Sign In

Nikhilman thoman

Software Engineer at Google

Highlights

Mar 24 · twitter

🖐️ micro-handpose: browser hand tracking faster than MediaPipe, pure WebGPU compute shaders.

15 WGSL shaders run palm detection + landmark inference on GPU. 74KB JS (17KB gzip) + 7.7MB weights vs MediaPipe's 34MB package. Zero dependencies.

M4 Pro Chrome / iPhone Safari: • micro-handpose — 2.2ms / 72ms (WebGPU) • MediaPipe GPU — 4.0ms / 97ms (WASM+WebGL) • MediaPipe CPU — 4.5ms / 136ms (WASM)

npm install @svenflow/micro-handpose

const handpose = await createHandpose() const hands = await handpose.detect(video) console.log(hands[0].keypoints.thumb_tip)

🖐️ micro-handpose: browser hand tracking faster than MediaPipe, pure WebGPU compute shaders.

15 WG
Jul 11 · twitter

I believe this but only because of the pool of users they chose.

The ones who are massively sped up are not amazing ICs, they’re ones with manager / TL skills where you’re constantly context switching and talking to 10 people about 10 things all day every day.

Related Questions

What inspired Nikhil Thorat to co-create TensorFlow.js?
How did Nikhil Thorat transition from Google to Databricks?
What challenges did Nikhil Thorat face while starting Lilac AI?
Can you tell me more about Nikhil Thorat's role at Google?
What is Know Your Data, and how did Nikhil Thorat contribute to it?
Nikhilman thoman
Add to my network

Experience

Software at Dupont Pioneer
Software Engineer at Campuslive Inc

Education

University Of Massachusetts
Mass Academy
Worcester Polytechnic Institute
Massachusetts Academy Of Math And Science At Wpi
Bachelor Of Science In Mathematics, Computer Science, University Of Massachusetts Amherst

Location

Northborough, Massachusetts, United States