LSIMasks

Proposal-free instance segmentation from Latent Single-Instance Masks

Installation (on linux)

If you plan to use the code to train your model, then you will need to install some extra packages:

• Clone the repository: git clone https://github.com/abailoni/LSIMasks.git
• cd LSIMasks
• chmod +x ./install_dependencies.sh
• To install the dependencies, you will need conda or miniconda
• Install the dependencies and the package by running ./install_dependencies.sh. While the script is running, you will need to confirm twice.
• The script will create a new conda environment called LSIMasks including all you need

Starting the training from scratch:

To start the training, run the following command:

CUDA_VISIBLE_DEVICES=0 ipython experiments/cremi/train_model.py -- <yourExperimentName> --DATA_HOMEDIR <path/to/the/cremi/data/you/downloaded> --inherit main_config.yml

(the one just given is a single command: for readability it was split into multiple lines)

Visualizing the training results in tensorboard

Go to the experiment folder (by default placed in the experiments/cremi/runs folder) and then start tensorboard:

tensorboard --logdir=./ --bind_all

For this, you will need to install tensorflow, with pip install --upgrade tensorflow