Contexts

AI Generated Images for Cognitive Tasks

Overview:

We systematically generated thousands of Context Images for our task’s backgrounds using the Dalle2 AI system, which creates realistic images from a description in natural language. This page details the steps we took to accomplish this.

Preparation / Excel Formation

• Create context categories.
• Create sub-categories and adjectives for each category.
• Select artistic styles to apply to each category.

• Create a numbering system for categories, sub-categories, adjectives, and art styles, giving each a unique identifier.

• Structure the data in excel to contain a row for each category/sub-category, creating individual combinations that are the basis for the Python loops. 
• Use excel formulas (V-Lookups) on the structured data to grab each categories corresponding ID and place it in a column next to the category name.  This allows the python loops to generate the contexts based on their names, and then save the files based on their corresponding IDs.

Python / Context Generation:

1. Convert your excel workbook into a CSV file.
2. Create a Python script:
   • Import Dalle2, CSV, and Requests.
   • Go to https://github.com/ezzcodeezzlife/dalle2-in-python and follow the setup directions, which will walk you through how to obtain your Bearer Token. Once obtained, add your bearer key to your Python script.
   • Add a string variable with the directory path of where to save the contexts.

Next, create a Python Method that:

• Reads in the converted CSV file,
• Loops through the string combinations, generating a Dalle2 image for each string, and saving the strings in the directory specified above.
• Note: We save each context with a file name matching the numbering system used in the excel file above. This allows for easy categorization and retrieval of specific contexts.