Thank you for your donation!


Withdrawn: No more blurry album covers!
#31
(09-18-2020, 01:56 AM)fmaxwell Wrote:
(09-17-2020, 10:14 PM)swizzle Wrote: The max image size is the space allocated on the screen to display the image, if you have a 2000x2000 image it’s not going to look any better than 700x700 if 700x700 is all the space there is. There may be some slight super sampling benefit from scaling down wrt smoothness but I think sharpness is the desire here.

The limitations in moOde 6.7x doesn't mean that 7.x and/or later can't be enhanced to take advantage of 8K or larger monitors using a larger size (the largest of which I suggested was 1000x1000).

Quote:[quote pid='24857' dateline='1600380840']
There will always be some degree of scaling because the image won’t ever be exactly the size of the space allocated for it, downscaling is obviously better than upscaling.

That's why it would be better to downscale it from an 800x800 thumbnail than to upscale it from a 400x400 thumbnail.

Quote:We don’t want people thinking they want maximum quality and choose the highest possible settings and end up with a slower system than if we’d just used the orginal cover art. 

All you have to worry about as a developer is choosing reasonable defaults that will work well for most users with typical setups and personal priorities.  Users of video games adjust anisotropic filtering, full screen anti-aliasing, texture resolutions, motion blur, view distance, fog quality, ambient occlusions, and shader quality, all while balancing off performance vs. display quality. So I think that moOde users can be trusted to adjust album art thumbnail size and quality, especially since it's only those users who are dissatisfied with the default settings.

Quote:A png thumb would be > 1MB for 800x800 and again wouldn’t offer any benefit over the original cover art.

I'll give you a perfect example of where an 800x800 PNG has an advantage.  Here's a 3.2MB, 1500x1500 JPEG of the album art for the soundtrack of 2001: A Space Odyssey.

Resized to an 800x800 PNG, it clocks in at under 1.5MB.

PNGs offer advantages over of JPEG, especially in images that include text, like most album covers.  

[Image: jpg_vs_png.png]

JPEG, unlike JPEG XS. is not designed for high multi-generation robustness.  It's DCT data compression algorithm is optimized for a single generation encoding from a lossless image source, much as MP3's data compression is optimized for a single generation encoding from a lossless audio source.  As you make JPEGs of JPEGs, you increase the artifacting in ways that become more visually apparent.

Note:  I understand that PNG is off the table for reasons of code complexity.  Nothing in this message should be interpreted as my suggesting that is the wrong choice.  It's just a technical discussion.
[/quote]

*** For some reason the quote levels on this are messed up ***
*** Best to reply not using quote ***

The gamer space is not really a good comparison to the moOde space since gamers are heavily $$$ invested into the highest levels of tech and are really into and benefit from all the options and tweaks.

The moOde space is one where we have to make tradeoffs in some of the features we offer so they work well on a variety of tech and don't create support nightmares.

The image you provided is super nice so let's see what happens at various resampling resolutions and qual levels.
Reply
#32
(09-18-2020, 02:25 AM)Tim Curtis Wrote: Running an Open Source project means relying on others. I do my part of the dev on Mac/Safari. Other parts are done by the moOde crew in their own dev environments. The Open Source community and our nice community of moOde users provides other parts of moOde.

No need to be shocked :-)

The Tongue was meant to indicate some humor in that statement.  I've done the engineering product development from idea to sales (small scale) a few times so I know the deal.  Opposite of you though Apple was always the annoyance to me as I didn't own or use any of their stuff and they always had to do things differently.  Even acquiring a Mac to test I still needed people who used the things more regularly to say yeah it's ok. 

That's why I tried to see if the problem happened on the Moode local display.  I figured it was the most likely common ground.

And somewhat on topic to the thread, I've generally found that when going from larger to smaller image size I often think things look better if some wavelet sharpening is applied after size reduction.
Reply
#33
(09-18-2020, 02:48 AM)Tim Curtis Wrote: The gamer space is not really a good comparison to the moOde space since gamers are heavily $$$ invested into the highest levels of tech and are really into and benefit from all the options and tweaks.

