency room created by a guy in Switzerland and asked him if you could open source this initiative said no 1 could just do whatever you want with it so at least it as open source and a colleague of mine help me ported to see but should super and now we can do this this a
saliency map of the political and to make it more visually astounding This is
the saliency map with only black-and-white this you can see the silence and the Press found out that the banana plant origin interesting and the guy interesting and the car and also at the sky and the sea where the background are not that's interesting when I'm not around this through a and
that encoder that goes tile volatile and assesses the GAP quality can change for the time I get something like this you can see that the encoder has found out that these pilots this the diagram tiles can be highly compressed without altering the message of the image too much and therefore they these tiles are exposed to high compression while the wider tiles are only exposed to to we compression which means these pilots will retain the individual details which is really need so this is what it acts has noted and this is what is currently being imported into more straight so I'm kind of happy about that somebody who knows image compression
better than me is picking this up and they're now they're building this into straight at some point which is really cool and it should is something where I'm seeing a computer
compression for what vision for images getting in more and more I don't know how many of you followed their course Computervision's challenges by Google and other companies that is the result of of 1 of them where a computer vision systems that use neural networks were actually very successful in detecting objects is an image successfully using a huge database of the bank to find out what is inside image now of course is a school you know for things like indexing images and in making images more accessible to people so this is really really nice this is also a huge benefit for the image compression
community because at this point we can do now is we can go and say hey look at this block there's a cat in here cats have fluorophores details and we might even read all the color of the flow and optimize for that we could go like this is a screen screens are usually black and boring so why not compress that more strongly because you know the screen might actually not incur compression artifacts because it's just the black shapes and and everything else that is not salient publish human I'm not important for the human eye perceive new images can be compressed
more strongly for example if you look at this as a supermarket website to buy to shop for food you don't give a crap if that table is highly compressible you interested in the food so I could go go and compressed that table texture very very strongly because they're not buying the table you buying food so computer vision is what I think is the is the future for image optimization then this is why we should follow this and with with interest and see what is going and all of these algorithms actually
over most of them are open source and I think almost all the teams that have those that have taken part in these challenges are about creating new start-ups because of that because they found out this has a high benefit for lots of use cases and so that got funding for this and now working on products to make to make these algorithms better the of course human computer computer vision
sometimes turns out it's right now it is that we have found who remembers this was called couple weeks ago yeah could be could be trained this but this is what computers come up with when they tried to you know dream up images are also very very weird even disturbing sometimes moments in our so think computer vision is a something we really interesting where we can expect a lot from the
future but it's all all not good news we've talked a lot about what Jake I can do now we can optimize J taken with its compression how we can shop Merisel interesting but there's just some things that you can't
really do very well the not yet anyway so there is a
working about right now the joint photographer take experts group is working on something new for Japan and that is
called jetpack xt typical xt there in comparison to the woodworking was trying to do what you X are was trying to do is fully backwards compatible that means every browser every view on the currently supports J. will be able to display wanted that xt will all put in the future which is also because this is where you want where a problem where p in a manner how well it performed was only able to be viewed in crawl in the open you you say you want open the photo and it saved
as let and the viewer you your operating system tells it uses scrolls has just we so J. peg xt will mechanic and make use of all the image processing tools that we have on average a big viewer is fully backwards compatible that's a huge class and it tries to address all the weak points that shape and as a as a form of it so that it can carry more bit depth but it will carry full of transparency not just Boolean transparency full of for transparency maps there it's can storage full of as much much more exciting features and working group
working on object also released the 1st couple of studies on how it's performing and according to the studies and outperforms everything we've seen so far this is from this year so it's very cool you can see that major projects are and Joe Jett et 2000 which is always you know cried after like 0 my god of middle region because of everything would be better no actually genetic can alter from all of them so it's
really really really need them and at this point which to all the CERN proprietary attempts by Google and Microsoft will just become redundant and the nice
