Recherche avancée

Sur d’autres sites (4621)

• Inside WebM Technology : VP8 Intra and Inter Prediction

  20 juillet 2010, par noreply@blogger.com (Lou Quillio)
  Continuing our series on WebM technology, I will discuss the use of prediction methods in the VP8 video codec, with special attention to the TM_PRED and SPLITMV modes, which are unique to VP8.

  
  

  First, some background. To encode a video frame, block-based codecs such as VP8 first divide the frame into smaller segments called macroblocks. Within each macroblock, the encoder can predict redundant motion and color information based on previously processed blocks. The redundant data can be subtracted from the block, resulting in more efficient compression.

  
  Image by Fido Factor, licensed under Creative Commons Attribution License.
  Based on a work at www.flickr.com

  
  

  A VP8 encoder uses two classes of prediction :

  • Intra prediction uses data within a single video frame
  • Inter prediction uses data from previously encoded frames

  The residual signal data is then encoded using other techniques, such as transform coding.

  
  

  VP8 Intra Prediction Modes

  VP8 intra prediction modes are used with three types of macroblocks :

  • 4x4 luma
  • 16x16 luma
  • 8x8 chroma

  Four common intra prediction modes are shared by these macroblocks :
  

  • H_PRED (horizontal prediction). Fills each column of the block with a copy of the left column, L.
  • V_PRED (vertical prediction). Fills each row of the block with a copy of the above row, A.
  • DC_PRED (DC prediction). Fills the block with a single value using the average of the pixels in the row above A and the column to the left of L.
  • TM_PRED (TrueMotion prediction). A mode that gets its name from a compression technique developed by On2 Technologies. In addition to the row A and column L, TM_PRED uses the pixel P above and to the left of the block. Horizontal differences between pixels in A (starting from P) are propagated using the pixels from L to start each row.

  For 4x4 luma blocks, there are six additional intra modes similar to V_PRED and H_PRED, but correspond to predicting pixels in different directions. These modes are outside the scope of this post, but if you want to learn more see the VP8 Bitstream Guide.

  
  

  As mentioned above, the TM_PRED mode is unique to VP8. The following figure uses an example 4x4 block of pixels to illustrate how the TM_PRED mode works :

  

  Where C, As and Ls represent reconstructed pixel values from previously coded blocks, and X₀₀ through X₃₃ represent predicted values for the current block. TM_PRED uses the following equation to calculate X_ij :

  
  

  X_ij = L_i + A_j - C (i, j=0, 1, 2, 3)

  
  

  Although the above example uses a 4x4 block, the TM_PRED mode for 8x8 and 16x16 blocks works in the same fashion.

  TM_PRED is one of the more frequently used intra prediction modes in VP8, and for common video sequences it is typically used by 20% to 45% of all blocks that are intra coded. Overall, together with other intra prediction modes, TM_PRED helps VP8 to achieve very good compression efficiency, especially for key frames, which can only use intra modes (key frames by their very nature cannot refer to previously encoded frames).

  
  

  VP8 Inter Prediction Modes

  
  

  In VP8, inter prediction modes are used only on inter frames (non-key frames). For any VP8 inter frame, there are typically three previously coded reference frames that can be used for prediction. A typical inter prediction block is constructed using a motion vector to copy a block from one of the three frames. The motion vector points to the location of a pixel block to be copied. In most video compression schemes, a good portion of the bits are spent on encoding motion vectors ; the portion can be especially large for video encoded at lower datarates.

  
  

  Like previous VPx codecs, VP8 encodes motion vectors very efficiently by reusing vectors from neighboring macroblocks (a macroblock includes one 16x16 luma block and two 8x8 chroma blocks). VP8 uses a similar strategy in the overall design of inter prediction modes. For example, the prediction modes "NEAREST" and "NEAR" make use of last and second-to-last, non-zero motion vectors from neighboring macroblocks. These inter prediction modes can be used in combination with any of the three different reference frames.

