#Prediction Enhancer for Mitochondrion Segmentation in EM

This model was trained to segment mitochondria in EM.
It predicts foreground and boundary probabilities.

## Training

The network was trained on data from [the VNC dataset](http://dx.doi.org/10.6084/m9.figshare.856713)
and trained using [torch_em](https://github.com/constantinpape/torch-em).

### Training Data

- Imaging modality: Electron Microscopy
- Dimensionality: 2D
- Source: http://dx.doi.org/10.6084/m9.figshare.856713

### Recommended Validation

It is recommended to validate the instance segmentation obtained from this model using intersection-over-union.
This model can be used in ilastik, deepimageJ or other software that supports the bioimage.io model format.

### Training Schedule

- n_epochs: 5
- batches_per_epoch: 500
- batch_size: 1
- loss_function: DiceLoss
- optimizer: Adam
- learning_rate: 0.0001
- n_train_images: None
- n_validation_images: None

## Contact

For questions or issues with this models, please reach out by:
- opening a topic with tags bioimageio and mitchondriaemsegmentation2d on [image.sc](https://forum.image.sc/)
- or creating an issue in https://github.com/constantinpape/torch-em