# HPA Cell segmentation model

This is the official baseline cell segmentation model provided to the participants of the [Human Protein Atlas - Single Cell Classification](https://www.kaggle.com/c/hpa-single-cell-image-classification/) competition held in 2021.

The model is a DPN-Unet adopted from the top winning model of [2018 Data Science Bowl](https://www.kaggle.com/c/data-science-bowl-2018) competition for nuclei segmentation. It's a U-Net variation predicting a two-channel border mask. See [here](https://www.kaggle.com/c/data-science-bowl-2018/discussion/54741) for more details.

In this implementation, the same DPN-Unet model was trained on around 300 cell images from the Human Protein Atlas with slightly different post processing. The full training dataset along with the trained weights were published at https://zenodo.org/record/4430893. To run this baseline model with the orignal pytorch weigths, see [hpa cell segmentator](https://github.com/CellProfiling/HPA-Cell-Segmentation).