Dockerizing FCN

[dcunhas]

Requirements:

1. Must be able to create Docker container and trigger learning/prediction with one command from host machine (or other docker container).
2. Settings must be set in settings file (currently considering .ini file).
3. Docker container run command (i.e. mounts, etc.) must be triggered based on paths and settings in file
4. Training/prediction must be correctly triggered based on settings file, depositing output into specified mounted directory.
5. Docker container must stop once training/prediction is complete.

Tasks

• Create sample .ini file with relevant and necessary values
• Create wrapper for FCN to run it based on settings in settings file
• Create wrapper for Docker run command to create instance based on settings

[dcunhas]

Sample settings.ini file.

[Docker Settings]
data_dir: 
; model_dir equivalent to logs_dir in FCN.py
model_dir: 
tensorboard_port:8050

[FCN Settings]
batch_size: 2
learning_rate: 1e-4
debug: False
image_augmentation: True
dropout: 0.5
; mode: train | visualize | predict | (test to be implemented)
mode: train 
max_iterations: 1e5 + 1
num_classes: 2
image_size: (224, 224)