This is the pytorch implementation of the paper:

Uwe Schmidt, Martin Weigert, Coleman Broaddus, and Gene Myers.
Cell Detection with Star-convex Polygons.
International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI), Granada, Spain, September 2018. Refer the StarDist repo before you continue.

All the necessary packages to generate ground truth can be found in the StarDist official repository. The code uses the deep learning library: PyTorch, instead of Keras/Tensorflow.
Optional: TensorboardX

• main.py: Start file for training the network. Specify path to dataset, tensorboard log directory etc here.
• dataloader.py: Simple basic pytorch dataloader for DSB2018.
• distance_loss.py : Loss function class, as defined in the paper.
• train.py: Trainer file. Code to load and save checkpoints in accordance with loss and learning rate.
• load_save_model.py: Boilerplate assisting code for the Trainer class.Used for loading and saving models.
• predict.py: Script for test set prediction. Specify path to test set and pretrained weights here. User can also change the probability threshold for NMS.
• metric.py: Unofficial evaluation metric script for calculating average precision of IoUs.

The performance of the model was evaluated on DSB2018 data set. The table shows average precision for several IoU thresholds when probability threshold was 0.4, calculated by metric.py script.

UNet code adapted from: https://github.com/milesial/Pytorch-UNet