Walk the aisle. Know what's on the plants.
DeepLeaf turns a phone video of a tomato row into fruit counts by ripeness, plants, and sizes, then forecasts the harvest for this week and the next two.
A hosted service: you film and sign in from a browser, and DeepLeaf runs the models, keeps the data, and fetches the weather.
- Guided capture while filming
- Eight-week forecast
- Accuracy per greenhouse
From a walk to a forecast in four steps
You do the filming and the weighing. DeepLeaf does the counting, the tracking, and the forecast.
-
Film with the live guide
Walk one side of the aisle at the height of the lowest ripe trusses. About once a second the app says move closer, step back, hold steady, or good. Filming guide
-
Upload one clip or several
Save the walk to a crop cycle and a week. A long aisle can go up as up to 12 clips in order, scored and counted as one walk.
-
Check the capture
Every walk opens with a capture check that says if the camera was too far, too close, or the frames were blurry, so you know how much to trust it.
-
Log the harvest
Harvest kilograms turn fruit counts into kilograms, and each week's harvest checks the forecast issued the week before.
One place for the walk, the records, and the forecast
Every number traces back to its source: a walk, a harvest you logged, or the weather for your greenhouse.
-
Scores the walk
Fruit counted by ripeness (green, turning, red), one plant per stem, and sizes in millimetres when a calibration board is in the frame.
-
Guides the capture
Film in the app with a framing guide and a live verdict on distance and focus. The check never counts as a walk.
-
Forecasts eight weeks
Kilograms for this week and the seven after it, with a range from this cycle's own past errors, and your harvest history when a walk is missing.
-
Reports its own accuracy
For each greenhouse and horizon: error, bias, range coverage, and a plain trust level, from Not enough weeks yet to Reliable.
-
Keeps the records
Farms, greenhouses, rows, and crop cycles, with harvests, irrigation, climate files, and weather. Edit and delete with a confirmation, and export to CSV.
-
Tracks water productivity
Harvest kilograms per cubic metre of irrigation, week by week, from the harvests and irrigation you log.
-
Access keys with roles
Admins create, rename, and revoke their organization's keys. Member keys do everything except keys and payment.
-
Pay by card or bank wire
Monthly or yearly by card payment, bank transfer, or a bank wire invoice. Bank payments are matched to the invoice automatically.
-
Keeps each organization apart
Every key sees only its own organization's data. The pages load no outside scripts, fonts, analytics, or trackers.
What we measured, and where it stops
These results come from public greenhouse tomato datasets: other growers' houses, cameras, and labelling rules. They are not a validation on your rows, so treat them as a starting point. Once you log harvests, the accuracy report measures the forecast on your own greenhouses.
| What | Result | Where it was measured |
|---|---|---|
| Ripeness agreement with labels | 92.9% to 96.6% | Two public image sets, 1,104 and 4,850 labelled fruit |
| Fruit found, large enough for ripeness (≥ 29 px) | 79% | 449 frames from a robot driving between tomato rows, normal aisle distance |
| Harvest-history forecast error | 32% to 48% | A six-compartment cherry tomato trial, this week to two weeks ahead |
| Close-range 4K phone frames | about 1.5× count | Phone frames filmed very close: one large tomato can get several boxes |
Limits
Close-range footage overcounts, so follow the filming guide. The capture check warns when the largest fruit (90th percentile) are over 60 px. The ripening forecast itself has not yet been validated on a full real season. Accuracy and limits
