正在加载视频...

视频加载失败

i built a real-time 3d positional tracker in the browser renders in ~8-10ms

35,343 次观看 • 1 天前 •via X (Twitter)

34 条评论

John Lindquist 的头像
John Lindquist1 天前

i did this 17 years ago 😉

vogel 的头像
vogel1 天前

nice!

John Lindquist 的头像
John Lindquist1 天前

i miss flash so much. imagine what astra could do with flash

vogel 的头像
vogel1 天前

try it out here:

maria 的头像
maria1 天前

ryan rouch, vogel browne

Salami 的头像
Salami1 天前

Stop with the indian ai linkden slop posts please there are multiple projects that already do this

Keegan 的头像
Keegan1 天前

Gotta pump those response numbers up

Caleb 的头像
Caleb1 天前

you should drop this as a space on 🤗

vogel 的头像
vogel1 天前

how

Caleb 的头像
Caleb1 天前

you can just ask your agent to deploy here, but saw the link above which is neat

vogel 的头像
vogel1 天前

nice:

🔥 nor 的头像
🔥 nor1 天前

which lab has the model "i" ?

Vladimir Cazacu | 𝕧𝕠𝕝𝕠𝕕▿ 的头像
Vladimir Cazacu | 𝕧𝕠𝕝𝕠𝕕▿1 天前

That's insanely slow. I was doing fiducial marker tracking via webcam at 2-3ms in 2007.

Dak 的头像
Dak1 天前

so cool haha

Arthur Katcher 的头像
Arthur Katcher1 天前

really cool

UNTITLED 的头像
UNTITLED1 天前

What if you put all frames into a buffer (seperrate mode), do the tracking and render per frame and then play those frames? It will be slightly delayed, but accuracy should be very high right?

R 的头像
R1 天前

ryanvoxel hehe

Gonza Pini 的头像
Gonza Pini1 天前

Welcome back Invizimals

bytesworms 的头像
bytesworms1 天前

so smooth 🔥

Kairo Salomo Chee Chin 的头像
Kairo Salomo Chee Chin1 天前

Interesting, i remember using some kind of synonym software to track my arm movements. completely forgot what it was.

Jonas 的头像
Jonas1 天前

How are you handling camera calibration? Normally for AprilTag tracking you take a bunch of photos of a calibration board at different positions and angles to calculate your camera intrinsics. If I were to guess, I assume you're just picking a reasonable assumption? That'll technically work, but AprilTag can actually have centimeter localization accuracy with good calibration, so it might be worth exploring using mrcal to generate good camera intrinsics.

Jason Ziegler 的头像
Jason Ziegler1 天前

Neat!

KiLee 的头像
KiLee1 天前

Open source ?

Flat 18 - Web Design & Development 的头像
Flat 18 - Web Design & Development1 天前

That is a remarkably tight render loop for a browser-based tracker. The interesting product question is how you keep the interaction feeling stable when the input gets noisy.

neamtu 的头像
neamtu1 天前

No Dax Demon on it? Wtf is this demo

Rohan Mukherjee 的头像
Rohan Mukherjee1 天前

pretty cool!

Sindre Reino Trosterud 的头像
Sindre Reino Trosterud1 天前

Now make it track a mosquito

Sm17 的头像
Sm171 天前

Man taking credit for clankers work!

Martin Solev 的头像
Martin Solev1 天前

But what's the use case? Very impressive btw!

Theo Ephraim 🧙‍♂️ 的头像
Theo Ephraim 🧙‍♂️1 天前

I remember when doing any kind of decent CV in the browser was absolutely black magic. I even wrote a paper about realtime face detection using flash in 2009 😂

Van Kram 的头像
Van Kram1 天前

Try doing it without the robo-QR codes

Taggart Maher 的头像
Taggart Maher1 天前

not fast enough man

MixRoute AI 的头像
MixRoute AI1 天前

What is it called?

Prince Kushwaha 的头像
Prince Kushwaha1 天前

Sub-10ms render times in the browser means you're hitting 100+ FPS locally—WebGPU and Wasm have made the browser a legitimate spatial computing engine.

相关视频