I can see clearly now - Adding vision to a rubber duck - DataConf 2024

Talk given at DataConf (Cape Town) - 29 August 2024

Notes

Slides are available here

Original Abstract

Miss Duckworth is a large 3D printed duck that is capable of having discussions with you and runs all of the AI models locally without internet access. Using machine learning, various image recognition libraries and multimodal AI models I want to add basic vision capabilities to the duck. Be able to ask it questions about what it ‘sees’ and have the duck answer back. Have the duck recognise people and try have conversations with them.

This talk covers how I got this to work, if it works, and all the tools and libraries required to do image processing offline using low powered computers without GPU’s etc.

See also