Harbo workflow
People wanting to help fight Varroa can start by collecting assay images that preserve the evidence from each inspected cell.
Open collaboration
Help build the Open Varroa Family Image Dataset.
Help create the labelled Harbo assay dataset needed for future Varroa decision-support tools: pupa in cell, pupa removed, original cell wall and floor, mite-family evidence, no-mite cells, false positives and uncertain cases.
Why this dataset
This is for opened brood-cell family evidence, not just mite detection.
Harbo/VSH scoring depends on linked images from the actual assay: pupa in cell, pupa removed, original cell wall and floor, mite family evidence, fecal patches, false positives and uncertain cases.
VSH labels
The label language needs to distinguish reproductive, non-reproductive, non-viable, multi-foundress and uncertain cases.
False positives
Larval skin, wax flakes, pollen, glare, dark comb and poor focus are useful examples when clearly marked.
Data before tools
Careful, permissioned Harbo assay data comes first. Shared images and labels give future review tools something reliable to learn from.
How it connects
The dataset trains eyes. The assay workflow records decisions.
The Open Varroa Family Image Dataset is for shared label confidence: R, NR, NV, no-mite cells, false positives and uncertain cases. BuzzTech's Harbo workflow is where those labels become hive records that can be compared with queen source, yield and survival.
Training evidence
Images help operators and reviewers agree what the cell shows.
Field record
The assay result belongs to the tagged hive and season history.
Review value
Uncertain examples are useful when they are labelled honestly.
Selection value
VSH labels matter most when linked back to business performance.
Assay reference images
Collect clear images from each inspected brood cell.
Strong submissions document the full review sequence: the cell before removal, the removed pupa, the original cell wall and floor, and any mite-family evidence or uncertain detail.
Image set needed
The best contribution is a linked set from the same inspected cell.
Labelled images are valuable. Unlabelled images are valuable too, especially if the cell relationship and basic context are intact.
Pupa in cell
Cell before removal, with brood stage context where possible.
Pupa removed
Pupa visible outside the cell for foundress, offspring, damage and stage review.
Cell wall and floor
The original empty cell, including wall, floor, fecal patch and hidden mite evidence.
Close-up detail
Foundress, egg, male, nymph, daughter mite, mite feces, debris or uncertain area.
Useful labels
Start with a label language humans can actually use.
A single image can carry multiple labels. The GitHub README covers filenames, metadata fields, reviewer status and suggested confidence values.
no_mite
foundress_only
reproductive
non_reproductive
non_viable
multi_foundress
mite_fecal_patch
pupa_damage
false_positive
uncertain_review
unlabelled
AI assist, not AI replacement
The first model target is triage and overlay support.
The useful near-term system highlights likely foundress mites, offspring, mite fecal patches, bad images and false-positive risks so a trained human can decide faster and more consistently.
Decision stays human The dataset is intended to support teaching, review and assay speed. Final VSH and breeding interpretation stays with expert scorers.
Accepted submissions
Raw images, reviewed images, unlabelled images, partial labels, uncertain cases, false positives and repeated views under different lighting.
Most useful metadata
Cell ID, filename, label status, lighting type, device or microscope, contributor permission and any brood-stage or Harbo timing context.
Permission standard
Only contribute images you own or have permission to share. A permissive research-friendly licence is preferred.
Contribution order
Read the README, start with a small labelled batch, then review label and metadata quality before sending more.
Contribute data
Have images, labels or review time to contribute?
The repo is public on GitHub. Use the README for the technical details, then use this form if you have labelled images, unlabelled images, or expert review time to offer.
Start collecting Anyone wanting to do something practical against Varroa can help by collecting Harbo assay images with labels, cell IDs, permission status and enough context to review later.