thing is also the discussion apologetic exceeds completely open it's a it's a working group you can join them like any other w 3 c working group so you can write an e-mail to that you this and explain why you want to be on that's on the group and just about chat and we'll get access to detect also you can discuss feature requests or blogs and just to become a productive member of the Japan working community and if
you don't have the the time however how you
can of course create PNG a can of course grades of transparency right now I'm not talking about PG 24 actually that if
you want to transparency right now 1 of the features that J peg kind cannot do natively you can always use PGA
to the flow of entrenchments transparency channels the something I always found out that the very very few people know people always think that PGA can only carry Boolean Alfonso full also from that she of PGH can carry floor transparency with 255 steps of transparency is the
old way of Hollywood have you wouldn't done 3 tools to create it from PG 24 18 G 8 with full of such but
don't unless there's a new way again by the average person in the on the whole world world in terms of image optimization Cornelissen's the there is a working on PG climbs to impinge 1 to can do this much much much better but here's a
PG 20 form all they the motorcycle and this PGA clients to by another sky with probably the in the cool thing about this is that it is much better in terms of color ring and then the Floyd Steinberg algorithm that is currently being applied when you convert from PG 24 to so the color gradients and object so much much better but when you use connotes version so strongly encourage you to check that out pGQL to also and get up also open source
another alternative that is really really weird is you can
go vector if you want to call for transparency for us image yeah throw stones at me if possible we do this by
creating an alpha transparency from you basically share the roster image and you rapid around and CG container and inside the SCG container there's also a transparency and this is being substracted and then you basically have a there are still image with the use of nice performance than a highly compressed shape it can carry well you also have full offer transparency of course has the slides downside that it only works in every browser that can do do SVG but honestly most of the models can do this as a tool for the 2 is
called sorrow ICT in check that out just dump AJ bank in there and creates the year of a transparent to reject that mask with SVG continue for you SCG
actually is becoming more and more popular so also keep all of my formal remind you of CG 1 of the things residues becoming more and has become more and more popular is what's called a clone technique and that was a year of the tool by a SW front wheel in where she used in SEG
container to ship multiple resolutions of JPEG images for response of images on responsive website before to authorize the saved us all by standardizing picture source and inside my middle major browsers so a message you got a lot of traction because of that because it solves the major pain point about development I also introduce cool because it replaced the horrible I forms that we should correct message renders very very performance but performing now and is it is it is much smaller and it's been unattainable I form so when web developers nowadays you use 1 membership icons they tend to go to SVG because residues just better from also SCG
compresses very well what you can see here is a tool that compresses uh SCG code which is just XML under the hood by rounding off integers in our removing all the cross to the commons and the unnecessary source code invisible layers the century centuries such as a couple of tools to Clinton I of of so the question is there is there is g performance load on Internet in comparison to break because you know every XML has to be passed in the processes put that out the answer is yes it used to be but not anymore so it when you open the message 4 years ago you
could basically the brawls agree that all this is difficult to process what are you doing to me can really see the image coming it was terrible but now g is hardware-accelerated when being processed when being rendered it so it's very very fast so SEGS pellucid you rendering involves has up significant and so if you want if you ownership SEG for the web where you wanna ship the smallest integer possible because you care about the users you should use it just would compresses well that there are several tools that can do this was in most famous 1 that 1 of the old days called school scholar a Python script so I would now actually encourage you all to check out as the or I always phenomena as we go because it's
kind of cool you know as we go yeah because if it is also and as the years and you know it is available as a module and several other iterations and it can be easily installed and integrated into deployment infrastructure and optimizes all the SCG is that you should it saves a lot of bytes that's you really need and OK begin at the end of
the day however at some points
in the moment how almost all strongly you encode how much you try to save things images don't become lighter anymore so what
can you do this is what called keep a no commodity the image placeholder that made
famous in was 1 was the thousand 13 to 14 by a former colleague of mine that would Johnny and is being
adopted also by Facebook will just put a nice post on there the developers blog about how they're doing that and how they ship 200 bytes low-resolution preview images to agree to increase the user experience so if you care about user experience and you have already deployed your J peg encoders and you make sure that your websites load faster and you still think you need to do better you could think about using something like a low-quality image placeholders to ship something that like
this that already slightly resembles the image that is balance to be shipped to the user and then when the when the high resolution images is downloaded on the slow due to g connection or whatever user as it gets replaced so this way the user still already can see 1 some part of the image the rendering engine inside a browser canola
