Recherche avancée

Sur d’autres sites (5044)

• The problems with wavelets

  27 février 2010, par Dark Shikari — DCT, Dirac, Snow, psychovisual optimizations, wavelets

  I have periodically noted in this blog and elsewhere various problems with wavelet compression, but many readers have requested that I write a more detailed post about it, so here it is.

  Wavelets have been researched for quite some time as a replacement for the standard discrete cosine transform used in most modern video compression. Their methodology is basically opposite : each coefficient in a DCT represents a constant pattern applied to the whole block, while each coefficient in a wavelet transform represents a single, localized pattern applied to a section of the block. Accordingly, wavelet transforms are usually very large with the intention of taking advantage of large-scale redundancy in an image. DCTs are usually quite small and are intended to cover areas of roughly uniform patterns and complexity.

  Both are complete transforms, offering equally accurate frequency-domain representations of pixel data. I won’t go into the mathematical details of each here ; the real question is whether one offers better compression opportunities for real-world video.

  DCT transforms, though it isn’t mathematically required, are usually found as block transforms, handling a single sharp-edged block of data. Accordingly, they usually need a deblocking filter to smooth the edges between DCT blocks. Wavelet transforms typically overlap, avoiding such a need. But because wavelets don’t cover a sharp-edged block of data, they don’t compress well when the predicted data is in the form of blocks.

  Thus motion compensation is usually performed as overlapped-block motion compensation (OBMC), in which every pixel is calculated by performing the motion compensation of a number of blocks and averaging the result based on the distance of those blocks from the current pixel. Another option, which can be combined with OBMC, is “mesh MC“, where every pixel gets its own motion vector, which is a weighted average of the closest nearby motion vectors. The end result of either is the elimination of sharp edges between blocks and better prediction, at the cost of greatly increased CPU requirements. For an overlap factor of 2, it’s 4 times the amount of motion compensation, plus the averaging step. With mesh MC, it’s even worse, with SIMD optimizations becoming nearly impossible.

  At this point, it would seem wavelets would have pretty big advantages : when used with OBMC, they have better inter prediction, eliminate the need for deblocking, and take advantage of larger-scale correlations. Why then hasn’t everyone switched over to wavelets then ? Dirac and Snow offer modern implementations. Yet despite decades of research, wavelets have consistently disappointed for image and video compression. It turns out there are a lot of serious practical issues with wavelets, many of which are open problems.

  1. No known method exists for efficient intra coding. H.264′s spatial intra prediction is extraordinarily powerful, but relies on knowing the exact decoded pixels to the top and left of the current block. Since there is no such boundary in overlapped-wavelet coding, such prediction is impossible. Newer intra prediction methods, such as markov-chain intra prediction, also seem to require an H.264-like situation with exactly-known neighboring pixels. Intra coding in wavelets is in the same state that DCT intra coding was in 20 years ago : the best known method was to simply transform the block with no prediction at all besides DC. NB : as described by Pengvado in the comments, the switching between inter and intra coding is potentially even more costly than the inefficient intra coding.

  2. Mixing partition sizes has serious practical problems. Because the overlap between two motion partitions depends on the partitions’ size, mixing block sizes becomes quite difficult to define. While in H.264 an smaller partition always gives equal or better compression than a larger one when one ignores the extra overhead, it is actually possible for a larger partition to win when using OBMC due to the larger overlap. All of this makes both the problem of defining the result of mixed block sizes and making decisions about them very difficult.

  Both Snow and Dirac offer variable block size, but the overlap amount is constant ; larger blocks serve only to save bits on motion vectors, not offer better overlap characteristics.

  3. Lack of spatial adaptive quantization. As shown in x264 with VAQ, and correspondingly in HCEnc’s implementation and Theora’s recent implementation, spatial adaptive quantization has staggeringly impressive (before, after) effects on visual quality. Only Dirac seems to have such a feature, and the encoder doesn’t even use it. No other wavelet formats (Snow, JPEG2K, etc) seem to have such a feature. This results in serious blurring problems in areas with subtle texture (as in the comparison below).

  4. Wavelets don’t seem to code visual energy effectively. Remember that a single coefficient in a DCT represents a pattern which applies across an entire block : this makes it very easy to create apparent “detail” with a DCT. Furthermore, the sharp edges of DCT blocks, despite being an apparent weakness, often result in a “fake sharpness” that can actually improve the visual appearance of videos, as was seen with Xvid. Thus wavelet codecs have a tendency to look much blurrier than DCT-based codecs, but since PSNR likes blur, this is often seen as a benefit during video compression research. Some of the consequences of these factors can be seen in this comparison ; somewhat outdated and not general-case, but which very effectively shows the difference in how wavelets handle sharp edges and subtle textures.

