Browser Based AI Prediction

This experiment integrates Vue.js with Tensorflow js and object detection with COCO-SSD


With Tensorflow JS, Deep Learning models can be instantiated in the browser, allowing you to analyze data streams captured in real-time by your own devices, or process a common data stream with individual user preferences.

The app above processes a video stream from your device's camera, analyzes the stream using an Object Detection prediction model running on the web page, and labels detected objects based on your preference - in this case, whether to pick out Everything, Living Things, or Non-living Things.

Use your device's rear camera to look for cats and dogs, or vehicles with the app. The app annotates vehicles recognized by the prediction model when you filter for Non-living Things. If there are no Living Things like dogs and cats in the scene, no annotations are drawn when filtering for Living Things.

Key takeaways:

  • Distributes the work of CNN prediction to the user device, and
  • Allows for a more personalized user experience, and
  • Is especially handy in use cases where different users wish to apply different preferences to identify different data points, even when analyzing data from a common input stream.

Need help with Machine Learning? Call us now!

133, New Bridge Road #24-01
Chinatown Point S(049513) •
Tel: 6-552-6826 •
Copyright Bumble Bee Laboratories Pte. Ltd. July 2020