already allocate space for it so that's kind of the time and then when the high resolution images is there we can just replace it on the fly and this is guideposts introduction of OK and this is the phase propose about 1 of the slides
of a new speaker that after this talk so you know we don't even need to take close and I am also working on
something really weird and I'm calling and this is the same as only that I'm going to be using vector because I think can be smaller than 200 bytes
so this is a roster image that is being forced to 200 bytes by the but Facebook which is community really cool but you know but everybody's Facebook and it's difficult to deploy this kind of infrastructure for example did not shipping the check care of
this with SVG uh I can be possible for anybody with binding to large infrastructure so I'm thinking as something and invest some more time to make things better so maybe at the end of the day we're going to have a scalable image placeholders which is going to be really cool and then we can replace the high resolution image once it's been offered on right
there was a lot of a lot of the and
so rapid that would take away we thought
about whether the is what the open-source yesterday this is where the was shipping yes I think it is this work he was helping no I don't think so aim because you know it has moved at the bank it doesn't need all of them also I think it was like a like a synergetic xt think where p is cool because it supported on chroma chromas alot of traffic right now but it will become redundant some points of if you're looking for a pet project of Image optimisation we want help that is open source I I wouldn't encourage you to go and help will with because not good enough and so it'll be replaced the same kind holds
true for generic from from Microsoft it's not really about whether it's an open source now that's where put the stop and here it we didn't used to be open source but now it's open source in the licensing issues have been sold it is it worth shipping to Microsoft crawlers if you have problems origin to detection yes it it will helping know the same argument the
dissimilarity index of basically any similarity index algorithms but it opposite is an open source yes is it worth should be modeled applied ship this when you images just used in with an image of course is it worth helping yes I think so because we're not there yet in creating the perfect measurement of an algorithm that can save an image has changed to the human eye yes or no the similarity indexes already much much better than the peak signal-to-noise ratio but is not good enough yet as he can see by all the publications coming out of so if you want you if you want to dive into human visual
systems and try to programmatically rebuilt this visual system does not have them go ahead I think the community can community can benefit from the and this saliency algorithm it is an open-source yes that was
shipping again it's not shipping that you know just used load this this is the maximum symmetric surrounding sentences was helping although it's my pet project no it's not worth incident help me with this because I'm not supporting it anymore so neither should you know there's a better of saliency maps already already Alden available this 1 is pretty neat but it's not the best so don't waste your time and PG
quantities law or ontological PG to because it was actually for is an open source yes is it was shipping Hellier schools to make a page is it worth helping know because it's already pretty good canal knows what he's doing and again with what's what's about to come I think also png is going to have a hard time holding on so PG might be on its way out and 2 or 3 years SCG all or any other
SVG compressor is an open-source yes it was shipping Hellier gunmaking USCG smaller cool is it was helping you know we already pretty good so unless you're when you're very very optimistic about what you can do with with smart mouth to to make curves more and more performant in terms of rendering or how to express the more performant smaller in XML I don't think you should invest their time helping something against a dual role pretty good is going to help you J. peg
xt is the new kid on the block that's going to be also is an open-source yes yes it is and so you know you can get the text analysis it's not it's not publicly released it but you can get the tech demos and all the sources with all the licensing issues for source have already been sold so it's open source is it worth shipping it will be because it's fully backwards compatible that's what's going to make it also is it was helping yeah definitely so I'm just going to let meeting subscribe there get the info and they just had a meeting and Warsaw where they worked on this for us and for some time so I think the following duplicates the development and also you know compiling reporting banks back and maybe even feature requests if you're confident now that is going to be right at this point because they just started discussing this heavily in the last couple of months and last
but not least also I think the most important part must take back the best input we have right now is an open source yes is it worth shipping hell yeah because it's going to make every agent that perhaps is about helping yes there's lots and lots of open issues and and the people can actually remember really benefit from some form some smart inside so you don't have to be a complete J. Peck energy to be able to help them so for example and stuff like that but we need to find out how to find how to discover what is an image has more grain in 1 corner than the other and they're looking for creative solutions so if if you feel confident that you can come up with creative solutions to interesting problems look through the modes J.