  
  

  In addition, VP8 has a very sophisticated, flexible inter prediction mode called SPLITMV. This mode was designed to enable flexible partitioning of a macroblock into sub-blocks to achieve better inter prediction. SPLITMV is very useful when objects within a macroblock have different motion characteristics. Within a macroblock coded using SPLITMV mode, each sub-block can have its own motion vector. Similar to the strategy of reusing motion vectors at the macroblock level, a sub-block can also use motion vectors from neighboring sub-blocks above or left to the current block. This strategy is very flexible and can effectively encode any shape of sub-macroblock partitioning, and does so efficiently. Here is an example of a macroblock with 16x16 luma pixels that is partitioned to 16 4x4 blocks :

  
  

  

  
  

  where New represents a 4x4 bock coded with a new motion vector, and Left and Above represent a 4x4 block coded using the motion vector from the left and above, respectively. This example effectively partitions the 16x16 macroblock into 3 different segments with 3 different motion vectors (represented below by 1, 2 and 3) :

  
  

  

  
  

  Through effective use of intra and inter prediction modes, WebM encoder implementations can achieve great compression quality on a wide range of source material. If you want to delve further into VP8 prediction modes, read the VP8 Bitstream Guide or examine the reconintra.c and rdopt.c files in the VP8 source tree.

  
  

  Yaowu Xu, Ph.D. is a codec engineer at Google.

  
  

  

• Conversion Rate Optimisation Statistics for 2024 and Beyond

  21 novembre 2023, par Erin — Analytics Tips

  Driving traffic to your website is only half the battle. The real challenge — once you’ve used a web analytics solution to understand how users behave — is turning more of those visitors into customers.

  That doesn’t happen by accident. You need to employ conversion rate optimisation strategies and tools to see even a small lift in conversion rates. The good news is that it doesn’t take much to see massive results. Raising your conversion rate from 1% to 3% can triple your revenue. 

  In even better news, you don’t have to guess at the best ways to improve your conversion rate. We’ve done the hard work and collected the most recent and relevant conversion rate optimisation statistics to help you. 

  General conversion rate optimisation statistics

  It appears the popularity of conversion rate optimisation is soaring. According to data collected by Google Trends, there were more people searching for the term “conversion rate optimization” in September 2023 than ever before. 

  As you can see from the chart below, the term’s popularity is on a clear upward trajectory, meaning even more people could be searching for it in the near future. (Source)

  

  Do you want to know what the average landing page conversion rate is ? According to research by WordStream, the average website conversion rate across all industries is 2.35%. 

  That doesn’t paint the whole picture, however. Better-performing websites have significantly higher conversion rates. The top 25% of websites across all industries convert at a rate of 5.31% or higher. (Source)

  Let’s break things down by industry now. The Unbounce Conversion Benchmark Report offers a detailed analysis of how landing pages convert across various industries.

  First, we have the Finance and Insurance industry, which boasts a conversion rate of 15.6%. 

  On the other end, agencies appears to be one of the worst-performing. Agencies’ landing pages convert at a rate of 8.8%. (Source)

  

  What about the size of the conversion rate optimisation industry ? Given the growth in popularity of the term in Google, surely the industry is experiencing growth, right ?

  You’d be correct in that assumption. The conversion rate optimisation software market was valued at $771.2 million in 2018 and is projected to reach $1.932 billion by 2026 — a compound annual growth rate (CAGR) of 9.6%.

  Statistics on the importance of conversion rate optimisation

  If you’re reading this article, you probably think conversion rate optimisation is pretty important. But do you know its importance and where it ranks in your competitors’ priorities ? Read on to find out. 

  Bounce rate — the number of people who leave your website without visiting another page or taking action — is the scourge of conversion rate optimisation efforts. Every time someone bounces from your site, you lose the chance to convert them.

  The questions, then, are : how often do people bounce on average and how does your bounce rate compare ? 

