Quickstart

This page shows how to segment an image with a pretrained InstanSeg model. It assumes you have already installed InstanSeg.

Note

The examples use HE_example.tif, an H&E image from the instanseg/examples folder of the repository. Replace it with the path to your own image.

Segment an image

Create an InstanSeg object with the name of a pretrained model, then call eval() on the path to an image:

from instanseg import InstanSeg

instanseg_brightfield = InstanSeg("brightfield_nuclei", image_reader="tiffslide", verbosity=1)

labeled_output = instanseg_brightfield.eval(
    image="HE_example.tif",
    save_output=True,
    save_overlay=True,
)

The model is downloaded the first time it is used. eval returns the labels as a torch.Tensor, where each object has its own integer ID and the background is 0. With save_output and save_overlay set, the labels and an overlay of the labels on the image are also saved next to the input image.

Work with arrays

For more control over each step, read the image yourself and pass the array to eval_small_image():

from instanseg.utils.utils import show_images

image_array, pixel_size = instanseg_brightfield.read_image("HE_example.tif")

labeled_output, image_tensor = instanseg_brightfield.eval_small_image(image_array, pixel_size)

display = instanseg_brightfield.display(image_tensor, labeled_output)

show_images(image_tensor, display, colorbar=False, titles=["Normalized image", "Image with segmentation"])

eval_small_image returns the labels and the normalised input image, which display() combines into an RGB overlay.

Next steps

• Running inference covers the pretrained models, pixel sizes, large images and the command line.

• Training explains how to train InstanSeg on your own data.

• The API reference documents every public class and function.