GAN

This subdirectory contains the files relevant to the Tensorflow Model of the Neural Network. The scatgan1, scatgan3 and vol2scat subdirectories contain:

• The (untrained) Tensorflow model for the network
• The training and validation datasets
• The Testing dataset on which the quantitative evaluations are run (Report and Paper)

respectively for the scatgan1, scatgan3 networks, and the network which processes full 3D volumes (vol2scat).

Sample Commands to Use the PIX2PIX GAN

The following commands allow one to:

Concatenate {Input, Target}-image pairs

python tools/process.py \
  --input_dir <input_image_dir> \
  --b_dir <target_image_dir> \
  --operation combine \
  --output_dir <dataset_dir_name>

Split into train/validation set

python tools/split.py \
  --dir <dataset_dir_name>

Train Network

python pix2pix.py \
  --mode train \
  --output_dir <dataset_dir_name>_train \
  --max_epochs 1000 \
  --input_dir <dataset_dir_name>/train \
  --which_direction AtoB \
  --display_freq 50

Test/Predict

python pix2pix.py \
--mode test \
--output_dir <test_output_dir> \
--input_dir <dataset_dir_name>/val \
--checkpoint <dataset_dir_name>_train

Tensorboard

tensorboard --logdir=<dataset_dir_name>_train --port=6006