  Siege Media analysed over 1.3 billion sessions from a range of traffic sources, including 700 million bounces, to calculate an average bounce rate of 50.9%. (Source)

  

  Bounce rates vary massively from website to website and industry to industry, however. Siege Media’s study unveils an array of average bounce rates across industries :

  • Travel – 82.58%
  • B2B – 65.17%
  • Lifestyle – 64.26%
  • Business and Finance – 63.51%
  • Healthcare – 59.50%
  • eCommerce – 54.54%
  • Insurance – 45.96%
  • Real Estate – 40.78%

  It won’t come as much of a surprise to learn that marketers are determined to reduce bounce rates and improve lead conversion. Today’s marketers are highly performance-based. When asked about their priorities for the coming year, 79% of marketers said their priority was generating quality qualified leads — the most popular answer in the survey. (Source)

  Just because it is a priority for marketers doesn’t mean that everyone has their stuff together. If you have a conversion rate optimisation process in place, you’re in the minority. According to research by HubSpot, less than one in five marketers (17%) use landing page A/B tests to improve their conversion rates. (Source)

  When it comes to personalisation strategies – a common and effective tool to increase conversion rates — the picture isn’t any rosier. Research by Salesforce found just over one-quarter of markets are confident their organisation has a successful strategy for personalisation. (Source)

  Conversion rate optimisation tactics statistics

  There are hundreds of ways to improve your website’s conversion rates. From changing the color of buttons to the structure of your landing page to your entire conversion funnel, in this section, we’ll look at the most important statistics you need to know when choosing tactics and building your own CRO experiments. 

  If you are looking for the best method to convert visitors, then email lead generation forms are the way to go, according to HubSpot. This inoffensive and low-barrier data collection method boasts a 15% conversion rate, according to the marketing automation company’s research. (Source)

  Where possible, make your call-to-actions personalised. Marketing personalisation, whether through behavioral segmentation or another strategy, is an incredibly powerful way of showing users that you care about their specific needs. It’s no great surprise, then, that HubSpot found personalised calls-to-actions perform a whopping 202% better than basic CTAs. (Source)

  If you want to boost conversion rates, then it’s just as important to focus on quantity as well as quality. Yes, a great-looking, well-written landing page will go a long way to improving your conversion rate, but having a dozen of these pages will do even more. 

  Research by HubSpot found companies see a 55% increase in leads when they increase the number of landing pages from 10 to 15. What’s more, companies with over 40 landing pages increase conversion by more than 500%. (Source)

  

  User-generated content (UGC) should also be high on your priority list to boost conversion rates. Several statistics show how powerful, impactful and persuasive social proof like user reviews can be. 

  Research shows that visitors who scroll to the point where they encounter user-generated content increase the likelihood they convert by a staggering 102.4%. (Source)

  Other trust signs can be just as impactful. Research by Trustpilot found that the following four trust signals make consumers more likely to make a purchase when shown on a product page :

  • Positive star rating and reviews (85% more likely to make a purchase)
  • Positive star rating (78%)
  • Positive customer testimonials (82%)
  • Approved or authorised seller badge (76%)

  (Source)

  Showing ratings and reviews has also increased conversion rates by 38% on home appliances and electronics stores. (Source)

  And no wonder, given that consumers are more likely to buy from brands they trust than brands they love, according to the 2021 Edelman Trust Barometer Special Report. (Source) 

  A lack of trust is also one of the top four reasons consumers abandon their shopping cart at checkout. (Source) 

  Traffic source conversion rate statistics

  What type of traffic works the best when it comes to conversions, or how often you should be signing up users to your mailing list ? Let’s look at the stats to find out. 

  Email opt-ins are one of the most popular methods for collecting customer information — and an area where digital marketers spend a lot of time and effort when it comes to conversion rate optimisation. So, what is the average conversion rate of an email opt-in box ?

  According to research by Sumo — based on 3.2 billion users who have seen their opt-in boxes — the average email opt-in rate is 1.95%. (Source)

  Search advertising is an effective way of driving website traffic, but how often do those users click on these ads ?

  WordStream’s research puts the average conversion of search advertising for all industries at 6.11%. (Source)

  The arts and entertainment industry enjoys the highest clickthrough rates (11.78%), followed by sports and recreation (10.53%) and travel (10.03%). Legal services and the home improvement industry have the lowest clickthrough rates at 4.76% and 4.8%, respectively.

  

  If you’re spending money on Google ads, then you’d better hope a significant amount of users convert after clicking them. 

