Recherche avancée

Sur d’autres sites (6555)

• What Every Programmer Should Know

  24 décembre 2012, par Multimedia Mike — General

  During my recent effort to force myself to understand Unicode and modern text encoding/processing, I was reminded that this is something that “every programmer should just know”, an idea that comes up every so often, usually in relation to a subject in which the speaker is already an expert. One of the most absurd examples I ever witnessed was a blog post along the lines of “What every working programmer ought to know about [some very specific niche of enterprise-level Java programming]“. I remember reading through the article and recognizing that I had almost no knowledge of the material. Disturbing, since I am demonstrably a “working programmer”.

  For fun, I queried the googles on the matter of what ever programmer ought to know.

  Specific Topics
  Here is what every programmer should know about : Unicode, time, memory (simple), memory (extremely in-depth), regular expressions, search engine optimization, floating point, security, basic number theory, race conditions, managed C++, VIM commands, distributed systems, object-oriented design, latency numbers, rate monotonic algorithm, merging branches in Mercurial, classes of algorithms, and human names.

  Broader Topics
  20 subjects every programmer should know, 97 things every programmer should know, 12 things every programmer should know, things every programmer should know (27 items), 10 papers every programmer should read at least twice, 10 things every programmer should know for their first job.

  Meanwhile, I remain fond of this xkcd comic whose mouseover text describes all that a person genuinely needs to know. Still, the new year is upon us, a time when people often make commitments to bettering themselves, and it couldn’t hurt (much) to at least skim some of the lists and find out what you never knew that you never knew.

  What About Multimedia ?
  Reading the foregoing (or the titles of the foregoing pieces), I naturally wonder if I should write something about what every programmer should know about multimedia. I think it would look something like a multimedia programming FAQ. These are some items that I can think of :

  1. YUV : The other colorspace (since most programmers are only familiar with RGB and have no idea what to make of the YUV that comes out of most video decoding APIs)
  2. Why you can’t easily seek randomly to any specific frame in a video file (keyframe/interframe discussion and their implications)
  3. Understand your platform before endeavoring to implement multimedia software (modern platforms, particularly mobile platforms, probably provide everything you need in the native APIs and there is likely little reason to compile libavcodec for the platform)
  4. Difference between containers and codecs (longstanding item, but I would argue it’s less relevant these days due to standardization on the MPEG — MP4/H.264/AAC — stack)
  5. What counts as a multimedia standard in this day and age (comparing the foregoing MPEG stack with the WebM/VP8/Vorbis stack)
  6. Trade-offs to consider when engineering a multimedia solution
  7. Optimization doesn’t always work the way you think it does (not everything touted as a massive speed-up in the world of computing — whether it be multithreaded CPUs, GPGPUs, new SIMD instruction sets — will necessarily be applicable to multimedia processing)
  8. A practical guide to legal issues would not be amiss
  9.  ???

  What other items count as “something multimedia-related that every programmer should know” ?

• Benefits and Shortcomings of Multi-Touch Attribution

  13 mars 2023, par Erin — Analytics Tips

  Few sales happen instantly. Consumers take their time to discover, evaluate and become convinced to go with your offer. 

  Multi-channel attribution (also known as multi-touch attribution or MTA) helps businesses better understand which marketing tactics impact consumers’ decisions at different stages of their buying journey. Then double down on what’s working to secure more sales. 

  Unlike standard analytics, multi-channel modelling combines data from various channels to determine their cumulative and independent impact on your conversion rates. 

  The main benefit of multi-touch attribution is obvious : See top-performing channels, as well as those involved in assisted conversions. The drawback of multi-touch attribution : It comes with a more complex setup process. 

  If you’re on the fence about getting started with multi-touch attribution, here’s a summary of the main arguments for and against it. 

  What Are the Benefits of Multi-Touch Attribution ?

  Remember an old parable of blind men and an elephant ?

  Each one touched the elephant and drew conclusions about how it might look. The group ended up with different perceptions of the animal and thought the others were lying…until they decided to work together on establishing the truth.

  Multi-channel analytics works in a similar way : It reconciles data from various channels and campaign types into one complete picture. So that you can get aligned on the efficacy of different campaign types and gain some other benefits too. 