pick it up a gender issues they have a lot of things when they're benefiting from input and I think we can all help them and make ChIP-X better right now I'm not sure this already on the on the way that's all from me thank you so much which is 5 minutes for questions the the at uh so the question is can talk about it but thousands of years and no and yet by 2010 a huge licensing problem that is why it was deployed widely after licensing problem it was sold J as normal Japan was already predominant and it was instead 2 thousand year that were not compatible not backwards compatible and wasn't used but sorry this is you know are OK so the question is what we use when gives stuff but as my slides have shown animated gives so there is something called a P animated PNG but it has 2 little support but also does any nation so you could use web was a with the gift that would work and if you if you only use interface and an nations like holding up menus up and down switching arrows etc cetera go with CSS transitions so so you can actually create very complex CSS transitions right now the nice thing about 3 instances as transitions as they support are supported by the GPU so there and a much more performance so are less less need something like my jumping gold than trying to find an assortment of on and then maybe where p animated for chroma everything link-based which actually make up like 60 per cent of its website traffic right now so it's it's worth investing in and then and then they give as a fallback for everybody else have well yes no to that so that's what I've mentioned above most shape because happened like 3 weeks ago many in 2 weeks ago there was a pattern that was expired and people just went to the get up page report detailed look this pattern just expired and the monster that people would acquire 0 and they really implemented this really really cool brothers and his aunt the
Metropolitan area network
Offene Menge
Software
Freeware
Benutzerbeteiligung
Forcing
Sprachsynthese
Wort <Informatik>
Reelle Zahl
Gruppoid
Ereignishorizont
Computeranimation
Zellularer Automat
Dateiformat
Oval
Bildgebendes Verfahren
Computeranimation
Metadaten
Benutzerbeteiligung
Bit
Pixel
Iteration
Web-Designer
Kantenfärbung
Bildgebendes Verfahren
Computeranimation
Flachbettscanner
Distributionstheorie
Web Site
Stichprobenumfang
Besprechung/Interview
Solitärspiel
Term
Pseudozufallszahlen
Whiteboard
Packprogramm
Computeranimation
Internetworking
Distributionstheorie
Shape <Informatik>
Minimierung
Elektronische Publikation
Term
Computeranimation
Objekt <Kategorie>
Metropolitan area network
System F
Stichprobenumfang
Decodierung
Kurvenanpassung
Quellencodierung
Bildgebendes Verfahren
Gammafunktion
Soundverarbeitung
Spannweite <Stochastik>
Benutzerbeteiligung
Multitasking
Vorlesung/Konferenz
Natürliche Sprache
Bildgebendes Verfahren
Computeranimation
Standardabweichung
Objekt <Kategorie>
Softwareentwickler
Bit
Total <Mathematik>
Güte der Anpassung
Decodierung
Kantenfärbung
Visuelles System
Term
Bildgebendes Verfahren
Computeranimation
Formale Grammatik
Metropolitan area network
Browser
Machsches Prinzip
Dateiformat
Vorlesung/Konferenz
Ordnung <Mathematik>
Kombinatorische Gruppentheorie
Dateiformat
Bildgebendes Verfahren
Computeranimation
Softwareentwickler
Open Source
Besprechung/Interview
Elektronischer Datenaustausch
Single Sign-On
Überlagerung <Mathematik>
Bitrate
Computeranimation
Metropolitan area network
Digitalsignal
Standardabweichung
Wort <Informatik>
