Visual AI in agriculture faces messy realities: near-duplicate drone frames, severe class imbalance, inconsistent labels, and domain shift across fields and seasons. 

 

This talk demonstrates a modern data-centric workflow using FiftyOne, the open-source platform for visual dataset curation, with an integrated AI agent handling the heavy lifting. Through a live demo on real agricultural imagery, we'll cover dataset exploration, embedding-based analysis, annotation quality auditing, and curated subset creation, all driven by natural language. 

 

Attendees will leave with a concrete picture of how agentic tooling is reshaping day-to-day computer vision work, and why data curation, not model complexity, remains the highest-leverage investment in applied AI.