WASD custom keycaps for Das 4: review

RedGlyph

24 Jan 2015, 12:02

Since there are no international keycaps on the Das Keyboard 4, I finally decided to make a custom one. Started with the blank keys but it was sometimes a bit annoying when I had to type with one hand, for confirmation codes and so on. Right, the perfect excuse ;)

[url=http://"http://support.wasdkeyboards.com/"]WASD Keyboards[/url] made a wonderful job, overall I'm very satisfied! They did have to reprint a couple of keys, but the difference from the specs was very minor, barely noticeable, and they were quick and reactive to the problem, even gave away a few stabilizer inserts I needed for free :)

The material is pleasant to the touch, which was my main concern. It's still ABS but I find the touch feeling on par with the original keys from Das Keyboard, even slightly better, perhaps a tad rougher and thus less slippery due to the UV print process - that might not be true if you compare to the printed keycaps of the Professional version, I couldn't say since I have the Ultimate.

If you want to swap the keycaps on the Das, don't forget to get those Costar stabilizer inserts, you'll need 6 pairs (the spacebar has Cherry stabilizers and doesn't need any); Metadot had the stupid idea of gluing their inserts to the keycaps, so most of them can't be recovered! Other than that, the keycaps are completely compatible with the Das Keyboard.

The colours are quite correctly rendered, though if you look the keycaps from a low angle you may notice process artefacts. It looks a little like the result of a dot printer: successive horizontal bands. But it's only the background, and if you look at extreme angles under a grazing light source, under normal circumstances it is not visible. You can see what I mean on the right of the photograph below, especially the numpad return key. The foreground has zero defect or artefact, and it's much sharper than laser etching of course.

For reference, the font I used is Overlock, this is an open-source font, easy to find. Don't forget to remove the fonts from the .svg when you send it to WASD, it's not mentioned on their website but you have to do it. Here's how:
In Inkscape:
Select all of your text/fonts, then go Path > Object to Path (Ctrl + Shift + C)

In Illustrator:
Select all of your text/fonts, then go Object > Expand… (Expand both Stroke and Fill)
AZERTY_HD.jpg
AZERTY_HD.jpg (239.43 KiB) Viewed 5906 times
I'm really happy with it, the only small regret is the initial choice of an Ultimate, for some obscure reason they decided not to print the "daskeyboard" logo and the media keys, the Professional version is looking better IMHO. At least they kept the red edge of the volume button.

User avatar
Daniel Beardsmore

24 Jan 2015, 14:38

On the assumption that it looks exactly like the photo, with fairly muted colours, that scheme works really well — very tasteful. However, I suspect that the JPEG compression has affected the colour and that in reality it's a lot less subtle (I always leave chroma sub-sampling disabled with JPEG images to avoid excessive colour damage).

It also comes as no surprise that Inkscape's terminology is far more logical than Illustrator.

RedGlyph

25 Jan 2015, 11:13

The photo hasn't been altered, though the light conditions gave a feeble blue-ish touch to the colours, it's snowy winter.

Chroma sub-sampling makes sense to compress images, given the rate you target it's an efficient way to remove information vs the perceived quality of the result (which is arguably subjective and hard to quantify). Disabling it means more alteration to the luminance, which will be worse. Since it was a quick shot and honestly I'm not equipped to take professional photographs, I didn't care to extract the raw file and apply a lossless compression, I don't think it would have added much to the post anyway :)

User avatar
Daniel Beardsmore

25 Jan 2015, 14:15

Why would disabling chroma sub-sampling cause more alteration to the luminance?

User avatar
Muirium
µ

25 Jan 2015, 15:14

Zero sum game. When compressing to JPEG and other lossy formats, images are split into luma and chroma components, and the luma is given a lot more bandwidth because our vision (from the retina to the large volume of brain set aside to it) is much more sensitive to that domain. We essentially see in monochrome with colours laid on top. We notice edges moving around physically far more than colour gradient changes. Chroma can be compressed quite dramatically with seemingly zero degradation in image quality.