  Another problem that periodically crops up is the visual aliasing that tends to be associated with wavelets at lower bitrates. Standard wavelets effectively consist of a recursive function that upscales the coefficients coded by the previous level by a factor of 2 and then adds a new set of coefficients. If the upscaling algorithm is naive — as it often is, for the sake of speed — the result can look quite ugly, as if parts of the image were coded at a lower resolution and then badly scaled up. Of course, it looks like that because they were coded at a lower resolution and then badly scaled up.

  JPEG2000 is a classic example of wavelet failure : despite having more advanced entropy coding, being designed much later than JPEG, being much more computationally intensive, and having much better PSNR, comparisons have consistently shown it to be visually worse than JPEG at sane filesizes. Here’s an example from Wikipedia. By comparison, H.264′s intra coding, when used for still image compression, can beat JPEG by a factor of 2 or more (I’ll make a post on this later). With the various advancements in DCT intra coding since H.264, I suspect that a state-of-the-art DCT compressor could win by an even larger factor.

  Despite the promised benefits of wavelets, a wavelet encoder even close to competitive with x264 has yet to be created. With some tests even showing Dirac losing to Theora in visual comparisons, it’s clear that many problems remain to be solved before wavelets can eliminate the ugliness of block-based transforms once and for all.

• Progress with rtc.io

  12 août 2014, par silvia

  At the end of July, I gave a presentation about WebRTC and rtc.io at the WDCNZ Web Dev Conference in beautiful Wellington, NZ.

  

  Putting that talk together reminded me about how far we have come in the last year both with the progress of WebRTC, its standards and browser implementations, as well as with our own small team at NICTA and our rtc.io WebRTC toolbox.

  

  One of the most exciting opportunities is still under-exploited : the data channel. When I talked about the above slide and pointed out Bananabread, PeerCDN, Copay, PubNub and also later WebTorrent, that’s where I really started to get Web Developers excited about WebRTC. They can totally see the shift in paradigm to peer-to-peer applications away from the Server-based architecture of the current Web.

  Many were also excited to learn more about rtc.io, our own npm nodules based approach to a JavaScript API for WebRTC.

  

  We believe that the World of JavaScript has reached a critical stage where we can no longer code by copy-and-paste of JavaScript snippets from all over the Web universe. We need a more structured module reuse approach to JavaScript. Node with JavaScript on the back end really only motivated this development. However, we’ve needed it for a long time on the front end, too. One big library (jquery anyone ?) that does everything that anyone could ever need on the front-end isn’t going to work any longer with the amount of functionality that we now expect Web applications to support. Just look at the insane growth of npm compared to other module collections :

  

  For those that – like myself – found it difficult to understand how to tap into the sheer power of npm modules as a font end developer, simply use browserify. npm modules are prepared following the CommonJS module definition spec. Browserify works natively with that and “compiles” all the dependencies of a npm modules into a single bundle.js file that you can use on the front end through a script tag as you would in plain HTML. You can learn more about browserify and module definitions and how to use browserify.

  For those of you not quite ready to dive in with browserify we have prepared prepared the rtc module, which exposes the most commonly used packages of rtc.io through an “RTC” object from a browserified JavaScript file. You can also directly download the JavaScript file from GitHub.

  

  So, I hope you enjoy rtc.io and I hope you enjoy my slides and large collection of interesting links inside the deck, and of course : enjoy WebRTC ! Thanks to Damon, JEeff, Cathy, Pete and Nathan – you’re an awesome team !

  On a side note, I was really excited to meet the author of browserify, James Halliday (@substack) at WDCNZ, whose talk on “building your own tools” seemed to take me back to the times where everything was done on the command-line. I think James is using Node and the Web in a way that would appeal to a Linux Kernel developer. Fascinating !!

• Multivariate Testing vs A/B Testing (Quick-Start Guide)

  7 mars 2024, par Erin

  Traditional advertising (think Mad Men) was all about slogans, taglines and coming up with a one-liner that was meant to change the world.

  But that type of advertising was extremely challenging to test, so it was hard to know if it worked. Most of the time, nobody knew if they were being effective with their advertising.

  Enter modern marketing : the world of data-driven advertising.

  Thanks to the internet and web analytics tools like Matomo, you can quickly test almost anything and improve your site.