  Unfortunately, conversion rates from Google ads decreased year-on-year for most industries in 2023, according to research by WordStream — in some cases, those decreases were significant. The only two industries that didn’t see a decrease in conversion rates were beauty and personal care and education and instruction. (Source)

  The average conversion rate for search ads across all industries is 7.04%. The animal and pet niche has the highest conversion rate (13.41%), while apparel, fashion and jewelry have the lowest conversion rate (1.57%). (Source)

  What about other forms of traffic ? Well, there’s good reason to try running interstitial ads on smartphone apps if you aren’t already. Ads on the iOS app see a 14.3 percent conversion rate on average. (Source)

  E-commerce conversion rate optimisation statistics (400 words)

  Conversion rate optimisation can be the difference between a store that sets new annual sales records and one struggling to get by. 

  The good news is that the conversion rate among US shoppers was the highest it’s ever been in 2021, with users converting at 2.6%. (Source)

  If you have a Shopify store, then you may find conversion rates a little lower. A survey by Littledata found the average conversion rate for Shopify was 1.4% in September 2022. (Source)

  What about specific e-commerce categories ? According to data provided by Dynamic Yield, the consumer goods category converted at the highest rate in September 2023 (4.22%), a spike of 0.34% from August. 

  Generally, the food and beverage niche boasts the highest conversion rate (4.87%), and the home and furniture niche has the lowest conversion rate (1.44%). (Source)

  If you’re serious about driving sales, don’t focus on mobile devices at the expense of consumers who shop on desktop devices. The conversion rate among US shoppers tends to be higher for desktop users than for mobile users. 

  

  In the second quarter of 2022, for instance, desktop shoppers converted at a rate of 3% on average compared to smartphone users who converted at an average rate of 2%. (Source)

  Increase your conversions with Matomo

  Conversion rate optimisation can help you grow your subscriber list, build your customer base and increase your revenue. Now, it’s time to put what you’ve learned into practice.

  Use the advice above to guide your experiments and track everything with Matomo. Achieve unparalleled data accuracy while harnessing an all-in-one solution packed with essential conversion optimisation features, including Heatmaps, Session Recordings and A/B Testing. Matomo makes it easier than ever to analyse conversion-focused experiments.

  Get more from your conversion rate optimisations by trying Matomo free for 21 days. No credit card required.

• Video concatenation puts sound out of sync

  9 août 2019, par mmorin

  (Cross-posted from Video Production, where the question received no answers and may be more technical than usual video production.)

  I have several MOV files from a DSLR camera. I concatenate them with directions from this thread :

  ffmpeg -safe 0 -f concat -i files_to_combine -vcodec copy -acodec copy temp.MOV

  where files_to_combine is :

  file ./DSC_0013.MOV
  
  ...
  
  file ./DSC_0019.MOV

  The result has image and sound in sync for the first clip and is out of sync by fractions of a second in the second clip, and out of sync by around a second for the last clip. It is probably related to this error from the log :

  [mov,mp4,m4a,3gp,3g2,mj2 @ 0x7f82dd802200] st: 0 edit list: 1 Missing key frame while searching for timestamp: 1000
  
  [mov,mp4,m4a,3gp,3g2,mj2 @ 0x7f82dd802200] st: 0 edit list 1 Cannot find an index entry before timestamp: 1000.
  
  [mov,mp4,m4a,3gp,3g2,mj2 @ 0x7f82dd802200] Auto-inserting h264_mp4toannexb bitstream filter

  How can I trim the frames to the available sound stream, then concatenate the two videos ?

