def _get_doc(ckpt, name, organelle):
training_summary = get_training_summary(ckpt, to_md=True, lr=1.0e-4)
model_tag = name.lower()
doc = f"""# U-Net for {organelle} segmentation in platynereis
This model segments {organelle} in electron microscopy images of Platynereis dumerilii.
## Training
The network was trained on data from [Whole-body integration of gene expression and single-cell morphology](https://doi.org/10.1016/j.cell.2021.07.017).
The training script can be found [here](https://github.com/constantinpape/torch-em/tree/main/experiments/platynereis).
This folder also includes example usages of this model.
### Training Data
- Imaging modality: electron microscopy
- Dimensionality: 3D
- Source: https://doi.org/10.1038/s41592-021-01249-6
### Recommended Validation
For validation, please refer to https://github.com/constantinpape/torch-em/tree/main/experiments/platynereis.
This model can also be used in ilastik, deepimageJ or other software that supports the bioimage.io model format.
### Training Schedule
{training_summary}
## Contact
For questions or issues with this models, please reach out by:
- opening a topic with tags bioimageio and {model_tag} on [image.sc](https://forum.image.sc/)
- or creating an issue in https://github.com/constantinpape/torch-em"""
return doc
def _get_doc(is_aff_model, ckpt, name):
if is_aff_model:
pred_type = "affinity maps"
pp = "The affinities can be processed with the Mutex Watershed to obtain an instance segmentation."
else:
pred_type = "boundary maps"
pp = (
"The boundaries can be combined with a nucleus segmentation to obtain an instance segmentation "
"via seeded watershed.")
training_summary = get_training_summary(ckpt, to_md=True, lr=1.0e-4)
model_tag = name.lower()
doc = f"""# U-Net for Cell Segmentation in IF
This model segments cells in immunofluorescence microscopy images.
It predicts {pred_type} and foreground probabilities and was trained on Vero E6 cells
imaged with a high-throughput-microscope, as part of a Covid19 antibody test.
{pp}
## Training
The network was trained on data from [this publication](https://doi.org/10.1002/bies.202000257).
The training script can be found [here](https://github.com/constantinpape/torch-em/tree/main/experiments/covid-if).
This folder also includes example usages of this model.
### Training Data
- Imaging modality: immunofluorescence images of Vero E6 cells
- Dimensionality: 2D
- Source: https://zenodo.org/record/5092850
### Recommended Validation
It is recommended to validate the instance segmentation obtained from this model using intersection-over-union.
See [the validation script](https://github.com/constantinpape/torch-em/tree/main/experiments/covid-if/validate_model.py).
This model can also be used in ilastik, deepimageJ or other software that supports the bioimage.io model format.
### Training Schedule
{training_summary}
## Contact
For questions or issues with this models, please reach out by:
- opening a topic with tags bioimageio and {model_tag} on [image.sc](https://forum.image.sc/)
- or creating an issue in https://github.com/constantinpape/torch-em"""
return doc
def _get_doc(is_aff_model, ckpt, name):
if is_aff_model:
pred_type = "affinity maps"
pp = "The affinities can be processed with the Mutex Watershed to obtain an instance segmentation."
else:
pred_type = "boundary maps"
pp = "The boundaries can be processed e.g. with Multicut or Watershed to obtain an instance segmentation."
training_summary = get_training_summary(ckpt, to_md=True, lr=1.0e-4)
model_tag = name.lower()
doc = f"""# U-Net for Nucleus Segmentation
This model segments nuclei in fluorescence microscopy images.
It predicts {pred_type} and foreground probabilities for nucleus segmentation in
different light microscopy modalities, mainly with DAPI staining.
{pp}
## Training
The network was trained on data from the Data Science Bowl Nucleus Segmentation Challenge.
The training script can be found [here](https://github.com/constantinpape/torch-em/tree/main/experiments/dsb).
This folder also includes example usages of this model.
### Training Data
- Imaging modality: different fluorescence light microscopy modalities.
- Dimensionality: 2D
- Source: DSB Kaggle Challenge
### Recommended Validation
It is recommended to validate the instance segmentation obtained from this model using intersection-over-union.
See [the validation script](https://github.com/constantinpape/torch-em/tree/main/experiments/dsb/validate_model.py).
This model can also be used in ilastik, deepimageJ or other software that supports the bioimage.io model format.
