MultiStar: Instance Segmentation of Overlapping Objects with Star-Convex Polygons

Link to the paper, presented at ISBI 2021.

How to use this repository:

• Install the required packages as a conda environment with MultiStar.yml. Additionally, you need this U-Net implementation, which needs to be stored in a folder /confnets in the same location as the other scripts (e.g. dataset.py, evaluation.py, ...).
• In the paper we evaluated our algorithm on two datasets: DSBOV and OSC-ISBI. They can be downloaded / recreated and stored in the proper format and location by running the scripts in /datagen. OSC-ISBI is composed of two datasets (ISBI14 and ISBI15), but this is taken care of by using the DatasetPlus class in dataset.py instead of the Dataset class. DatasetPlus accesses data from both datases.
• Example usage for the DSBOV dataset with gpu available is demonstrated in usage_example.py
• Results are saved in the directory experiments.