  The full log from the ffmpeg command is :

  ffmpeg version 4.1.3 Copyright (c) 2000-2019 the FFmpeg developers
  
    built with Apple LLVM version 10.0.1 (clang-1001.0.46.4)
  
    configuration: --prefix=/usr/local/Cellar/ffmpeg/4.1.3_1 --enable-shared --enable-pthreads --enable-version3 --enable-hardcoded-tables --enable-avresample --cc=clang --host-cflags='-I/Library/Java/JavaVirtualMachines/adoptopenjdk-11.0.2.jdk/Contents/Home/include -I/Library/Java/JavaVirtualMachines/adoptopenjdk-11.0.2.jdk/Contents/Home/include/darwin' --host-ldflags= --enable-ffplay --enable-gnutls --enable-gpl --enable-libaom --enable-libbluray --enable-libmp3lame --enable-libopus --enable-librubberband --enable-libsnappy --enable-libtesseract --enable-libtheora --enable-libvorbis --enable-libvpx --enable-libx264 --enable-libx265 --enable-libxvid --enable-lzma --enable-libfontconfig --enable-libfreetype --enable-frei0r --enable-libass --enable-libopencore-amrnb --enable-libopencore-amrwb --enable-libopenjpeg --enable-librtmp --enable-libspeex --enable-videotoolbox --disable-libjack --disable-indev=jack --enable-libaom --enable-libsoxr
  
    libavutil      56. 22.100 / 56. 22.100
  
    libavcodec     58. 35.100 / 58. 35.100
  
    libavformat    58. 20.100 / 58. 20.100
  
    libavdevice    58.  5.100 / 58.  5.100
  
    libavfilter     7. 40.101 /  7. 40.101
  
    libavresample   4.  0.  0 /  4.  0.  0
  
    libswscale      5.  3.100 /  5.  3.100
  
    libswresample   3.  3.100 /  3.  3.100
  
    libpostproc    55.  3.100 / 55.  3.100
  
  [mov,mp4,m4a,3gp,3g2,mj2 @ 0x7f82dc00e000] Auto-inserting h264_mp4toannexb bitstream filter
  
  Input #0, concat, from 'files_to_combine':
  
    Duration: N/A, start: -0.592000, bitrate: 36888 kb/s
  
      Stream #0:0(eng): Video: h264 (High) (avc1 / 0x31637661), yuvj420p(pc, smpte170m/bt709/bt470m), 1920x1080, 35352 kb/s, 50 fps, 50 tbr, 50k tbn, 100 tbc
  
      Metadata:
  
        handler_name    : VideoHandler
  
      Stream #0:1(eng): Audio: pcm_s16le (sowt / 0x74776F73), 48000 Hz, stereo, s16, 1536 kb/s
  
      Metadata:
  
        handler_name    : SoundHandler
  
  Output #0, mov, to 'temp.MOV':
  
    Metadata:
  
      encoder         : Lavf58.20.100
  
      Stream #0:0(eng): Video: h264 (High) (avc1 / 0x31637661), yuvj420p(pc, smpte170m/bt709/bt470m), 1920x1080, q=2-31, 35352 kb/s, 50 fps, 50 tbr, 50k tbn, 50k tbc
  
      Metadata:
  
        handler_name    : VideoHandler
  
      Stream #0:1(eng): Audio: pcm_s16le (sowt / 0x74776F73), 48000 Hz, stereo, s16, 1536 kb/s
  
      Metadata:
  
        handler_name    : SoundHandler
  
  Stream mapping:
  
    Stream #0:0 -> #0:0 (copy)
  
    Stream #0:1 -> #0:1 (copy)
  
  Press [q] to stop, [?] for help
  
  [mov,mp4,m4a,3gp,3g2,mj2 @ 0x7f82dd802200] st: 0 edit list: 1 Missing key frame while searching for timestamp: 1000
  
  [mov,mp4,m4a,3gp,3g2,mj2 @ 0x7f82dd802200] st: 0 edit list 1 Cannot find an index entry before timestamp: 1000.
  
