VeinGAN

Synthetic Image Generation of Veins using Generative Models

Use this script to install the requirements for this implementation.

pip install -r requirements.txt

To download the dataset locally, use this script as follows.

python -m scripts.download --dataset=<dataset> --target=<target>

Parameter

dataset: A Kaggle Dataset Name or Key
- kaggle-fv: Kaggle Finger Vein Dataset (Default)
target: Target Directory for Dataset (Default: ./veingan-tmp/dataset/)

To train the model and generate synthetic images, use the following command and adjust the given parameters.

python -m scripts.generate <model> <dataset> <target> --verbose --epoch=<epoch> --configuration=<configuration>

Parameter

model: The GAN model to generate the image-
- gan: Deep Convolutional GAN
- vae: Variational Autoencoders
- cyclegan: CycleGAN
dataset: The dataset to train on
target: The target directory to save the images
verbose: Argument to enable verbose output of model progress
epoch: The amount of cycle to train on
configuration: Configuration Specifc for Model
- vae_128+cpu: VAE CPU Mode (Default)
- vae_128+gpu: VAE GPU Mode
- gan128_64+cpu: GAN CPU Mode (Default)
- gan128_64+gpu: GAN GPU Mode
- gan128_64+full: GAN GPU Mode with (Full Train)
- cyclegan128_1+cpu+train: CycleGAN CPU Mode (Default)
- cyclegan128_1+gpu+train: CycleGAN GPU Mode
- cyclegan128_1+full+train: CycleGAN GPU Mode (Full Train)
- cyclegan128_1+infer: CycleGAN Inference Mode

Evaluate some sample of the dataset using various methods.

python -m scripts.evaluate <method> <target> --configuration=<configuration>

Parameter

method Method of Evaluation
- entropy: Calculates Entropy of Sample Sets
- laplace: Calculates Laplace Gradient of Sample Sets
- inception: Calculates Inception Score of Sample Sets
- snapshot: Preview a snapshot of Sample Sets
target : Target Directory and Label of Evaluation Sets (Example: "dir:label;dir:label")
configuration : Specific Configuration for Evaluation Method

Provide feedback