RoBee FrameVision

AI draft first. Beekeeper judgment last.

The goal is not to fake certainty. The goal is to give RoBee a clean first pass, show confidence visually, and capture high-value corrections that teach the system faster.

Current capability

FrameVision estimates the visible comb region, drafts a cell lattice, and lets you confirm mite and queen signals before saving reviewed training data.

Detected now

Upload a frame to estimate visible comb coverage, cell size, and approximate cell count before classifying eggs, brood, pollen, nectar, or honey.

Next passes

Egg density becomes credible when every visible cell is mapped.
Brood irregularity becomes measurable as percentages by cell type.
Risk cues can flag queen cells, stress pockets, and anomalies against the lattice.
Live review surface

Frame inspection viewport

Tap mites in mite mode. Drag a box in queen mode. Save only after the frame looks right.
Browse source folders
Operator workflow

Training loop

This should feel like a simulator run: load a real frame, inspect in order, then commit the correction only when the teaching signal is clean.
Step 1
waiting
Load or select a frame
Use a real frame photo from the queue or upload one from the field.
Step 2
waiting
Inspect the AI draft
Check comb coverage first, then annotate mites or queen only in the active layer.
Step 3
waiting
Commit training labels
Save only after the frame is corrected well enough to become reusable training data.
Reference imagery

Source images inside the training view

Spot-check the actual source folders without leaving the review flow.
Loading source preview…