This Repository was created to test Deep ICY

Image from PanNuke Dataset       Nuclei segmentation result

The model used is a simple UNet with 3 channels as input and 3 channels as output. model was trained on PanNuke Dataset with the TI loss (https://github.com/TopoXLab/TopoInteraction):

The goal is to segment nuclei.

There are 3 classes as output:

0: Background pixels

1: Nuclei contours

2: Nuclei

Pre-processing:

just normalize the image between [0,1] by dividing by 255.

Post-processing:

Apply exponential to the output of the model.

Test

In test_deep_icy_model.ipynb you can try the model with and example image.