retina-vessel-segmentation-competetion

Repository dedicated to training retinal vessel segmentation models

Container operations

inside docker/ folder

Build docker container

0. container configuration with port mappings and container naming can be found in container_source file
1. ./build.sh

Run container

0. container configuration with port mappings and container naming can be found in container_source file
1. ./container_up.sh

Attach to a running container's shell

1. ./bash_exec.sh

View running container logs

1. ./container_logs.sh

Run jupyter notebook inside a running container

1. ./jupyter_exec.sh

Shutdown and remove running container

1. ./container_rm.sh

Train model

execute everything in running docker container

1. configure experiment in config.py
2. ./train.sh 0 first argument has to be a number pointing to an existing gpu_id in PCI_BUS_ID order

Experiment configuration, logs, epoch metrics, tensorboard data and checkpoints will be stored in experiment folder in experiments/

Evaluate model

1. ./eval.sh <path to experiment configuration file .json> <device_id>

Evaluation result will be placed in selected model's experiment folder. Submission .zip file will be created too.