Comment by TeMPOraL | Hacker News Reader

Comment by TeMPOraL | Hacker News Reader

Parent Full thread

> The idea was to put a rocker switch on the side, to rock between "dot" and "dash", and then use Morse code. To "read" Morse, the phone could vibrate, or at least decode it as text to read. You could text without walking into a tree (which I have done.)

I can imagine this working. Kind of like CW for HAMs, except with a 3-state switch and couple dozen more layers of abstraction between you and the wire :). I wish my Morse code wasn't so rusty though - and it makes me wonder, are there better text encodings that could be used for such purpose?

The method for texting without looking I described was only limited to writing, but it's also something I've done in practice before smartphones, in particular on a Sony Ericsson K800i. This was possible because of good-quality keyboard (and joystick), and good UI decisions. I don't know if they put an actual soft-realtime OS in there (I suspect so), but all the UI delays were always the same - so I naturally memorized sequences like "unlock, menu, texting, wait ~0.5 second for it to load, open thread, type in message, send". It probably takes teenage-level or business-professional-level of use to get that kind of muscle memory, though.

> Instead of dividing up the picture into rectangles, divide it up into triangles. (...) This will prevent any banding or that ugly blockiness.

Is the insight here that a triangle-based subdivision like this will make areas of detail not align on horizontal/vertical lines, unless the underlying image has its details aligned on a band? Or am I misunderstanding this?

> The method for texting without looking I described was only limited to writing, but it's also something I've done in practice before smartphones, in particular on a Sony Ericsson K800i. This was possible because of good-quality keyboard (and joystick), and good UI decisions. I don't know if they put an actual soft-realtime OS in there (I suspect so), but all the UI delays were always the same - so I naturally memorized sequences like "unlock, menu, texting, wait ~0.5 second for it to load, open thread, type in message, send". It probably takes teenage-level or business-professional-level of use to get that kind of muscle memory, though.

Wow, this threw me back. :) Indeed this was not only possible, but actually easy. I don't think it took a lot of effort or any conscious effort even, actually, the UI was so good, but most importantly there were no unpredictable delays and no unpredictable inconsistencies!

Why is modern software so bad in comparison? Sure it can do more things, but it does them worse. The interface of everything is constantly changing for the unfounded whims of designers who simply think "this looks prettier to me", things are not laid out logically, adverts permeate the interface of most products from phones to TVs...

I've never attempted to learn Morse, but it looks like a few hours of practice should do it. Anyone can learn it, but I expect younger people would be more willing to do it.

> the insight

The vertices will be on pixel addresses. The edges of the triangle will have no discontinuities themselves, and the triangles sharing an edge will have no discontinuities. Hence, no banding or blockiness. A naive renderer would just render the colors of the triangle as a straight linear interpolation. A more advanced one could look at the adjoining triangles, and render it so the first derivative would also be zero across the line. I.e. curve fit it.

The selection of where the vertices go would likely be best selected where the colors are shifting fastest, i.e. an edge in the picture.

The neato thing is one could render a low res image on a high res display and it'll still look good. You wouldn't have these stray pixels scattered around the text like jpg does. The other neato thing is more vertices can be put in the complex areas of the image, rather than wasting bandwidth on the parts of it that are smooth.

Progressive sharpening of the rendering also works, as more vertices can be added at any time.

Hexagonal, equivalently, equilateral triangular tessellations indeed have many advantages over using squares. Your scheme reminded me of Delaunay triangulation and its dual, Voronoi tessellation.

I have a faint suspicion that your familiarity with finite element models had something to do with your choice of triangles.

Triangles are used by game engines for good reason. Hexagonal cannot be mapped without discontinuities. Equilateral triangles are not necessary.

I've done finite element models, but they were pretty unsophisticated.

That's right, simplicial complexes of the right dimension would give one the most fidelity in terms of continuity. Triangles happen to be the one for 2D. For 3D one would get similar benefits using tetrahedra.

I was alluding the fact that barring boundary effects hexagonal tessellation and equilateral tessellation are essentially the same.

The method actually reminds me of a simple h-refinement scheme for FEM.

Reply on news.ycombinator.com