  Better Understanding of Customer Journeys 

  On average, it takes 8 interactions with a prospect to generate a conversion. These interactions happen in three stages : 

  • Awareness : You need to introduce your company to the target buyers and pique their interest in your solution (top-of-the-funnel). 
  • Consideration : The next step is to channel this casual interest into deliberate research and evaluation of your offer (middle-of-the-funnel). 
  • Decision : Finally, you need to get the buyer to commit to your offer and close the deal (bottom-of-the-funnel). 

  You can analyse funnels using various attribution models — last-click, fist-click, position-based attribution, etc. Each model, however, will spotlight the different element(s) of your sales funnel. 

  For example, a single-touch attribution model like last-click zooms in on the bottom-of-the-funnel stage. You can evaluate which channels (or on-site elements) sealed the deal for the prospect. For example, a site visitor arrived from an affiliate link and started a free trial. In this case, the affiliate (referral traffic) gets 100% credit for the conversion. 

  This measurement tactic, however, doesn’t show which channels brought the customer to the very bottom of your funnel. For instance, they may have interacted with a social media post, your landing pages or a banner ad before that. 

  Multi-touch attribution modelling takes funnel analysis a notch further. In this case, you map more steps in the customer journey — actions, events, and pages that triggered a visitor’s decision to convert — in your website analytics tool.

  

  Then, select a multi-touch attribution model, which provides more backward visibility aka allows you to track more than one channel, preceding the conversion. 

  For example, a Position Based attribution model reports back on all interactions a site visitor had between their first visit and conversion. 

  A prospect first lands at your website via search results (Search traffic), which gets a 40% credit in this model. Two days later, the same person discovers a mention of your website on another blog and visits again (Referral traffic). This time, they save the page as a bookmark and revisit it again in two more days (Direct traffic). Each of these channels will get a 10% credit. A week later, the prospect lands again on your site via Twitter (Social) and makes a request for a demo. Social would then receive a 40% credit for this conversion. Last-click would have only credited social media and first-click — search engines. 

  The bottom line : Multi-channel attribution models show how different channels (and marketing tactics) contribute to conversions at different stages of the customer journey. Without it, you get an incomplete picture.

  Improved Budget Allocation 

  Understanding causal relationships between marketing activities and conversion rates can help you optimise your budgets.

  First-click/last-click attribution models emphasise the role of one channel. This can prompt you toward the wrong conclusions. 

  For instance, your Facebook ads campaigns do great according to a first-touch model. So you decide to increase the budget. What you might be missing though is that you could have an even higher conversion rate and revenue if you fix “funnel leaks” — address high drop-off rates during checkout, improve page layout and address other possible reasons for exiting the page.

  

  By knowing when and why people abandon their purchase journey, you can improve your marketing velocity (aka the speed of seeing the campaign results) and your marketing costs (aka the budgets you allocate toward different assets, touchpoints and campaign types). 

  Or as one of the godfathers of marketing technology, Dan McGaw, explained in a webinar :

  “Once you have a multi-touch attribution model, you [can] actually know the return on ad spend on a per-campaign basis. Sometimes, you can get it down to keywords. Sometimes, you can get down to all kinds of other information, but you start to realise, “Oh, this campaign sucks. I should shut this off.” And then really, that’s what it’s about. It’s seeing those campaigns that suck and turning them off and then taking that budget and putting it into the campaigns that are working”.

  More Accurate Measurements 

  The big boon of multi-channel marketing attribution is that you can zoom in on various elements of your funnel and gain granular data on the asset’s performance. 

  In other words : You get more accurate insights into the different elements involved in customer journeys. But for accurate analytics measurements, you must configure accurate tracking. 

  Define your objectives first : How do you want a multi-touch attribution tool to help you ? Multi-channel attribution analysis helps you answer important questions such as :

  • How many touchpoints are involved in the conversions ? 
  • How long does it take for a lead to convert on average ? 
  • When and where do different audience groups convert ? 
  • What is your average win rate for different types of campaigns ?

  Your objectives will dictate which multi-channel modelling approach will work best for your business — as well as the data you’ll need to collect. 

