toerror

Software developer, Manchester UK. On extended sabbatical from corporate dev while I work on some personal projects, but it looks like someone set fire to the real world while I've been busy... Currently working on a space game project that is frankly too large for one person.

Cryptomnesia: sorry
Pronouns: he/him.

Last Notes

2026-08-08 13:42:18 UTC

- reply

I found with height field data, it tends to smush together the quantisation stair cases as if they were plasticine, and you pushed across them smoothing them out, but you can find it wrecks other details that you'd rather keep.

2026-08-08 13:40:46 UTC

- reply

So "Bilateral" filtering is essentially an edge preserving blur. It performs a bunch of samples and integrates them together like a weighted Gaussian blur that preserves details.
Here's an example performing an image blur -
https://www.shadertoy.com/view/4dfGDH

2026-08-08 13:08:40 UTC

- reply

Ah, fair enough - I ended up implementing bilateral to get rid of exactly that quantisation, which it does with some success ( not complete removal ), but it does also attack desirable details to a slightly lesser degree. In the end I managed to get higher res data which removed the need, so if that's an option then that's the best approach. Worth remembering though if you find yourself stuck with display data.

2026-08-08 11:47:46 UTC

- reply

It might be worth experimenting with bilateral smoothing of the height field data to try to smooth out some of those more flat regions so you don't get as much of that layering kind of effect. ( Apologies if you've tried this.. )

2026-07-15 12:12:37 UTC

- reply

Yeah - I think there are different categories of these things - flat earth strikes me as a bit of a cult where membership has a requirement that you be willing to prove your devotion by rejecting reality, or the reality of those outside the cult. This also has the effect of alienating members from the outside world.

2026-07-15 12:01:16 UTC

- reply

Off topic, but my theory is it's because they're unprovable - or at least, it's the unprovable ones that stay around as conspiracy theories as opposed to established facts and therefore have longer to gain social inertia than other things which lose interest once they've been clarified.
Conspiracy theorists them selves appear to be attracted to the things that make them paranoid, reinforcing their anxiety surrounding the topic in question.

2026-07-02 09:06:11 UTC

- reply

Inference costs aren't negligible at present though - or do you mean for limited context single queries as opposed to e.g iterative vibe coding?

2025-09-23 12:05:26 UTC

- reply

It looks like a project designed to fail after everyone concerned has got rich and retired.

2024-08-16 12:42:16 UTC

- reply

They look like a horror concept.. I wonder if the original idea was something to do with the creepy security risks of smart appliances, and someone got hold of the concept and recycled it out of context. It just seems like such a weirdly inappropriate idea to come up with out of the blue.