Explore the crucial role of data in developing tiny computer vision models in this tinyML Summit 2022 talk. Discover techniques for curating optimal training datasets, designing effective data sampling strategies, and measuring model robustness in diverse real-world environments. Learn about Plumerai's specialized data infrastructure and tooling, built to produce efficient tiny computer vision models. Gain insights into the challenges of creating production-worthy models that require large quantities of high-quality training data while avoiding shortcuts that don't generalize in practice.
Tiny Models with Big Appetites: Cultivating the Perfect Data Diet for Computer Vision
Overview
Syllabus
Introduction
plumeri
how did we build it
why not just use public data
design a good data set
plumeria infrastructure
Data unit tests
Data curation
Visual similarity
Recap
Data tooling
Vertical integration
Questions
Sponsors
Taught by
tinyML