  At the highest level, you need to collect two data points :

  • Conversions : Desired actions from your prospects — a sale, a newsletter subscription, a form submission, etc. Record them as tracked Goals. 
  • Touchpoints : Specific interactions between your brand and targets — specific page visits, referral traffic from a particular marketing channel, etc. Record them as tracked Events. 

  Your attribution modelling software will then establish correlation patterns between actions (conversions) and assets (touchpoints), which triggered them. 

  The accuracy of these measurements, however, will depend on the quality of data and the type of attribution modelling used. 

  Data quality stands for your ability to procure accurate, complete and comprehensive information from various touchpoints. For instance, some data won’t be available if the user rejected a cookie consent banner (unless you’re using a privacy-focused web analytics tool like Matomo). 

  Different attribution modelling techniques come with inherent shortcomings too as they don’t accurately represent the average sales cycle length or track visitor-level data, which allows you to understand which customer segments convert best.

  Learn more about selecting the optimal multi-channel attribution model for your business.

  What Are the Limitations of Multi-Touch Attribution ?

  Overall, multi-touch attribution offers a more comprehensive view of the conversion paths. However, each attribution model (except for custom ones) comes with inherent assumptions about the contribution of different channels (e.g,. 25%-25%-25%-25% in linear attribution or 40%-10%-10%-40% in position-based attribution). These conversion credit allocations may not accurately represent the realities of your industry. 

  Also, most attribution models don’t reflect incremental revenue you gain from existing customers, which aren’t converting through analysed channels. For example, account upgrades to a higher tier, triggered via an in-app offer. Or warranty upsell, made via a marketing email. 

  In addition, you should keep in mind several other limitations of multi-touch attribution software.

  Limited Marketing Mix Analysis 

  Multi-touch attribution tools work in conjunction with your website analytics app (as they draw most data from it). Because of that, such models inherit the same visibility into your marketing mix — a combo of tactics you use to influence consumer decisions.

  Multi-touch attribution tools cannot evaluate the impact of :

  • Dark social channels 
  • Word-of-mouth 
  • Offline promotional events
  • TV or out-of-home ad campaigns 

  If you want to incorporate this data into your multi-attribution reporting, you’ll have to procure extra data from other systems — CRM, ad measurement partners, etc, — and create complex custom analytics models for its evaluation.

  Time-Based Constraints 

  Most analytics apps provide a maximum 90-day lookback window for attribution. This can be short for companies with longer sales cycles. 

  

  Marketing channels can be overlooked or underappreciated when your attribution window is too short. Because of that, you may curtail spending on brand awareness campaigns, which, in turn, will reduce the number of people entering the later stages of your funnel. 

  At the same time, many businesses would also want to track a look-forward window — the revenue you’ll get from one customer over their lifetime. In this case, not all tools may allow you to capture accurate information on repeat conversions — through re-purchases, account tier updates, add-ons, upsells, etc. 

  Again, to get an accurate picture you’ll need to understand how far into the future you should track conversions. Will you only record your first sales as a revenue number or monitor customer lifetime value (CLV) over 3, 6 or 12 months ? 

  The latter is more challenging to do. But CLV data can add another depth of dimension to your modelling accuracy. With Matomo, you set up this type of tracking by using our visitors’ tracking feature. We can help you track select visitors with known identifiers (e.g. name or email address) to discover their visiting patterns over time. 

  

  Limited Access to Raw Data 

  In web analytics, raw data stands for unprocessed website visitor information, stripped from any filters, segmentation or sampling applied. 

  Data sampling is a practice of analysing data subsets (instead of complete records) to extrapolate findings towards the entire data set. Google Analytics 4 applies data sampling once you hit over 500k sessions at the property level. So instead of accurate, real-life reporting, you receive approximations, generated by machine learning models. Data sampling is one of the main reasons behind Google Analytics’ accuracy issues. 

  In multi-channel attribution modelling, usage of sampled data creates further inconsistencies between the reports and the actual state of affairs. For instance, if your website generates 5 million page views, GA multi-touch analytical reports are based on the 500K sample size aka only 90% of the collected information. This hardly represents the real effect of all marketing channels and can lead to subpar decision-making. 

  With Matomo, the above is never an issue. We don’t apply data sampling to any websites (no matter the volume of traffic) and generate all the reports, including multi-channel attribution ones, based on 100% real user data. 

