Train your own AI, with ease
Build a model custom to your images — starting with a single file, one labeled tile, and no code.
Start with a single image
Upload at least one image to begin — there's no minimum dataset required to get started. Once it's imported, choose a segmentation type and tile size for labeling, based on the size of the objects you're detecting.
Label one tile, and you can start training
Add your first class, then label just one tile containing a few objects — that's enough to kick off your first training run. AI-assisted labeling helps from the very first tile, and once your model is trained, use it to label the rest of your images automatically — correcting only where needed instead of labeling everything by hand.
Want a more robust model? Label more tiles and more objects before your first training run. Exactly how much you need varies by model — a simple, low-variability structure might need only a handful of examples, while a more complex or variable one benefits from a larger, more diverse labeled set.
Train, validate, and refine
Train your model directly in the platform, then validate its predictions against your labeled ground truth. If it's not accurate enough yet, label a few more tiles and retrain — each round improves the model without starting over.
Every version saved, switch active anytime
Every training round creates a new version — nothing is overwritten or lost. Keep every version on record, and choose which one is active for analysis while the rest stay deactivated. Compare performance across versions, and switch back to an earlier one anytime if a newer version doesn't perform the way you expect.
Your model detects the objects — you decide what gets measured
Training gets your model to find the right cells, tissue regions, or structures. What happens after detection — which metrics are computed, how object classes relate to each other, how regions are transformed — is configured separately.
Design your measurements →