There's plenty of overlap and I'm a good example of that, being tech savvy and heavily invested in both high-end audio/video and high-end computing.  And there are plenty of casual gamers who just play the games at the default settings. 

Much of the tech support concern can be addressed with informational text on the setup screens that says something like:  "Note: Larger, higher quality images can result in slower performance and greater RAM usage without a significant visual improvement."

Alternatively, you could ask a question of the users like "According to Nyquist's Theorem, what is the highest audio frequency, in KHz, that can be encoded with 48KHz sampling?"  If they answer wrong three times in a row, put up a message that reads "You don't need to access the hidden controls. Just go back to the album view and look at all the pretty pictures."  
Smile 

Quote:The image you provided is super nice so let's see what happens at various resampling resolutions and qual levels.

It's the first album shown in my alphabetized library and it also really jumped out at me as being degraded and soft in the thumbnail view.  The quantity of small text, the stars, the image complexity and contrast, the hard lines, and the levels of detail provide a real challenge for resampling and compression algorithms.

albumartexchange.com is the best source of high-quality album art that I've found.  I can hunt up more examples if you want, but if you get the 2001 soundtrack album art looking really good. I expect everything else will look fine, too.  

Thanks, Tim.
Reply
#34
(09-18-2020, 03:27 PM)fmaxwell Wrote: Alternatively, you could ask a question of the users like "According to Nyquist's Theorem, what is the highest audio frequency, in KHz, that can be encoded with 48KHz sampling?"  If they answer wrong three times in a row, put up a message that reads "You don't need to access the hidden controls. Just go back to the album view and look at all the pretty pictures."  
Smile

That's a vague and tricky question.  Nyquist's theorem says nothing about audio frequency in particular as that's a biology thing about what's audible and for sampling in general Nyquist's theorem does not imply an absolute maximum frequency that can be sampled and reconstructed for a given sample rate.
Reply
#35
(09-18-2020, 04:33 PM)seashell Wrote:
(09-18-2020, 03:27 PM)fmaxwell Wrote: Alternatively, you could ask a question of the users like "According to Nyquist's Theorem, what is the highest audio frequency, in KHz, that can be encoded with 48KHz sampling?"  If they answer wrong three times in a row, put up a message that reads "You don't need to access the hidden controls. Just go back to the album view and look at all the pretty pictures."  
Smile

That's a vague and tricky question.  

It is neither vague nor tricky.  The answer is 24KHz.  No album art controls for you!

Quote:Nyquist's theorem says nothing about audio frequency in particular as that's a biology thing about what's audible and for sampling in general Nyquist's theorem does not imply an absolute maximum frequency that can be sampled and reconstructed for a given sample rate.

You are misinformed.  The Nyquist theorem states unequivocally that an analog signal waveform can be reconstructed from samples taken at equal time intervals if the sampling rate is equal to, or greater than, twice the highest frequency component in the analog signal.  It applies to all frequencies, including audio.

See:  http://sites.music.columbia.edu/cmc/Musi.../02_03.php

Therefore, the maximum frequency you could record with a 48KHz sampling rate is 24KHz, a sound outside of your hearing range but well within the hearing range of many animals including dogs and cats (https://upload.wikimedia.org/wikipedia/c...ge.svg.png).
Reply
#36
Sadly no, that's the common misunderstanding of Nyquist's theorem. It is not based on the highest frequency but rather the bandwidth of the signal. The sampling rate (for uniform sampling) must be greater than twice the bandwidth of the signal you wish to sample and reconstruct. If your signal happens to exist from 0 to (Sampling rate)/2 Hz you're all good. If your signal is centered higher than (Sampling Rate)/2 there are additional complications with aliases that must be considered to determine which sampling rates will work. But a trivial case with no worries about aliasing are the Nyquist bands. For 48 kHz sampling the Nyquist bands are 0-24kHZ, 24-48kHz, 48-72KHz,...,N*24Khz-(N+1)*24kHz where N is any integer >= 0. Of course the odd bands will be spectrally inverted but that's easily fixed. So with 48 kHz sampling rate I can easily sample and reconstruct an audio frequency that is greater than 24 kHz. I just have to bandbass filter the signal appropriately before I sample. (It's just another form of anti-aliasing filter than the the low pass that would be used for 0-24kHz.)