  AI Application 

  On the other hand, websites with smaller traffic volumes often have limited sampling datasets for building attribution models. Some tracking data may also be not available because the visitor rejected a cookie banner, for instance. On average, less than 50% of users in Australia, France, Germany, Denmark and the US among other countries always consent to all cookies. 

  To compensate for such scenarios, some multi-touch attribution solutions apply AI algorithms to “fill in the blanks”, which impacts the reporting accuracy. Once again, you get approximate data of what probably happened. However, Matomo is legally exempt from showing a cookie consent banner in most EU markets. Meaning you can collect 100% accurate data to make data-driven decisions.

  Difficult Technical Implementation 

  Ever since attribution modelling got traction in digital marketing, more and more tools started to emerge.

  

  Most web analytics apps include multi-touch attribution reports. Then there are standalone multi-channel attribution platforms, offering extra features for conversion rate optimization, offline channel tracking, data-driven custom modelling, etc. 

  Most advanced solutions aren’t available out of the box. Instead, you have to install several applications, configure integrations with requested data sources, and then use the provided interfaces to code together custom data models. Such solutions are great if you have a technical marketer or a data science team. But a steep learning curve and high setup costs make them less attractive for smaller teams. 

  Conclusion 

  Multi-touch attribution modelling lifts the curtain in more steps, involved in various customer journeys. By understanding which touchpoints contribute to conversions, you can better plan your campaign types and budget allocations. 

  That said, to benefit from multi-touch attribution modelling, marketers also need to do the preliminary work : Determine the key goals, set up event and conversion tracking, and then — select the optimal attribution model type and tool. 

  Matomo combines simplicity with sophistication. We provide marketers with familiar, intuitive interfaces for setting up conversion tracking across the funnel. Then generate attribution reports, based on 100% accurate data (without any sampling or “guesstimation” applied). You can also get access to raw analytics data to create custom attribution models or plug it into another tool ! 

  Start using accurate, easy-to-use multi-channel attribution with Matomo. Start your free 21-day trial now. No credit card requried. 

• ffprobe newer version detect audio codec incorrectly

  16 janvier, par alancc

  I find a strange problem.

  


  I have a test video with h264 video codec and aac audio codec. It is at https://drive.google.com/file/d/1YAyz5cO0kb9r0MgahCpISR4bZ_1_n8PL/view?usp=sharing

  


  I build a ffmpeg version by myself, its version is :

  


  ffprobe version 7.0.2 Copyright (c) 2007-2024 the FFmpeg developers
  built with gcc 14.1.0 (Rev3, Built by MSYS2 project)
  configuration: --enable-shared
  libavutil      59.  8.100 / 59.  8.100
  libavcodec     61.  3.100 / 61.  3.100
  libavformat    61.  1.100 / 61.  1.100
  libavdevice    61.  1.100 / 61.  1.100
  libavfilter    10.  1.100 / 10.  1.100
  libswscale      8.  1.100 /  8.  1.100
  libswresample   5.  1.100 /  5.  1.100


  


  I then use ffprobe to get its info :

  


  ffprobe -v quiet -print_format ini -show_streams -show_packets test_h264.mp4 > test_h264.ini


  


  Then I get an ini file which shows the audio codec as MP2 :

  


  [streams.stream.0]
index=0
codec_name=mp2
codec_long_name=MP2 (MPEG audio layer 2)
profile=unknown
codec_type=audio
codec_tag_string=mp4a
codec_tag=0x6134706d
sample_fmt=fltp
sample_rate=44100
channels=2
channel_layout=stereo
bits_per_sample=0
initial_padding=0
id=0x1
r_frame_rate=0/0
avg_frame_rate=0/0
time_base=1/44100
start_pts=2788
start_time=0.063220
duration_ts=435455
duration=9.874263
bit_rate=127706
max_bit_rate=N/A
bits_per_raw_sample=N/A
nb_frames=378
nb_read_frames=N/A
nb_read_packets=378


  


  Another developer he uses his version of ffprobe :

  


  ffprobe version 2023-02-22-git-d5cc7acff1-full_build-www.gyan.dev Copyright (c) 2007-2023 the FFmpeg developers  


