I noted that we would need Labellerr to fine‑tune detection models for snow and other obstacles.
I emphasized on 2024‑05‑27 that high‑quality labeled data must be context‑aware for the target use‑case, using autonomous driving as an example where a stop‑sign label is only useful if the sign lies on the vehicle’s path. I decided to prioritize “context and relevance of the annotations” over generic labeling to improve model accuracy and real‑world usefulness.
I decided to prioritize collection of nuanced, domain‑specific labels for agricultural robotics because generic object tags (e.g., “plant present”) do not enable actionable decisions such as watering or harvesting; instead I need annotations of growth stage, disease/pest presence, and environmental context.