And before you think raw files are the saviour: camera image sensors are also designed to physically reflect this priority. They are higher resolution in luma than chroma. The latter being averaged over several neighbouring luma pixels.

Anyway, this is all really about white balance. And it's independent of compression. Even an ideal camera could shoot stinky hues if the light's not there!

User avatar
Daniel Beardsmore

25 Jan 2015, 16:17

OK, so why does disabling chroma sub-sampling result in a larger image? The resulting increase in image size is far less than if you raised image quality instead, so for me it's a better compromise.

Digital cameras are certainly not designed to reflect this priority. The problem with digital cameras is that they sense luma, so colour filters are required to restrict the sensing by colour, and that comes at a cost of lost accuracy and lost light. The Foveon X3 sensor can sense all three primary colours at the same spot, however.

And no, it's nothing to do with white balance. My query is whether the JPEG compression has muted the colours, because the colour areas are so small.

User avatar
Muirium
µ

25 Jan 2015, 16:23

My brother works in the image sensing industry, and although I'm not well versed in the details myself, I can assure you that claims about cameras having 1:1 chroma to luma resolution are bollocks. Design is about trade offs on every level. And so all current (and developing) technologies concentrate heavily on luma, and interpolate chroma separately. The various software libraries in use across the industry for making RAW files at the firmware level and for applying noise reduction and dead pixel masking are also built on this assumption.

User avatar
Daniel Beardsmore

25 Jan 2015, 17:14

Wait, what claim are you refuting here?

User avatar
Muirium
µ

25 Jan 2015, 17:25

Okay, we're not on the same frequency at all. I'll cut out!

RedGlyph

26 Jan 2015, 19:18

Daniel Beardsmore wrote: OK, so why does disabling chroma sub-sampling result in a larger image? The resulting increase in image size is far less than if you raised image quality instead, so for me it's a better compromise.
I'm not sure how you quantified that in order to reach this conclusion.

If you stress the compression ratio, you'll see that keeping the chroma & luma on par will see the overall quality collapse very badly, wheras chroma subsampling maintains an acceptable quality level. At some point you can't even compress to some rates unless you subsample them.

I've done some tests at higher rates and the distance was still better by subsampling, at equal rates. I did it on R, G, B values however, so here I'm looping back to the quality criterion, there are many ways to quantify the quality of compression, none is universal, neither perfect. But at the end of the day, chroma subsampling is more robust.

User avatar
Daniel Beardsmore

26 Jan 2015, 22:35

I'm about to hit integer overflow on the number of things I encounter that cannot be explained by any available knowledge.

There are two key problems that I discovered, and one of them led me to find that chroma subsampling was the solution. I think I took the problem to Thorsten Lemke, who I presume suggested disabling what I imagine was just called "susbsampling" in GraphicConverter.

One of the two issues is occurs when trying to save screenshots to JPEG, typically those of high-colour desktops where PNG wasn't having much effect (this was years ago and I probably still had dial-up). It would cause primarily red to smear out all around small red objects, such as icons. I would have to ramp up the compression level drastically to remove that mess, while disabling chroma subsampling (possibly something Thorsten Lemke taught me?) would prevent the smearing without increasing the file size anywhere near as much.

The other one is red corruption. Take the attached image, that I stole from some post recently (attached to stop phpBB from altering it).

Observe how the top of the keycap beyond the cherry exhibits banding. There are dark bands across it. This is caused by chroma subsampling (they're not there if you convert to greyscale), which for some reason always causes large red areas to show dark bands and blotches. If you imagine what the image started out like, there's no reason for the chroma to be downsampled to include dark bands. I've even seen this across huge red areas with nothing dark in sight.

This is probably the issue that I showed Thorsten when I wanted to know what was going on.

There are some pretty nasty problems caused by chroma subsampling, and the drop in image size isn't great enough for me to worry about switching it on and off on a per-image basis. I leave it permanently off. Quality of course is program-specific; I use IrfanView, and 75% quality for most images.

It's been years since I studied the effects of quality and subsampling, but I found that subsampling just did more harm than it ever did good.

