Training

Improve cell outlines with your reviewed examples.

Start a new model

Start from fresh uploads or reuse a previous completed analysis job so you can go straight into correction and training.

What do you want to improve?

Choose the workflow that matches the result you need.

Training learns which pixels belong to each cell. It improves segmentation; it does not teach a new biological measurement.

Cell-outline training does not validate tube-network measurements. Use tube formation analysis to measure length, branches and junctions.

Open tube formation analysis

Use images with known viability to calibrate the live/dead threshold.

Open viability calibration

Choose the matching analysis before training. Improving cell outlines alone does not validate another measurement.

Choose an analysis

Model name

Pick a name now. This is what your model will be saved as. You can edit it before training.

Culture type

Picks the best segmentation approach for your images.

Tip: for best training results, upload images without burned-in scale bars. Models that see scale bars at train-time can learn them as a feature, which then hurts inference on bar-free images. Upload guide.