### Training Schedule
{training_summary}
## Contact
For questions or issues with this models, please reach out by:
- opening a topic with tags bioimageio and {model_tag} on [image.sc](https://forum.image.sc/)
- or creating an issue in https://github.com/constantinpape/torch-em"""
return doc
def _get_doc(is_aff_model, ckpt, name):
if is_aff_model:
pred_type = "affinity maps"
pp = "The affinities can be processed with the Mutex Watershed to obtain an instance segmentation."
else:
pred_type = "boundary maps"
pp = "The boundaries can be processed with Multicut segmentation to obtain an instance segmentation."
training_summary = get_training_summary(ckpt, to_md=True, lr=1.0e-4)
model_tag = name.lower()
doc = f"""# U-Net for EM Neuron Segmentation
This model segments neurons in electron microscopy images. It predicts {pred_type} segmenting neural membranes.
{pp}
## Training
The network was trained on data from the CREMI Neuron Segmentation Challenge.
The training script can be found [here](https://github.com/constantinpape/torch-em/tree/main/experiments/neuron_segmentation/cremi).
This folder also includes example usages of this model.
### Training Data
- Imaging modality: serial section transmission electron microscopy data of fruit-fly neural tissue.
- Dimensionality: 3D
- Source: CREMI Neuron Segmentation Challenge
### Recommended Validation
It is recommended to validate the instance segmentation obtained from this model using metrics derived from the RandIndex
or the Variation of Information.
See [the validation script](https://github.com/constantinpape/torch-em/tree/main/experiments/neuron_segmentation/cremi/validate_model.py).
This model can also be used in ilastik, deepimageJ or other software that supports the bioimage.io model format.
### Training Schedule
{training_summary}
## Contact
For questions or issues with this models, please reach out by:
- opening a topic with tags bioimageio and {model_tag} on [image.sc](https://forum.image.sc/)
- or creating an issue in https://github.com/constantinpape/torch-em"""
return doc
def _get_doc(is_aff_model, ckpt, name):
if is_aff_model:
pred_type = "affinity maps"
pp = "The affinities can be processed with the Mutex Watershed to obtain an instance segmentation."
else:
pred_type = "boundary maps"
pp = "The boundaries can be processed e.g. with Multicut or Watershed to obtain an instance segmentation."
training_summary = get_training_summary(ckpt, to_md=True, lr=1.0e-4)
model_tag = name.lower()
doc = f"""# U-Net for Livecell Segmentation
This model segments cells in phase-contrast microscopy images, which are often used in live-cell imaging.
It predicts {pred_type} and foreground probabilities. {pp}
## Training
The network was trained on data from the [LiveCell publication](https://doi.org/10.1038/s41592-021-01249-6).
The training script can be found [here](https://github.com/constantinpape/torch-em/tree/main/experiments/livecell).
This folder also includes example usages of this model.
### Training Data
- Imaging modality: phase-contrast microscopy
- Dimensionality: 2D
- Source: https://doi.org/10.1038/s41592-021-01249-6
### Recommended Validation
It is recommended to validate the instance segmentation obtained from this model using intersection-over-union.
See [the validation script](https://github.com/constantinpape/torch-em/tree/main/experiments/livecell/validate_model.py).
This model can also be used in ilastik, deepimageJ or other software that supports the bioimage.io model format.
### Training Schedule
{training_summary}
## Contact
For questions or issues with this models, please reach out by:
- opening a topic with tags bioimageio and {model_tag} on [image.sc](https://forum.image.sc/)
- or creating an issue in https://github.com/constantinpape/torch-em"""
return doc
def _get_doc(is_aff_model, ckpt, name):
if is_aff_model:
pred_type = "affinity maps"
pp = "The affinities can be processed with the Mutex Watershed to obtain an instance segmentation."
else:
pred_type = "boundary maps"
pp = "The boundaries can be processed with Multicut to obtain an instance segmentation."
training_summary = get_training_summary(ckpt, to_md=True, lr=1.0e-4)
model_tag = name.lower()
doc = f"""# U-Net for Mitochondria Segmentation
This model segments mitochondria in electron microscopy images. It predicts {pred_type} and foreground probabilities. {pp}
## Training
The network was trained on data from the [MitoEM Segmentation Challenge](https://mitoem.grand-challenge.org/).
The training script can be found [here](https://github.com/constantinpape/torch-em/tree/main/experiments/mito-em).
This folder also includes example usages of this model.
### Training Data
- Imaging modality: serial blockface electron microscopy
- Dimensionality: 3D
- Source: https://mitoem.grand-challenge.org/
### Recommended Validation
It is recommended to validate the instance segmentation obtained from this model using intersection-over-union.
See [the validation script](https://github.com/constantinpape/torch-em/tree/main/experiments/mito-em/validate_model.py).
This model can also be used in ilastik, deepimageJ or other software that supports the bioimage.io model format.
### Training Schedule
{training_summary}
## Contact
For questions or issues with this models, please reach out by:
- opening a topic with tags bioimageio and {model_tag} on [image.sc](https://forum.image.sc/)
- or creating an issue in https://github.com/constantinpape/torch-em"""
return doc