Of course, it wouldn't surprise me if my visual perception is different to everyone else's.
Attachments
Eye test.jpg
Eye test.jpg (20.88 KiB) Viewed 5752 times

User avatar
Muirium
µ

27 Jan 2015, 13:18

That image deserves an award for incompetence! A well composed plate of horrible.

I see the dark junk on the red area, too. But look at the noise in the grey background. That noise is because whoever shot this did so with about a tenth as much light as they should have. Yet noise is actually quite demanding of a compressing algorithm. You know how percussion, and especially cymbals (which produce a nice rich range of noise) squelch badly in low bitrate mp3s? In fact, they drag down the quality of the other instruments whenever they are hit. The same is true for all high frequency domain stuff in digital compression. And camera sensors produce a load of it, which dominates the true signal when underexposed. That noise wasn't filtered out in this case, so it hogged the bandwidth! And I suspect it's smeared across the red area, leaving that wake of destruction.

Red and blue are the hardest colours for sensors, of course. As they lie at the extreme ends of the RGB sample range. Pure red is black in two channels. And black is badly affected by foolishly compressed noise, which is what I suspect is going on here.

User avatar
LLRnR
\m/

27 Jan 2015, 13:46

Muirium wrote: That image deserves an award for incompetence! A well composed plate of horrible.
Gee, I didn't take that photo, but this is precisely the kind of comment that makes me think twice about posting one. :( I have a very modest camera so I seldom take photos, anyway.

User avatar
Muirium
µ

27 Jan 2015, 13:52

Nothing personal. We all take shitty photos. That's how you learn to take ones that suck less, until they're good!

What that picture needs is LIGHT. Cropping out the background would be good, too. And maybe getting closer to the cap, if at all possible with the current lens. Three big things, dominated by one that's very wrong indeed.

To their credit, the angle's nice and straight! Usually, bad pictures are shot all skewed, by someone who very clearly doesn't care about what they're doing. (Ebay sellers!!) This one makes me think someone was trying, and could improve.

User avatar
Daniel Beardsmore

27 Jan 2015, 20:19

I am not sure why the person who took that photo is due so much ridicule. It's not fair to expect everyone to be a visual person, and it's just a random photo from the classifieds that caught my eye because of the damage caused by chroma subsampling. It's not like it's a wiki photo!

Here's an example that doesn't involve noise. Simple test image involving red and white, which you can inspect.

I've saved two JPEG images from it, one with chroma subsampling enabled (at whatever setting IrfanView uses), and with it disabled.

With chroma subsampling enabled, all the red areas get an inner fringe of dark red. Also, you'll see that the chroma spills out around the red areas, with the white becoming pale red. With chroma subsampling disabled, the edges are sharper and cleaner, with no red spill and no dark fringing.

I don't pretend to understand it; I only know empirically that chroma subsampling causes undue damage to images. This is why I expressed concern that the legends shown in the photo may not be the colour you see, because of alteration of apparent luma, as well as chroma spill, in addition to the obvious problems with JPEG at any hard boundary.

There are so many things that aren't explained. Here's another one:

http://telcontar.net/Misc/random/colourblends.php

Most people blend colours in RGB mode, but the midpoint of red-blend-green usually comes out brown, with the apparent luma not preserved. Windows GDI has a colour blending algorithm that provides correct apparent luma — I'm guessing it operates in a different colour space to RGB. In greyscale, the effects of an RGB blend looks fine, but to the eye, it looks fringed.
Attachments
Source image in lossless format
Source image in lossless format
Source image.png (1.33 KiB) Viewed 5689 times
IrfanView JPEG, 75%, chroma subsampling enabled
IrfanView JPEG, 75%, chroma subsampling enabled
Test image 1 -- subsampling enabled.jpg (8.1 KiB) Viewed 5689 times
IrfanView JPEG, 75%, chroma subsampling disabled
IrfanView JPEG, 75%, chroma subsampling disabled
Test image 2 -- subsampling disabled.jpg (12.2 KiB) Viewed 5689 times