Chi-Quadrat-Verteilung
Bildgebendes Verfahren
Standardabweichung
Lesen <Datenverarbeitung>
Web Site
Punkt
Graph
Indexberechnung
Flüssiger Zustand
Menge
Raum-Zeit
Packprogramm
Computeranimation
Homepage
Rechenzentrum
Spezialrechner
Metropolitan area network
Last
Last
Gruppe <Mathematik>
Total <Mathematik>
Server
Bandmatrix
Speicher <Informatik>
Quellencodierung
Korrelationsfunktion
Bildgebendes Verfahren
Binärdaten
Web Site
Subtraktion
Graph
Bitrate
Computeranimation
Gebundener Zustand
Homepage
Summengleichung
Last
Benutzerbeteiligung
Last
Vorlesung/Konferenz
Quellencodierung
Bildgebendes Verfahren
Korrelationsfunktion
Formale Grammatik
Mittelwert
Web Site
Datentyp
Gemeinsamer Speicher
Güte der Anpassung
Singularität <Mathematik>
Biprodukt
Solitärspiel
Computeranimation
Homepage
Portscanner
Metropolitan area network
Spezialrechner
Bildschirmmaske
Benutzerbeteiligung
Total <Mathematik>
Vorlesung/Konferenz
Bildgebendes Verfahren
Metropolitan area network
Bit
Verschlingung
Open Source
Mosaicing <Bildverarbeitung>
Browser
Besprechung/Interview
Vorlesung/Konferenz
Kantenfärbung
Bildgebendes Verfahren
Computeranimation
Knotenmenge
Deklarative Programmiersprache
Schaltnetz
Besprechung/Interview
Dateiformat
Vorlesung/Konferenz
Solitärspiel
Codec
Dateiformat
Bildgebendes Verfahren
Standardabweichung
Internetworking
Bildverstehen
Subtraktion
Besprechung/Interview
Term
Visuelles System
Quick-Sort
Computeranimation
Mapping <Computergraphik>
Metropolitan area network
Physikalisches System
Rechter Winkel
Stichprobenumfang
Visuelles System
Kantenfärbung
Maschinelles Sehen
Bildgebendes Verfahren
Magnetbandlaufwerk
Einfügungsdämpfung
Datentyp
Gruppe <Mathematik>
Gruppenoperation
Besprechung/Interview
t-Test
Wärmeübergang
Transformation <Mathematik>
Diskrete Gruppe
Computeranimation
Portscanner
Rechter Winkel
Total <Mathematik>
Datentyp
Vorlesung/Konferenz
Kantenfärbung
Information
Trigonometrische Funktion
Quellencodierung
Bildgebendes Verfahren
Widerspruchsfreiheit
Metropolitan area network
Linienelement
Spieltheorie
Güte der Anpassung
Besprechung/Interview
Inhalt <Mathematik>
Gleitendes Mittel
Standardabweichung
Benutzerbeteiligung
Punkt
Menge
Minimierung
Besprechung/Interview
Decodierung
Kontrast <Statistik>
Kantenfärbung
Marketinginformationssystem
Verkehrsinformation
Computeranimation
Tabelle <Informatik>
Physikalisches System
Zentrische Streckung
Subtraktion
Typentheorie
Programmbibliothek
Schreiben <Datenverarbeitung>
Physikalisches System
Kantenfärbung
Materiewelle
Bildgebendes Verfahren
Computeranimation
Drucksondierung
Trennungsaxiom
Subtraktion
Wellenlehre
Besprechung/Interview
Vorlesung/Konferenz
Decodierung
Schwebung
Drucksondierung
Soundverarbeitung
Metropolitan area network
Nichtlinearer Operator
Generizität
Einfügungsdämpfung
Huffman-Code
Minimierung
Besprechung/Interview
Skript <Programm>
Sprachsynthese
Computeranimation
Eins
Tabelle <Informatik>
Beobachtungsstudie
Metropolitan area network
Minimierung
Koeffizient
Besprechung/Interview
Vorlesung/Konferenz
Decodierung
Natürliche Sprache
Computeranimation
Monster-Gruppe
Facebook
Globale Optimierung
Open Source
Huffman-Code
Decodierung
Tropfen
Softwareentwickler
Bildgebendes Verfahren
Tabelle <Informatik>
Geschwindigkeit
Decodierung
Minimierung
Open Source
Objekt <Kategorie>
Monster-Gruppe
Matrizenring
Geometrische Quantisierung
Arithmetische Folge
Last
Geschlecht <Mathematik>
Koeffizient
Mereologie
Decodierung
Quellencodierung
Normalspannung
Bildgebendes Verfahren
Auswahlaxiom
Tabelle <Informatik>