  


  Based on the year, my version(2024) should be newer than his(2023), but his version of ffprobe can get the audio codec properly :

  


  [streams.stream.1]
index=1
codec_name=aac
codec_long_name=AAC (Advanced Audio Coding)
profile=LC
codec_type=audio
codec_tag_string=mp4a
codec_tag=0x6134706d
sample_fmt=fltp
sample_rate=44100
channels=2
channel_layout=stereo
bits_per_sample=0
initial_padding=0
id=0x2
r_frame_rate=0/0
avg_frame_rate=0/0
time_base=1/44100
start_pts=1764
start_time=0.040000
duration_ts=436480
duration=9.897506
bit_rate=111733
max_bit_rate=N/A
bits_per_raw_sample=N/A
nb_frames=427
nb_read_frames=N/A
nb_read_packets=427
extradata_size=5


  


  Why ?

  


  I also tried a ffprobe version on ubuntu with the following version :

  


  ffprobe version 6.1.1-3ubuntu5 Copyright (c) 2007-2023 the FFmpeg developers
  built with gcc 13 (Ubuntu 13.2.0-23ubuntu3)
  configuration: --prefix=/usr --extra-version=3ubuntu5 --toolchain=hardened --libdir=/usr/lib/x86_64-linux-gnu --incdir=/usr/include/x86_64-linux-gnu --arch=amd64 --enable-gpl --disable-stripping --disable-omx --enable-gnutls --enable-libaom --enable-libass --enable-libbs2b --enable-libcaca --enable-libcdio --enable-libcodec2 --enable-libdav1d --enable-libflite --enable-libfontconfig --enable-libfreetype --enable-libfribidi --enable-libglslang --enable-libgme --enable-libgsm --enable-libharfbuzz --enable-libmp3lame --enable-libmysofa --enable-libopenjpeg --enable-libopenmpt --enable-libopus --enable-librubberband --enable-libshine --enable-libsnappy --enable-libsoxr --enable-libspeex --enable-libtheora --enable-libtwolame --enable-libvidstab --enable-libvorbis --enable-libvpx --enable-libwebp --enable-libx265 --enable-libxml2 --enable-libxvid --enable-libzimg --enable-openal --enable-opencl --enable-opengl --disable-sndio --enable-libvpl --disable-libmfx --enable-libdc1394 --enable-libdrm --enable-libiec61883 --enable-chromaprint --enable-frei0r --enable-ladspa --enable-libbluray --enable-libjack --enable-libpulse --enable-librabbitmq --enable-librist --enable-libsrt --enable-libssh --enable-libsvtav1 --enable-libx264 --enable-libzmq --enable-libzvbi --enable-lv2 --enable-sdl2 --enable-libplacebo --enable-librav1e --enable-pocketsphinx --enable-librsvg --enable-libjxl --enable-shared
  libavutil      58. 29.100 / 58. 29.100
  libavcodec     60. 31.102 / 60. 31.102
  libavformat    60. 16.100 / 60. 16.100
  libavdevice    60.  3.100 / 60.  3.100
  libavfilter     9. 12.100 /  9. 12.100
  libswscale      7.  5.100 /  7.  5.100
  libswresample   4. 12.100 /  4. 12.100
  libpostproc    57.  3.100 / 57.  3.100


  


  It will detect the audio as aac properly, but with different parameters, for example, bit_rate is 111733(developer) but 110399(ubuntu). But this parameter comes from the same file so should be the same.

  


  [streams.stream.1]
index=1
codec_name=aac
codec_long_name=AAC (Advanced Audio Coding)
profile=LC
codec_type=audio
codec_tag_string=mp4a
codec_tag=0x6134706d
sample_fmt=fltp
sample_rate=44100
channels=2
channel_layout=stereo
bits_per_sample=0
initial_padding=0
id=0x2
r_frame_rate=0/0
avg_frame_rate=0/0
time_base=1/44100
start_pts=0
start_time=0.000000
duration_ts=441353
duration=10.008005
bit_rate=110399
max_bit_rate=N/A
bits_per_raw_sample=N/A
nb_frames=432
nb_read_frames=N/A
nb_read_packets=432
extradata_size=5