How we used Face Detection to create a delightful User Experience
summary
Mixtiles replaces a crude center crop with a smart crop that unions face bounding boxes plus padding so people end up centered instead of chopped off. Frontend inference runs with face-api.js built on TensorFlow, choosing the tiny face detector (200 KB) over a 5 MB SSD model for speed. Model load costs ~1.5 seconds, detections cost milliseconds, and the feature cuts complaint volume dramatically and drives roughly $6,000,000/year; about 10% of images lack faces and need fallback logic.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:25Bad image crops problem
- 3:09Smart cropping with ML
- 4:55Face API JS intro
- 6:19Face detection in React
- 9:06Smart crop algorithm
- 10:32Results and impact
- 11:34Challenges and future
- 12:36Summary
related talks
31mFace Recognition
49mReal Time Face Detection through a Webcam Using AI
32mSelf-Portraits in JavaScript
31mFace detection in PHP
51mRealTime Exoskeleton-Mapping and Body Segmentation on the Browser with Tensorflow.js
57mFoundations of Facial Recognition
33mDo I Know You? Real time facial recognition with an Apache stack
15mMachine learning for next gen web apps with TensorFlow.js