Building self-hosted computer vision software with tinygrad
Computer vision allows computers to "see", but how?
What can computer vision do for me?
And how can it be used on me by others?
This talk will show you how computer vision models work, what they can do for you, what they can do against you, what they cannot do, and how to not send a 24/7 broadcast of your home to an AI company.
This talk will be around Clearcam, an (my) open source NVR with computer vision functionality built in. While open source models are on the rise, so too is the cost of hardware. Tinygrad is a machine learning framework similar to PyTorch, but with the goal to "commoditize the petaflop" rather than simply running as fast as possible on a specific hardware architecture, with token support for others.
Clearcam aims to have the ease of use and functionality of a cloud connected doorbell camera, but without the monthly (with fluctuations) cost, cloud dependance, and privacy concerns. You do not need to be a rich, Nvidia GPU (Mac Minis are double in price now too!) owning, hobbyist linux administrator to reap the benefits of open source!
Software Engineer specialising in computer vision and ML inference