Bildauflösung
Standardabweichung
Codec
Gewicht <Mathematik>
Pixel
Besprechung/Interview
Decodierung
Quellencodierung
Computeranimation
Schätzwert
Besprechung/Interview
Zahlenbereich
Ähnlichkeitsgeometrie
Automatische Indexierung
Vorlesung/Konferenz
Visuelles System
Decodierung
Datenstruktur
Grundraum
Normalspannung
Bildgebendes Verfahren
Auswahlaxiom
Softwaretest
Knotenmenge
Subtraktion
Differential
Mittelwert
Benutzerfreundlichkeit
Mathematisierung
Gruppenkeim
Zahlenbereich
Ähnlichkeitsgeometrie
Decodierung
Bildgebendes Verfahren
Lineare Abbildung
Subtraktion
Wasserdampftafel
Besprechung/Interview
Ähnlichkeitsgeometrie
Algorithmische Programmiersprache
Computeranimation
Metropolitan area network
Menge
Automatische Indexierung
Quellencodierung
Bildgebendes Verfahren
Chatbot
Funktion <Mathematik>
Wellenpaket
Prozess <Physik>
Schärfentiefe
Geschlecht <Mathematik>
Schreiben <Datenverarbeitung>
Vorlesung/Konferenz
Strömungsrichtung
Quellencodierung
Computeranimation
Anpassung <Mathematik>
Gradient
Einfach zusammenhängender Raum
Overloading <Informatik>
Softwareschwachstelle
Wort <Informatik>
Ideal <Mathematik>
Quellencodierung
Bildgebendes Verfahren
Computeranimation
Bildauflösung
Digitale Photographie
Güte der Anpassung
Besprechung/Interview
Systemaufruf
Vorlesung/Konferenz
Biprodukt
Bildgebendes Verfahren
Konturfindung
Algorithmus
Digitale Photographie
Besprechung/Interview
Vorlesung/Konferenz
Decodierung
Quellencodierung
Bildgebendes Verfahren
Computeranimation
Richtung
Konturfindung
Objekt <Kategorie>
Metropolitan area network
Demo <Programm>
Rechter Winkel
Besprechung/Interview
Skript <Programm>
Vorlesung/Konferenz
p-Block
Decodierung
Bildgebendes Verfahren
Computeranimation
Extrempunkt
Open Source
Schaltnetz
Systemaufruf
Hausdorff-Raum
Extrempunkt
Dialekt
Computeranimation
Computervisualistik
Data Mining
Algorithmus
Flächeninhalt
Kantenfärbung
Ganze Funktion
Bildgebendes Verfahren
Hilfesystem
Mapping <Computergraphik>
Diagramm
Parkettierung
Decodierung
Ausgleichsrechnung
Quellencodierung
Chi-Quadrat-Verteilung
Message-Passing
Bildgebendes Verfahren
Computeranimation
Objekt <Kategorie>
Punkt
Automatische Indexierung
Datenhaltung
Computervision
Gebäude <Mathematik>
Vorlesung/Konferenz
Physikalisches System
Quellencodierung
Maschinelles Sehen
Bildgebendes Verfahren
Neuronales Netz
Computervisualistik
Web Site
Shape <Informatik>
Punkt
Minimierung
p-Block
Computervisualistik
Textur-Mapping
Algorithmus
Vorlesung/Konferenz
Kantenfärbung
Computerunterstützte Übersetzung
Quellencodierung
Maschinelles Sehen
Bildgebendes Verfahren
Informationssystem
Touchscreen
Tabelle <Informatik>
Algorithmus
Momentenproblem
Open Source
Mereologie
Besprechung/Interview
Biprodukt
Maschinelles Sehen
Bildgebendes Verfahren
Computervisualistik
Expertensystem
Sichtenkonzept
Spider <Programm>
Digitale Photographie
Browser
Gruppenkeim
Vorlesung/Konferenz
Paarvergleich
Quellencodierung
Computeranimation
Beobachtungsstudie
Bit
Punkt
Klasse <Mathematik>
Gruppenkeim
Bildanalyse
Viewer
Computeranimation
Mapping <Computergraphik>
Objekt <Kategorie>
Metropolitan area network
Bildschirmmaske
Mittelwert
Netzbetriebssystem
Grundsätze ordnungsmäßiger Datenverarbeitung
Projektive Ebene
Bildschirmsymbol
Boolesche Algebra
Uniforme Struktur
Punkt
Web log
Gruppenkeim
E-Mail
Computeranimation
Portscanner
Besprechung/Interview
Übertrag
Boolesche Algebra
Solitärspiel
Datenfluss
Computeranimation
Gradient
Minimierung
Open Source
Versionsverwaltung
Kraft
Term
Computeranimation
Gradient
