Usage

Installation

To install ELD:

(.venv) $ pip install .

Then you have to install torch==1.9.0, torchvision==0.10.0, and torchaudio==0.9.0 with the right CUDA version from PyTorch Previous Versions. Here’s an example:

(.venv) $ pip install torch==1.9.0+cu111 torchvision==0.10.0+cu111 torchaudio==0.9.0 -f https://download.pytorch.org/whl/torch_stable.html

Training

For training Visdom is used to easily following the training process. To start a Visdom server, do the following

(.venv) $ python -m visdom.server -port 9006

You can train models using different types of data: unimodal, 3D, and multimodal. The following sections describe how to use each type.

Unimodal

To train a model using unimodal data, use the --model unimodal flag. Here is an example command:

eld-train --elastic_sigma 5 --cuda 1 --port 9006 --data_path ../marcoAnalysis/MOB_HE/ --npts 16 --o scratch --elastic True --step_size 5 --ws 0 --gamma 0.9 --angle 8 --model unimodal

3D Data

To train a model on 3D tissue landmarks, use the --model 3d flag. The image files for this model should be named 0.png, 1.png,...,n.png where the number indicates the image’s position in the stack. Here is an example command:

eld-train --elastic_sigma 5 --cuda 1 --port 9006 --data_path ../marcoAnalysis/CODA_prostate/ --npts 16 --o scratch --elastic True --step_size 5 --ws 0 --gamma 0.9 --angle 8 --model 3d

Multimodal Data

To train a model using multimodal data, use the --model multimodal flag. The corresponding files should end with *_mod1.png or *_mod2.png, which indicate their respective modality. Here is an example command:

eld-train --elastic_sigma 5 --cuda 1 --port 9006 --data_path ../marcoAnalysis/CODA_prostate/ --npts 16 --o scratch --elastic True --step_size 5 --ws 0 --gamma 0.9 --angle 8 --model multimodal