Code to Run AuDrA Model to Assess Originality in the Incomplete Shape Drawing Task

This repository contains code from Patterson et al.'s (2023) paper AuDrA: An automated drawing assessment platform for evaluating creativity.

Github Repo Navigation

The following is the top-level directory layout of this repo:

.
├── \__pycache__
├── \user_images                                  # Folder to store drawings to evaluate originality
├── .gitignore                                    
├── AuDrA tutorial.pdf                            # A detailed tutorial on how to use pre-trained AuDrA model
├── AuDrA_Class.py                                # A Pytorch class of AuDrA model
├── AuDrA_DataModule.py                           # A class related to loading data for model training
├── audra_environment_cpu.yml                     # Configure environment for running on CPU
├── audra_environment_gpu.yml                     # Configure environment for running on GPU
├── audra_EXACT_ENVIRONMENT_USED_GPU_LINUX.yml    # Configure environment for running on CPU (exact environment of original training)
├── AuDrA_predictions.csv                         # Predicted originality rating (output after running `AuDrA_run.py `)
├── AuDrA_run.py                                  # Main script to assess originality in the incomplete shape drawing task
├── AuDrA_trained.ckpt (gitignored)               # Checkpoint file of pre-trained AuDrA model
├── datafuncs.py                                  # A helper function
├── invert.py                                     # A helper function
├── store_user_images.py                          # A script to store user images from `analysis/data/drawings/png` direcotry in the `user_images` directory
├── README.md

Environment Activation and Code Running

After installing miniconda, the first step is to create the virtual environment based on where you plan to run the code (CPU or GPU). For instance, for running on CPU: bash conda env create -f audra_environment_cpu.yml This will create a virtual environment called audra_cpu.

The next step is to activate the created environment (still suppose using CPU): bash conda activate audra_cpu

Afterwards, we need to copy drawings from the analysis/data/drawings/png directory into the user_images directory in this repository: bash python store_user_images.py

Finally, run AuDrA_run.py file (using the Python interpreter located at the virtual environment), which calculates originality ratings of drawings stored in \user_images and outputs the results into AuDrA_predictions.csv (default name of output csv file): bash python AuDrA_run.py

OSF Access

To acceess the the trained AuDrA model (with tutorial), analysis script, and ratings, go to this website.

References

@article{patterson2023audra, title={AuDrA: An automated drawing assessment platform for evaluating creativity}, author={Patterson, John D and Barbot, Baptiste and Lloyd-Cox, James and Beaty, Roger E}, journal={Behavior research methods}, pages={1--18}, year={2023}, publisher={Springer} }