This is a release of the stimuli (in both single-image and multiple-images formats) used in the evaluation of our paper. Stimuli for the KiVA-adults benchmark are prefixed with "adults_". Stimuli for the KiVA-compositionality benchmark are prefixed with "compo_".
For further details please refer to the documentation in the README of the project.
Notes:
-
Each trial is presented as a multiple-choice question and is repeated three times, with the image choices in each repetition assigned to randomized choice labels.
-
For each transformation concept, we provide a JSON file that details the specific transformation, the trial number, the number of trial repetitions, the correct image label, the incorrect image label, the no-change image label, as well as the input and output parameters for both the training and test transformations.