  The question is, should you do multivariate testing or A/B testing ?

  While both have their advantages, each has a specific use case.

  In this guide, we’ll break down the differences between multivariate and A/B testing, offer some pros and cons of each and show you some examples so you can decide which one is best for you.

  What is A/B testing ?

  A/B testing, or split testing, is testing an individual element in a medium against another version of the same element to see which produces better results.

  

  A/B tests are conducted by creating two different versions of a digital landmark : a website, landing page, email, or advertisement.

  The goal ? Figure out which version performs better.

  Let’s say, for example, you want to drive more sales on your core product page.

  You test two call-to-action buttons : “Buy Now” and “Add to Cart.”

  After running the test for two weeks, you see that “Buy Now” produced 1.2% conversions while “Add to Cart” produced 7.6%.

  In this scenario, you’ve found your winner : version B, “Add to Cart.”

  By conducting A/B tests regularly, you can optimise your site, increase engagement and convert more visitors into customers.

  Keep in mind that A/B testing isn’t perfect ; it doesn’t always produce a win.

  According to Noah Kagan, founder of AppSumo, only 1 out of 8 A/B tests his company conducts produces significant change.

  Advantages of A/B testing

  A/B testing is great when you need to get an accurate result fast on a specific element of your marketing efforts.

  Whether it’s a landing page or product page, you can get quick results without needing a lot of traffic.

  A/B testing is one of the most widely accepted and used testing methods for marketers and business owners.

  When you limit the number of tracked variables used in a test, you can quickly deliver reliable data, allowing you to iterate and pivot quickly if necessary.

  This is a great way to test your marketing methods, especially if you’re a newer business or you don’t have substantial traffic yet.

  Splitting up your traffic into a few segments (like with multivariate testing) will be very challenging to gain accurate results if you have lower daily traffic.

  One final advantage of A/B testing is that it’s a relatively easy way to introduce testing and optimising to a team, decision-maker, or stakeholder since it’s easy to implement. You can quickly demonstrate the value with a simple change and tangible evidence.

  Disadvantages of A/B testing

  So, what are the downsides to A/B testing ?

  Although A/B testing can get you quick results on small changes, it has limitations.

  A/B testing is all about measuring one element against another.

  This means you’re immediately limited in how many elements you can test. If you have to test out different variables, then A/B testing isn’t your best option since you’ll have to run test after test to get your result.

  If you need specific information on how different combinations of elements interact with one another on a web page, then multivariate is your best option.

  What is multivariate testing ?

  If you want to take your testing to the next level, you’ll want to try multivariate testing.

  Multivariate testing relies on the same foundational mechanism of A/B testing, but instead of matching up two elements against one another, it compares a higher number of variables at once.

  Multiple + variations = multivariate.

  Multivariate testing looks at how combinations of elements and variables interact.

  Like A/B testing, traffic to a page is split between different web page versions. Multivariate testing aims to measure each version’s effectiveness against the other versions.

  Ultimately, it’s about finding the winning combination.

  

  When to use multivariate testing

  The quick answer on when to use multivariate testing is if you have enough traffic.

  Just how much traffic, though ?

  While there’s no set number, you should aim to have 10,000 visitors per month or more, to ensure that each variant receives enough traffic to produce meaningful results within a reasonable time frame.

  Once you meet the traffic requirement, let’s talk about use cases.

  Let’s say you want to introduce a new email signup.

  But you want to create it from scratch and aren’t sure what will make your audience take action.

  So, you create a page with a signup form, a header, and an image.

  To run a multivariate test, you create two lengths of signup forms, four headlines, and two images.

  Next, you would create a test to split traffic between these sixteen combinations.

  Advantages of multivariate testing

  If you have enough traffic, multivariate testing can be an incredible way to speed up your A/B testing by testing dozens of combinations of your web page.

  This is handy when creating a new landing page and you want to determine if specific parts of your design are winners — which you can then use in future campaigns.

  Disadvantages of multivariate testing

  The main disadvantage of multivariate testing is that you need a lot of traffic to get started.

  If you try to do a multivariate analysis but you’re not getting much traffic, your results won’t be accurate (and it will take a long time to see accurate data).

  Additionally, multivariate tests are more complicated. They’re best suited for advanced marketers since more moving parts are at play.

  Key differences between multivariate and A/B testing

  Now that we’ve covered what A/B and multivariate tests are, let’s look at some key differences to help clarify which is best for you.