Bildschirmmaske
Client
Unterring
Algorithmus
Kantenfärbung
Bildgebendes Verfahren
Rechenschieber
Shape <Informatik>
Informationsmodellierung
Browser
Landau-Theorie
Jensen-Maß
Äußere Algebra eines Moduls
Vektorraum
Bildgebendes Verfahren
Computeranimation
Web Site
Punkt
Browser
Open Source
Besprechung/Interview
Element <Mathematik>
Bildschirmsymbol
Computeranimation
Verdeckungsrechnung
Bildschirmmaske
Multiplikation
Residuum
Endogene Variable
Volumenvisualisierung
Web-Designer
Speicherabzug
Softwareentwickler
Normalspannung
Message-Passing
Bildgebendes Verfahren
Klon <Mathematik>
Bildauflösung
Prozess <Physik>
Quellcode
Paarvergleich
Extrempunkt
Code
Computeranimation
Internetworking
Metropolitan area network
Benutzerbeteiligung
Ganze Zahl
Last
Skript <Programm>
Gravitationsgesetz
Cloud Computing
Quellencodierung
Message-Passing
Bildgebendes Verfahren
Portscanner
Metropolitan area network
Punkt
Freier Parameter
Momentenproblem
Besprechung/Interview
Ablöseblase
Iteration
Extrempunkt
Modul
Bildgebendes Verfahren
Computeranimation
Einfach zusammenhängender Raum
Web Site
Facebook
Freier Parameter
Browser
Solitärspiel
Data Mining
Summengleichung
Mereologie
Decodierung
Softwareentwickler
Bildgebendes Verfahren
Bildauflösung
Rechenschieber
Facebook
Systemaufruf
Vorlesung/Konferenz
Vektorraum
Bildgebendes Verfahren
Phasenumwandlung
Raum-Zeit
Computeranimation
Bildauflösung
Metropolitan area network
Skalierbarkeit
Freier Parameter
Globale Optimierung
Rechter Winkel
Open Source
Projektive Ebene
Hilfesystem
Bildgebendes Verfahren
Computeranimation
Bildauflösung
Parametersystem
Algorithmus
Automatische Indexierung
Spider <Programm>
Open Source
Visuelles System
Ähnlichkeitsgeometrie
Einflussgröße
Bildgebendes Verfahren
Computeranimation
Ontologie <Wissensverarbeitung>
Extrempunkt
Open Source
Inzidenzalgebra
Gesetz <Physik>
Computeranimation
Homepage
Mapping <Computergraphik>
Algorithmus
Last
Vorlesung/Konferenz
Visuelles System
Projektive Ebene
Cloud Computing
Hilfesystem
Demo <Programm>
Punkt
Verbandstheorie
Open Source
Information
p-Block
Softwareentwickler
Kurvenanpassung
Solitärspiel
Term
Computeranimation
Analysis
ATM
Shape <Informatik>
Open Source
Gruppenoperation
Familie <Mathematik>
Ein-Ausgabe
Komplex <Algebra>
Computeranimation
Homepage
Monster-Gruppe
Rechenschieber
Energiedichte
Benutzerbeteiligung
Bildschirmmaske
Offene Menge
Geschlecht <Mathematik>
Mustersprache
Mereologie
Zeitrichtung
Normalvektor
Verkehrsinformation
Bildgebendes Verfahren
Schnittstelle
Instantiierung
Offene Menge
Software
Freeware
Computeranimation

Metadaten

Formale Metadaten

Titel The JPEG is dead! Long live the JPEG!
Serientitel FrOSCon 2015
Teil 22
Anzahl der Teile 80
Autor Baldauf, Tobias
Lizenz CC-Namensnennung 3.0 Unported:
Sie dürfen das Werk bzw. den Inhalt zu jedem legalen Zweck nutzen, verändern und in unveränderter oder veränderter Form vervielfältigen, verbreiten und öffentlich zugänglich machen, sofern Sie den Namen des Autors/Rechteinhabers in der von ihm festgelegten Weise nennen.
DOI 10.5446/19621
Herausgeber Free and Open Source software Conference (FrOSCon) e.V.
Erscheinungsjahr 2015
Sprache Englisch

Inhaltliche Metadaten

Fachgebiet Informatik
Abstract A JPEG is a JPEG and nothing changed since 1992? Think again! Automated compression quality detection, advanced dithering algorithms, saliency mapping and computer vision - get a glimpse at what the image optimization open source community has been up to! Tobias Baldauf

Ähnliche Filme

Loading...