Dataset Builder
Generate labeled datasets by searching nearest neighbors in HSV and/or MobileNet embeddings.
Anchor filename
Exact filename of the anchor image to search around.
Site ID (optional)
Optional. Leave blank to search across all sites.
Model selection
HSV
MobileNet
Both
Embedding space used for nearest-neighbor search.
Class label
baseline
milky
surcharge
fog
glare
occluded
debris
other
Label to assign to exported dataset rows.
Other label (if selected)
Only used when class label is 'other'.
Max results
Maximum number of rows returned after merge, sorting, and dedupe.
Min results
Minimum desired row count. Warning shown if fewer are found.
Radius (optional)
Optional max embedding distance (raw vector distance). Lower is stricter.
Per-model K (when BOTH)
When model=Both, fetch this many candidates per model before merge.
Time window days (optional)
Optional +/- day window around anchor timestamp.
Dedupe minutes
Keep best match per time bucket of this many minutes. 0 disables dedupe.
Random grid size (max 100)
Number of random sample images shown in grid preview.
Export destination
Download
Store in DB
Download CSV now or store dataset rows in DB first.
Re-run search
Save dataset build
Export CSV
Summary
Run a search to see summary info.