  [mov,mp4,m4a,3gp,3g2,mj2 @ 0x7f82dd802200] Auto-inserting h264_mp4toannexb bitstream filter
  
  frame=41886 fps=547 q=-1.0 Lsize= 3789826kB time=00:13:58.75 bitrate=37014.8kbits/s speed=10.9x    
  
  video:3631879kB audio:157123kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: 0.021759%

  Update (1 July 2019)

  I thought that the files had a problem at the beginning or at the end, so I
  trimmed one second from each end, but it still had the sound out of sync :

  FILES=files_to_combine
  
  OUTPUT=show2.MOV
  
  rm $FILES
  
  for i in 3 4 5 6 7 8 9; do
  
      rm ${i}.MOV
  
      duration=$(ffprobe -v 0 -show_entries format=duration -of compact=p=0:nk=1  DSC_001${i}.MOV)
  
      trimmed=$(echo $duration - 1 | bc)
  
      ffmpeg -ss 1 -t $trimmed -i DSC_001${i}.MOV -vcodec copy -acodec copy ${i}.MOV
  
      echo file ./${i}.MOV >> $FILES
  
  done
  
  
  
  rm $OUTPUT
  
  ffmpeg -safe 0 -f concat -i $FILES -vcodec copy -acodec copy $OUTPUT

  When I trim a single file near the end, the sound and video do not seem out of sync :

  ffmpeg -ss 00:09:20 -t 20 -i DSC_0014.MOV -vcodec copy -acodec copy end.MOV

  When I concatenate only 30 seconds from each video, the result seems OK :

  FILES=files_to_combine
  
  OUTPUT=show2.MOV
  
  rm $FILES
  
  for i in 3 4 5 6 7 8 9; do
  
      rm ${i}.MOV
  
      duration=$(ffprobe -v 0 -show_entries format=duration -of compact=p=0:nk=1  DSC_001${i}.MOV)
  
      start=$(echo $duration - 30 | bc)
  
      end=$(echo $duration - 1 | bc)
  
      ffmpeg -ss $start -t $end -i DSC_001${i}.MOV -vcodec copy -acodec copy ${i}.MOV
  
      echo file ./${i}.MOV >> $FILES
  
  done
  
  
  
  rm $OUTPUT
  
  ffmpeg -safe 0 -f concat -i $FILES -vcodec copy -acodec copy $OUTPUT

  This last concatenation gives this error multiple times :

  [mov @ 0x7fc3c7837400] Non-monotonous DTS in output stream 0:0; previous: 9080205, current: 9080200; changing to 9080206. This may result in incorrect timestamps in the output file.

  So I am guessing that the problem is small differences in timestamps that
  accumulate and become more noticeable with longer durations and the
  concatenation of multiple files.

  For reference, the DSLR that shot these clips is a Nikon D3300 and the result
  of ffprobe on one of the files is :

  $ ffprobe DSC_0017.MOV -hide_banner
  
  [mov,mp4,m4a,3gp,3g2,mj2 @ 0x7fab70003800] st: 0 edit list: 1 Missing key frame while searching for timestamp: 1000
  
  [mov,mp4,m4a,3gp,3g2,mj2 @ 0x7fab70003800] st: 0 edit list 1 Cannot find an index entry before timestamp: 1000.
  
  Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'DSC_0017.MOV':
  
    Metadata:
  
      major_brand     : qt  
  
      minor_version   : 537331968
  
      compatible_brands: qt  niko
  
      creation_time   : 2019-06-12T23:52:37.000000Z
  
    Duration: 00:09:53.58, start: 0.000000, bitrate: 36843 kb/s
  
      Stream #0:0(eng): Video: h264 (High) (avc1 / 0x31637661), yuvj420p(pc, smpte170m/bt709/bt470m), 1920x1080, 35300 kb/s, 50 fps, 50 tbr, 50k tbn, 100 tbc (default)
  
      Metadata:
  
        creation_time   : 2019-06-12T23:52:37.000000Z
  
      Stream #0:1(eng): Audio: pcm_s16le (sowt / 0x74776F73), 48000 Hz, 2 channels, s16, 1536 kb/s (default)
  
      Metadata:
  
        creation_time   : 2019-06-12T23:52:37.000000Z

  Update (9 August 2019)

  I concatenated the files in iMovie and the sound and image are not as out of sync as with FFMPEG. Maybe iMovie aligns the timestamps at the end of each clip instead of concatenating the audio and image streams separately.

  I ran the concatenation again with the latest ffmpeg 4.1.4_1 on these files and others from the same camera. The audio and image are in sync in one case (the results lasts 46 minutes) out of sync in another (the result lasts 48 minutes).