Farm AI data, labeled by people who grow things.
Agricultural models face changing light, seasons and regions, and the cost of a wrong call is a lost crop. They need varied field data labeled by people who know what they are looking at.
Agriculture, with proof.
Agriculture teams build models for disease and pest detection, weed identification, yield estimation, livestock monitoring, advisory chat and field robotics. Field conditions are the hard part. A leaf spot looks different at dawn and noon, in one variety and another, early and late in the season. Many models trained on a narrow set of fields falter when deployed somewhere new. HUMXN sources field imagery, video and recordings from consenting growers and creators across the conditions you specify, and commissions specialists to label them. A pilot batch comes first.
Diagnosis is expert work. A nutrient deficiency, a fungal infection and herbicide drift can look alike in a photograph, and a generalist annotator will guess. Our agronomists, plant pathologists, entomologists and veterinarians label to your taxonomy, note growth stage and severity, and pass uncertain calls to a second expert. Labels record whether they came from a creator, an AI suggestion or expert review. Advisory models can be fine-tuned and evaluated on expert-written answers to the questions growers actually ask.
Farm imagery has its own privacy questions. Location metadata from phones and drones never reaches buyers, and property releases are bound to each item where relevant. Images are fingerprinted with crop-tolerant perceptual hashes and checked for duplicates, so the same field photographed twice does not skew your class balance. Field-robot recordings can be captured by trained operators with synchronized video and sensor data. Every item carries an Ed25519-signed provenance record and is listed on a signed receipt.
Built to your specification.
Every engagement starts from a written spec and a pilot batch. These are the most common requests we source for agriculture.
Crop disease and pest imagery
Field and close-up photographs labeled for disease, pest, deficiency, growth stage and severity by plant specialists.
Weed and crop segmentation
Pixel-level masks separating crop from weed species, for spot-spraying and robotic weeding models.
Livestock monitoring data
Video and images of animals labeled for behavior, body condition and visible health signs by veterinarians.
Field-robot demonstrations
Teleoperation and sensor recordings for harvesting and field tasks, captured by trained operators with outcome labels.
Agronomic advisory answers
Expert-written answers and graded model responses to grower questions on crops, soil, pests and timing.
Drone and aerial imagery
Aerial imagery labeled for stand counts, stress and damage, with location metadata stripped before delivery.
01Field variability
Light, season, variety and region shift what symptoms look like. Sourcing across specified conditions gives models variation to learn from.
02Lookalike symptoms
Deficiency, disease and chemical damage can look alike. Specialist labels and second-expert review keep diagnoses accurate.
03Farm location privacy
Location metadata from phones and drones is stripped, and property releases are bound to each item where relevant.
Agriculture: asked often.
Can you source imagery for specific crops and regions?
We source against a written specification of crops, varieties, growth stages, conditions and regions, from consenting growers and creators and through commissioned work. Where a crop or condition is hard to source, we say so during scoping and in the spec. The pilot batch shows the coverage you can expect before you scale.
Who labels agricultural data?
Credential-checked agronomists, plant pathologists, entomologists, crop scientists and veterinarians, matched to the task. Uncertain diagnoses go to a second expert. Each label records whether it came from a creator, an AI suggestion or expert review, so your team can filter by confidence when training.
Is farm location data shared with buyers?
No. Location and personal metadata never reach buyers, including GPS data from phones and drones. Property releases are bound to each item where relevant, and works showing identifiable people are offered for training only with consent on file. If you need region-level labels, we agree them in the spec.
What formats do you deliver?
Images with labels in COCO format or WebDataset shards, segmentation masks and metadata in JSON Lines or Parquet, and robot and sensor episodes in Parquet with video shards. Croissant 1.0 metadata can describe the dataset, and approved buyers can use the API.
How is agricultural data priced?
By sourcing difficulty, annotation type, the specialist required, review depth and licence. Dense segmentation and expert diagnosis cost more than image-level tags. Seasonal standing orders can spread sourcing across the year, and volume reduces the unit price. We quote after scoping.
Tell us what your model needs to learn.
Send a brief in five minutes. A data lead replies within one business day with questions and a first sourcing plan.