I was hoping there would be some whacky embedding in 2.5D space for this, but it just seems randomly assigned.

A lot of AI agent swarm management is being automated. People keep pitching these to me with a straight face.

This is covered in the plot of Terminator 2. The liquid metal tries really hard.

yes! I learned React first, and then Svelte. Svelte seems to have all the benefits without any of the React abstractions that are more pain than they’re worth.

https://en.wikipedia.org/wiki/Zipf%27s_law may be what you're looking for

It's simpler. The verbosity, the line of code count, is what has gotten people excited about it. So everyone optimizes for verbosity.

This does not strike me as true at all. Nobody's crowing about the verbosity of their AI-generated code. Rather, it seems to me they're impressed by the functionality or capabilities relative to time/cost.

You're thinking of making a manual annotation front-end to add the action points as determined by refs?

No, action points I get either from parsing a video capture of a scoreboard or by pulling them over API, depending on which scoreboard/match management system is used.

I do plan to have a front-end for manual annotation of referee performance: correct relative position, timing, determining attacking/defending wrestler etc.

Input: video of an exercise/workout/dance. Output: single collage image as summary.

Me+team have been working on a large scope problem this past year. We're hardening some of our internal tools, and choose this small-scope problem about a month ago. Results have been encouraging so far, though not yet non-sloppy enough to share. We'll release in about a month.

Going into this, I thought I'd be spending most of my time tuning computer vision. Instead, the majority of the time has been handling codec and ffmpeg behaviour edge cases.

this is interesting for sports i'm involved in. does it sort images by various symmetries? being able to "green line" (don't search that) an image to show where the expected vs. realized symmetry should be would be very useful for athletes.

Which sports are you into? I'll be looking for closed alpha testers shortly. Email me at dustin at hyperpan.cv

Automatically drawing hint lines has been a request from a few people. When I talk to personal trainers or athletes (new or serious) in various sports, I've tested out this concept by asking them what they'd want drawn. Turns out, it's so contextual, opinionated and subjective, I don't think there's a great approach unless I assume we have an unrealistic amount of context.

e.g., one case is kettlebell swings. You might want to see how consistent your swings are in a set, and between sets. To do this well, you'd need to reliably detect the kettlebell swing exercise using some N-way classifier. And then, correct for camera position. And then, center the relevant part of the body (is this shoulder? hips? turns out people have different opinions).

Instead, we're going to use a more generic visualization approach that doesn't try to be opinionated. But, I'm expecting to open-source the visualization front-end so it's tweakable by users.

a back squat form checker would be a product in itself. dressage is probably too niche, but an interesting control or test case. yoga would be a large market.

Oh man, this sounds fun. But, the button is stuck at "Loading" and never completes. Is HN hugging it to death?

Yeah, I think so. It's running on a pretty small VPS and I never implemented any caching. Should have thought about that before posting I guess. I see the CPU is currently pegged. I was able to get it to load at least 1 trip myself though, so maybe retry in a bit

We could make a license plate big enough that aliens could see it and know we are sophisticated.

Nah, let's just put a big sign "Kilroy was here", so they don't bother with the "mostly harmless" planet.

If four out of seven digits were blurred out, then they would really see how sophisticated we are: we have the concept of privacy!