  

  1. Variation of combinations

  The major difference between A/B and multivariate testing is the number of combinations involved.

  With A/B testing, you only look at one element (no combinations). You simply take one part of your page (i.e., your headline copy) and make two versions.

  With multivariate testing, you’re looking at combinations of different elements (i.e., headline copy, form length, images).

  2. Number of pages to test

  The next difference lies in how many pages you will test.

  With an A/B test, you are splitting traffic on your website to two different pages : A and B.

  However, with multivariate testing, you will likely have 4-16 different test pages.

  This is because dozens of combinations can be created when you start testing a handful of elements at once.

  For example, if you want to test two headlines, two form buttons and two images on a signup form, then you have several combinations :

  • Headline A, Button A, Image A
  • Headline A, Button A, Image B
  • Headline A, Button B, Image A
  • Headline A, Button B, Image B
  • Headline B, Button A, Image A
  • Headline B, Button A, Image B
  • Headline B, Button B, Image A
  • Headline B, Button B, Image B

  In this scenario, you must create eight pages to send traffic to.

  3. Traffic requirements

  The next major difference between the two testing types is the traffic requirements.

  With A/B testing, you don’t need much traffic at all.

  Since you’re only testing two pages, you can split your traffic in half between the two types.

  However, if you plan on implementing a multivariate test, you will likely be splitting your traffic at least four or more ways.

  This means you need to have significantly more traffic coming in to get accurate data from your test. If you try to do this when your traffic is too low, you won’t have a large enough sample size.

  4. Time requirements

  Next up, just like traffic, there’s also a time requirement.

  A/B testing only tests two versions of a page against each other (while testing a single element). This means you’ll get accurate results faster than a multivariate test — usually within days.

  However, for a multivariate test, you might need to wait weeks. This is because you’re splitting your traffic by 4, 8, 12, or more web page variations. This could take months since you need a large enough sample size for accuracy.

  5. Big vs. small changes

  Another difference between A/B testing and multivariate testing is the magnitude of changes.

  With an A/B test, you’re looking at one element of a page, which means changing that element to the winning version isn’t a major overhaul of your design.

  But, with multivariate testing, you may find that the winning combination is drastically different than your control page, which could lead to a significant design change.

  6. Accuracy of results

  A/B tests are easier to decipher than multivariate testing since you only look at two versions of a single element on a page.

  You have a clear winner if one headline yields a 5% conversion rate and another yields a 1.2% conversion rate.

  But multivariate testing looks at so many combinations of a page that it can be a bit trickier to decipher what’s moving the needle.

  Pros and cons : Multivariate vs. A/B testing

  Before picking your testing method of choice, let’s look at some quick pros and cons.

  

  A/B testing pros and cons

  Here are the pros and cons of A/B testing :

  Pros

  • Get results quickly
  • Results are easier to interpret
  • Lower traffic requirement
  • Easy to get started

  Cons

  • You need to be hyper-focused on the right testing element
  • Requires performing test after test to optimise a web page

  Multivariate testing pros and cons

  Here are the pros and cons of multivariate testing :

  Pros

  • Handy when redesigning an entire web page
  • You can test multiple variables at once
  • Significant results (since traffic is higher)
  • Gather multiple data insights at once

  Cons

  • Requires substantial traffic
  • Harder to accurately decipher results
  • Not as easy to get started (more advanced)

  Use Matomo to start testing and improving your site

  

  You need to optimise your website if you want to get more leads, land more conversions and grow your business.

  A/B testing and multivariate testing are proven testing methods you can lean on to improve your website and create a better user experience.

  You may prefer one testing method now over the other, and that’s okay.

  The main thing is you’re starting to test. The best marketers and analysts in the world find what works through testing and double down on their winning tactics.

  If you want to start improving your website with testing today, get started with Matomo for free.

  With Matomo, you can conduct A/B tests and multivariate tests easily, accurately, and ethically. Unlike other web analytics tools, Matomo prioritises privacy, providing
  100% accurate data without sampling, and eliminates the need for cookie consent
  banners (except in the UK and Germany).

  Try Matomo free for 21-days. No credit card required.

  Try Matomo for Free

  21 day free trial. No credit card required.

  
  

  Your email address
  

  Your website address
  

  Your Analytics subdomain will be .matomo.cloud (change)

  Choose your analytics subdomain
  

  .matomo.cloud

  I accept the terms and conditions and data processing agreement (DPA)

  
  Your information will be used to create an account on our cloud service. For more information please consult our privacy policy.
  
  
  

  Start improving your websites now