Tutorial: Overview of the Command Line Interface (CLI)

The XAI library comes with its own CLI tool that allows you to execute XAI algorithms on your desired models and images directly from the command shell. Note that for the v0.1.0 release, Grad-CAM is the only XAI algorithm supported in the XAI library and hence the CLI tool only executes Grad-CAM.

This tutorial serves as an overview of the CLI tool and its arguments. For a step-by-step guide on using the CLI tool for the Food-Non-Food (FNF) and Food Scoring (FS) models, refer to the following:

Explanation of Arguments

Invoke the XAI tool on the command line through the XAI executable. Online --help is available on the command line. Enter the following to list commands or options for a given command with a short description.

$ XAI --help

usage: XAI [-h] [--batch_size BATCH_SIZE] [--relu_attributions]
       [--return_metadata] [--plot_gradcam]
       {FNF,FS} path_to_csv pretrained_weights_folder output_folder
       target_layer_name

Performs Grad-CAM for FoodDX.

positional arguments:
{FNF,FS}              The use case for Grad-CAM. Either 'FNF' (Food-Non-
                      Food) or 'FS' (Food Scoring).
path_to_csv           The absolute or relative path to the 'data.csv' file.
pretrained_weights_folder
                      The absolute or relative path to the folder that
                      contains the checkpoint (.pth) file(s). These .pth
                      files should contain the pretrained weights to be
                      loaded into the model. For Food-Non-Food (FNF), there
                      should be only 1 .pth file in the folder. For Food
                      Scoring (FS), there should be 3 .pth files in the
                      folder, one for each model.
output_folder         The absolute or relative path to save the Grad-CAM
                      outputs to. E.g. '/save/outputs/to/this/folder'.
target_layer_name     The name of the layer to use when performing Grad-CAM.

optional arguments:
-h, --help            show this help message and exit
--batch_size BATCH_SIZE
                      The batch size to split the dataset into. Default=10.
--relu_attributions   Indicates whether to apply a ReLU operation on the
                      final attribution, returning only non-negative
                      attributions. Setting this flag to True matches the
                      original GradCAM algorithm, otherwise, by default,
                      both positive and negative attributions are returned.
--return_metadata     If True, metadata corresponding to the input image
                      along with the Grad-CAM attribution maps will be
                      generated. If False, NaN values will be generated in
                      place of each metadata instead. Default=False.
--plot_gradcam        If this optional argument is called, the resulting
                      Grad-CAM arrays will be plotted and saved as .png
                      files in the folder specified in `save_output_path`.
                      By default, this optional argument is not called, i.e.
                      set to False

Boolean Arguments

The following are optional boolean arguments:

  • --relu_attributions

  • --return_metadata

  • --plot_gradcam

Boolean arguments are set to False by default and have two forms:

  • --this-option sets the argument to True

  • Not including the argument in the command sets it to False

General CLI Tool Workflow

The general workflow of the CLI tool is as such:

  1. Specify whether you are performing Grad-CAM on the FNF or FS model using the {FNF,FS} argument.

  2. Loads the model together with the pretrained weights.

    • In v0.1.0, the model is loaded from a class that contains the FNF and FS models. i.e., this is the only model that is available for use if you use the CLI tool.

    • The pretrained weights are loaded from the given folder in pretrained_weights_folder. The CLI tool will check that only 1 checkpoint file is given if you are performing Grad-CAM for FNF and only 3 checkpoint files are given if you are performing Grad-CAM for FS.

  3. Loads the .csv file in path_to_csv and creates a custom dataset using the images listed in the .csv file.

  4. Performs the Grad-CAM on the model and images and saves the output in the output_folder.

For specifics on how to use the CLI tool for FNF and FS models, refer to the following: