This project requires:

• Python 3.8.18 (other versions may work, but this is the tested version)

git clone https://github.com/trishamazu/complexity-experiment-final.git

pip install -r requirements.txt

• THINGS (Hebart et al., 2020)^[1]
• Bistable-Control
• Savoias-Dataset (Saraee et al., 2018)^[2]. The Ground truth subfolder contains human complexity ratings.

The embeddings folder contains two folders with embeddings for all three aforementioned datasets obtained from the respective models:

• CLIP-CBA (built using CLIP-IQA (Wang et al., 2022^[3])
• CLIP-HBA

• The Experiments folder contains the code for the THINGS ranking experiment and the bistable 2-AFC experiment.
• The Data subfolder contains the human complexity ratings for both experiments.

• The Bayesian subfolder contains the code used to extract the optimal weights from a csv of embeddings and a target set of complexity scores.
• The BestWeights subfolder contains the best weights obtained from running the optimization on the three datasets.
• HBAOptimizations and CBAOptimizations contain notebooks that can be used to run the optimization on different datasets.
• The Neural subfolder contains code that can be used to create RDMs and compare them with THINGS MEG data.