Python based (no gui) way of estimating diameters

[mat10d]

I would like to replicate the diameter calibration outside of the GUI, for the pretrained models provided. I am running:

def estimate_cell_diameters(data, channels=[2,3], gpu=False):
    """Estimate diameter directly using Cellpose model's size estimator"""
    model = Cellpose(model_type="cyto3", gpu=gpu)
    diams, _ = model.sz.eval(data, channels=channels)
    return np.maximum(5.0, diams)

This is based on the calibrate size function in the GUI, found here. Is this a reliable way of getting optimal diameters in a purely python way?

Furthermore, every time I run this function, I get different calibrated diameters. Is this expected?