Spring Annuals Trial HarnessHabitat NZ · technical demo by EmerTech

CoFly-WeedDB — real UAV frame 7 (4.9 m AGL)

cofly-07 · 1280 × 720 px · 5.74 mm/px · 4.9 m AGLCC BY 4.0CoFly-WeedDB (Krestenitis et al.), DJI Phantom 4 Pro over cotton field, Larissa, Greece — zenodo.org/records/6697343

0 existing marks← All frames
cursor off framezoom 1.0×1280 × 720 pxclick to start a Pygmy buttercup mark

Nothing selected. Click near a mark to inspect it — every record carries its observer, timestamp and the method it was drawn with, because a count is only as defensible as its provenance.

Your marks

0 marks on this frame — 0 plants, 0 hard negatives.

Pygmy buttercupCeratocephala pungens0
Dwarf forget-me-notMyosotis brevis0
NZ mousetailMyosurus minimus subsp. novae-zelandiae0
Assumed Myosotis hikuwaiSpecies D — to confirmto confirm with Habitat NZ0
Hard negativesMust never be counted as a plant0

Agreement vs known truth

This frame ships with no ground truth.

Nothing to compare against: on a real sensor frame, truth only exists once a human draws it. Your marks here are the truth every model on this frame gets scored against.

Keyboard

Everything is one keystroke away — hands stay on the frame.

1 – 4
pick species
X
hard-negative mode
M
swap diameter / corners
Esc
cancel the first click, or deselect
Del
delete the selected mark
F
fit the frame
Space + drag
pan — or middle-drag, or just drag
Wheel
zoom about the cursor

Feeds the bake-off

The workbench is not a side quest — it is the input to the exemplar lane.

These boxes are exactly what the exemplar model lane consumes — annotate 3, then run the bake-off. Hard negatives go in as negative exemplars, which is how a detector is told that a schist chip is not a plant.