User avatar
Muirium
µ

28 Jan 2015, 22:01

Daniel Beardsmore wrote: I am not sure why the person who took that photo is due so much ridicule.
Not the person, but the picture. We all shoot shitty photos. But without feedback, we might never improve any skill. My criticism is always with suggestions for practical improvements, if I know of any. I've shot a lot worse than that, after all! Namecalling, meanwhile, is pure bullying and best left to tearful boys…

Seeing your objective tests, here's a couple of jpegs I just made on Mac OS X from your png reference file:
Red 20k.jpg
Red 20k.jpg (19.86 KiB) Viewed 5654 times
Red 12k.jpg
Red 12k.jpg (11.86 KiB) Viewed 5654 times
Can your app tell you if they have the chroma downsampling you're focussing on? Because the only difference between them that I made was pulling down Preview's slider to force a smaller file size with the second one. Looks like simple over-compression to me.

The mathematical details of how jpeg goes about this are on Wikipedia. All looks quite hairy and well worth using wisely, in moderation instead of extreme!

User avatar
seebart
Offtopicthority Instigator

28 Jan 2015, 22:34

I´m not going to get into any technical arguments here but as some of you know I shoot quite a bit of pictures and I have to agree with Mu the cherry keycap picture is lousy. Less so the composition or subject but rather the technical side of it. The whole image is one carpet of noise.
I am not sure why the person who took that photo is due so much ridicule.
It´s not about that at all, it`s not about the person, it´s about the picture.

User avatar
chzel

28 Jan 2015, 23:13

I haven't posted many photos myself, so surely my worse is still to come!
Muirium wrote: <snip>...We all shoot shitty photos..<snip>...I've shot a lot worse than that, after all!
May I propose this example of dreadful noise?
Image

EDIT: Scottish weather is not an excuse!!! Get some legs!

User avatar
seebart
Offtopicthority Instigator

28 Jan 2015, 23:18

chzel wrote: I haven't posted many photos myself, so surely my worse is still to come!
Muirium wrote: <snip>...We all shoot shitty photos..<snip>...I've shot a lot worse than that, after all!
May I propose this example of dreadful noise?

EDIT: Scottish weather is not an excuse!!! Get some legs!
Uhhh chzel quite tough tonight, I wonder why? :mrgreen:

User avatar
chzel

28 Jan 2015, 23:21

Just venting some built-up steam on unsuspecting innocents! ;)
Muirium don't take my post too seriously! :oops:

User avatar
Muirium
µ

29 Jan 2015, 00:48

Like I said in the quote you excerpted! Mind, I still like that picture. The noise is jarring in the shadow, and I would be better off shooting with a tripod (I have one, but setting it up…) in the long, low winter light. But I'm a sucker for bokeh, which that shot has in spades, as well as the physical sense of that wee board in this murky season.

User avatar
chzel

29 Jan 2015, 01:14

I like it too, although I think I'd like it more if the focus was on the Esc. That F-row leads my eye to an out of focus cap which dominates with its colour! And that bokeh is indeed sweeeeet!

On a side note, we are really really good at steering threads off topic!

User avatar
Muirium
µ

29 Jan 2015, 01:27

That's the fun in this place! All the best stuff is OT.

The focus in that picture is intentional. I shot it to illustrate Mr Interface's Alps to MX adapters. He's making a second batch sometime soon, which I'll gladly test again and compare with the (tight) originals.

The reason I got so much bokeh is I shoot in shutter priority mode when the light's low, so I don't get motion blur, and the camera was compensating for the gloom with a full open f/2.8

RedGlyph

17 Feb 2015, 22:03

Daniel Beardsmore wrote: It's been years since I studied the effects of quality and subsampling, but I found that subsampling just did more harm than it ever did good.

Of course, it wouldn't surprise me if my visual perception is different to everyone else's.
It could, but not that much ;) I found this document today, and thought it could help shed some light on the physiological understanding of why chroma subsampling is a valid strategy.

Post Reply

Return to “Keyboards”