imageatm 0.0.1

Image classification for everyone

Image ATM (Automated Tagging Machine)

The Image ATM system is a one-click tool that automates the workflow of a typical image classification pipeline, this includes:

• preprocessing and validating input images and labels
• starting/terminating cloud instance with GPU support
• training
• model evaluation

For more details see the Image ATM confluence page.

Set up conda environment

conda env create -f environment.yml
conda activate image-atm

Testing

Run pytest -vs src/tests

Building

Run python setup.py install

Command line client

To see all available options and commands run

image-atm --help

Each component can be run individually or a pipeline can be run, which then checks in the config file which components to run.

TODOs:

• we are currently using Keras 2.2. The plan is to use tf.keras once TF 2.0 is out. Currently tf.keras is buggy, especially with model saving/loading (https://github.com/tensorflow/tensorflow/issues/22697)