This repository contains data and code associated with Maniquet, T., Op de Beeck, H. & Costantino., A. (2025) Recurrent issues with deep neural network models of visual recognition (see paper).

Human and DNN data is contained within the data folder:

• human_data.csv contains the preprocessed and aggregated data from all 218 participants.
• nn_data*.csv files contain the aggregated performance results from all DNN models used in the experiment, split into <100mb chunks.
• The processed folder contains the data aggregated, average, and otherwise processed to run the analyses contained in scripts.

All the code necessary to reproduce the analyses and figures is contained in the scripts folder. Scripts are divided into sequential steps that follow the structure of the results section.

All the stimuli from the human experiment are contained within the stimuli section. Images are divided into main manipulation conditions: control, clutter, occlusion and phase scrambling. Each challenging manipulation contains sub-conditions, contained in separate sub-folders.