Date: Tue, 21 Jun 2022 01:36:40 -0400
From: Luke Iannini
Subject: Re: The world's tiniest page
We have replication at Fremont lab!

Just some notes on prerequisites if you want to try it out — the per-camera 4K switch has been in the console since last August or so, but you'll need the Kurokesu-upgraded BRIOs to use their optical zoom to get the best pixel coverage. You need to manually update "max dot size" parameter to see classic pages from the usual 15 to 30 to account for the extra pixels in 4K mode at the moment. Full performance perhaps relies on the new Phantom Canyon NUCs, but once you have all nano-scale pages you can crank down the "max dot size" parameter back down to get a speedup, and I'm planning on trying out a down-resolution dot pass to catch classic dots without needing to march every pixel.

Shawn also reports that nanopages require more light to recognize, which makes sense and you can counter it by cranking the physical aperture on the Kurokesu lens open, increasing exposure time, and with a silly-but-incredibly-effective new feature I'll be sending along with the next release.


On Jun 21, 2022, at 12:36 AM, Luke Iannini wrote:

Sorry, make that 1x1 stamp sized… the stamp was too big and I had to trim it to get it to fit… I give you the Dynastamp
<image1.jpeg>

On Jun 21, 2022, at 12:31 AM, Luke Iannini wrote:

I mean come on
<image0.jpeg>

And yeah, they totally track
<Video.mov>


On Jun 21, 2022, at 12:20 AM, Luke Iannini wrote:

With a second pass at Scene Kit shaping up nicely (i.e. '3D proteins on the table' and friends, release coming shortly!) I've been unable to stop myself from going full steam ahead into Recognition Kit 2: This Time It's For Real.

4K CV pipelines are coming along great, I've got a SIFT prototype functional, and the new shape matching algorithm has begun to introduce its first shapes to each other, so I wanted a peg in the ground on Apriltags to finish the picture... but as I pondered how small we could get the tags... I had to know... just how small of a dot can we actually see at 4K?

To the Print Proofer! First I printed the Realtalk equivalent of a visual acuity test...
<IMG_8842.jpeg><IMG_8844.jpeg><IMG_8845.jpeg>


Looks like we can see line 3 just fine! (Line 2 is semi-functional with dots flickering in and out, Line 1 is a goner)

Well dang, that looks pretty good. That seems to be right at the theoretical 0.25 scale we should expect from a 1080p -> 4K jump... I wonder... I wonder if...
<IMG_8846.jpeg>


<IMG_8847.jpeg>


Now come on that's just silly. There's no way it can see that.

<IMG_8849.jpeg>


Well... I'll... be. That's the blank, that I wrote the page proof on, to generate a tiny version of itself, being recognized, and generating itself again (complete with tiny little blank page text).

I really thought I'd have to go fix some hardcoded values somewhere the dot frame recognizer, but no, it just works (I'm sure there's something in there to fix though!).

So... ~2x2 postage-stamp-sized fully-fledged Realtalk pages have been apparently been sitting waiting patiently for us since I did the 4K upgrade sometime last year! Amazing.