Find your objects
Work with supported raster imagery. Use classification and object extraction to identify candidates for review.
Built for agronomic research · Guided pilots
Your field knowledge.
A model that can learn it.
Turn trial imagery into reviewed training chips and custom trait-detection models. Dakota Agros connects the work between collecting images and teaching a model what matters.
For breeding teams, field-trial researchers, and agricultural imaging specialists.
From field to training set01 — 04
Candidate
Review
Label
Export
Human expertise, in the loop
Find it. Review it. Teach it.
Workflow illustration · not a prediction or accuracy claim
Agronomic trial imagery → Reviewed labels → Training chips → Custom models
The bottleneck is between the images and the model.
You know the trait.
Make it teachable.
Field imagery contains the detail your research needs. Turning that detail into consistent examples for machine learning takes careful selection, labeling, and review. Agros brings those steps into a connected workflow so your team's agronomic judgment becomes usable training data.
A connected research workflow
From a field image
to a training dataset.
Start with one trial and a clearly defined visual target. Keep researchers involved where their judgment matters most.
Put expertise into labels
Inspect candidate shapes, correct boundaries, and remove false positives before they become training examples.
Build training chips
Export reviewed objects as image chips with YOLO labels and a dataset configuration for model training.
Train, evaluate, refine
Run a custom training workflow, inspect results on held-out imagery, and use the errors to guide the next labeling pass.
Start where the workflow is strongest
Built around research trials.
Open to the next application.
Agronomic research
Scope visual targets such as sunflower blooms, plant objects, and canopy features. Build datasets around a defined crop, sensor, growth stage, and trial.
Repeatable dataset preparation
Give drone and geospatial teams a structured path from imagery to reviewed chips that a machine-learning workflow can use.
Other visual detection tasks
Forestry, ecological surveys, and inspection imagery may fit the same review-and-train process. New applications require a feasibility review and validation.
A practical first engagement
Prove the workflow
on your own imagery.
Start with a scoped pilot before committing to a larger deployment. We agree on the data, deliverables, and cost before work begins.
One dataset. One clear target.
From $1,500 USD / project
A focused engagement to test whether Agros fits your research workflow.
- Imagery and target-trait feasibility review
- Labeling workflow setup and guided chip review
- A scoped training dataset export
- Baseline training and results review, where feasible
- Recommendations for the next iteration
Final quote depends on data volume, labeling effort, and compute. No accuracy guarantee or unlimited training.
Keep building after the pilot.
Custom scope
For repeat studies, new traits, and teams that need a longer-term workflow.
- Additional datasets, crops, or target classes
- Repeat training and evaluation cycles
- Deployment and storage planning
- Workflow adaptation and team support
Hosting, GPU usage, support, and any integration work are specified in your proposal. Self-service subscriptions are not available yet.
Before you begin
A few things worth knowing.
Can I sign up and start uploading data today?
Agros is available through guided pilot engagements. We first review your imagery and agree on a working environment. Public self-service accounts and online billing are not available yet.
What exactly is a training chip?
A chip is a small image cut from a larger source image. Paired with reviewed labels, it gives a model an example of the object or visual feature you want it to learn.
Which traits can Agros detect?
The current workflow focuses on visible objects and features in agronomic imagery, including sunflower bloom and plant examples. Each new trait, crop, sensor, and season needs suitable labels and evaluation. Traits that are not visible in the imagery cannot be inferred reliably just by labeling chips.
How will we know whether it works?
We agree on evaluation criteria during scoping, such as labeling time, review effort, and detection quality on held-out imagery. Model results depend on imagery quality, label consistency, and how well the training data represents your trial conditions.
Where will my research data go?
Data transfer, access, storage, retention, and deliverable rights are agreed in your pilot proposal before imagery is shared. Please use the inquiry form for project details only; do not send research files or confidential data through it.
Let's start with your trial
What do you want
your model to see?
Tell us the crop or application, imagery format, target trait, and approximate dataset size. We'll discuss fit and put a clear scope around the first pilot.
Prefer email? [email protected]