TensorFlow.js: From prototype to production, what's new in 2022?
summary
TensorFlow.js adds concrete production-ready features: Coral Edge TPU support via tfjs-tflite-node for Node.js, a new selfie depth model for per-pixel distance, and a 3D pose model outputting 33 keypoints with light/full/heavy variants (light hits ~150 FPS on modern desktops). Body segmentation runs >120 FPS, selfie segmentation >200 FPS, WebGL init lag is fixed, and 3.13.0 introduces tensor.dataToGPU; the Web GPU alpha backend halves body-pic latency from 44.6ms to 21.7ms.
AI-written summary · about
chapters (8)
- 0:00Start
- 1:59Coral Edge TPU support in tfjs-tfite-node release
- 7:22Selfie Depth Estimation
- 8:45Full Body Segmentation
- 9:53Hand Pose Estimation
- 10:22Face Landmark Detection
- 12:42WebGL Parallel Compilation
- 13:26Keeping data on GPU
related talks
20mTensorFlow.js Bringing Machine Learning to the Web and Beyond
41mTensorFlow.js 101: ML in the Browser and Beyond
15mMachine learning for next gen web apps with TensorFlow.js
24mTensorFlow.js + Node
37mTensorFlow.js - Bringing ML and Linear Algebra to Node.js
48mTensorFlow.js
41mTensorFlow ever higher
31mDeep Learning at the Edge with TensorFlow.js