Look up bandpass sampling theory if you want more information. No this isn't typically used in the audio world but it is used commonly at RF frequencies.

And yes some animals can hear above 24kHz so in that context those are audio frequencies.

Thus your question is vague and tricky. Your question is actually along the lines of questions I use frequently to see how well DSP job applicants understand sampling theory. The 24kHz answer is a disappointing one.
Reply
#37
From the 2006 textbook DSP Software Development Techniques for Embedded and Real-Time Systems by Robert Oshana:

Quote:One of the most important rules of sampling is called the Nyquist Theorem. This theorem states that the highest frequency which can be represented accurately is one half of the sampling rate. 

This speaks to the author's expertise:  https://executive.engr.utexas.edu/bio/oshana.html

I asked about the Nyquist theorem, not the later Nyquist-Shannon Sampling theorem to which you were obviously referring.  Shannon contributed the idea that the sampling frequency needs to be at least twice the signal bandwidth and not twice the maximum frequency component.  It's what is commonly referred to now as "undersampling."  Your reply was a disappointing one.
Reply
#38
Ok, now you're getting into semantics of random text book authors, and I don't even mean the one you cited. "Nyquist's Theorem" as a name was a Bell Labs term applied after Shannon wrote his theorem, citing Whittaker's work. At least according to wikipedia, with references, but still wikipedia (so not the most authoritative). I'm happily not a theorem name historian. It seems you understand sampling. Good for you. Most people don't and think < fs/2 is THE RULE. Sad. But hey it's your question so fine I won't pick your album art size. You win.

I hope in the future you will cite the Nyquist-Shannon theorem so that you can help spread knowledge of the more nuanced and correct limit. Then maybe more people will answer correctly when I ask them how fast they have to sample the bandpass signal I draw for them. Too many textbooks get hung up on the semantics of "Nyquist's Theorem" and beat incomplete information into people's heads.

P.S. If we're really going to be pedantic, your question should use kHz, not KHz.
Reply
#39
(09-19-2020, 02:58 AM)seashell Wrote: Ok, now you're getting into semantics of random text book authors, and I don't even mean the one you cited.  "Nyquist's Theorem" as a name was a Bell Labs term applied after Shannon wrote his theorem, citing Whittaker's work.  At least according to wikipedia, with references, but still wikipedia (so not the most authoritative).  I'm happily not a theorem name historian.  It seems you understand sampling.  Good for you.  Most people don't and think < fs/2 is THE RULE.  Sad.  But hey it's your question so fine I won't pick your album art size.  You win.

I hope in the future you will cite the Nyquist-Shannon theorem so that you can help spread knowledge of the more nuanced and correct limit.  Then maybe more people will answer correctly when I ask them how fast they have to sample the bandpass signal I draw for them.  Too many textbooks get hung up on the semantics of "Nyquist's Theorem" and beat incomplete information into people's heads.

seashell, you're a good guy and a lot smarter than almost any I run across on the 'net.  I started my engineering career in the very early 1980s developing firmware and software for NIR spectrophotometers, so embedded systems and DSP are not new to me.  

The problem with citing Shannon when it comes to audio is that it's hard enough to get most people to understand Nyquist's theorem without throwing in concepts of undersampling and having the ADC serve a dual role in which it also acts as a downconverter or mixer.  Eyes just glaze over since many of them think that the analog waveforms coming out of CD players have stairsteps.

Quote:P.S. If we're really going to be pedantic, your question should use kHz, not KHz.

You are, of course, correct.  I've become sloppy on that and appreciate the reminder.  

P.S.  Robert Oshana is more than a random textbook author and you should check out his bio.  He's been in senior engineering positions at NXP, Freescale, TI, Raytheon, etc.  If you're an embedded systems person, you may well have read something by him at some time.
Reply
#40
It's all good @fmaxwell . Nothing wrong with a little technical debate from time to time. Thanks for being cool about it. That doesn't always happen on the internet especially.
Reply


Forum Jump: