ViSpa

ViSpa constructs high-dimensional vector spaces of vision-based representations by extracting feature vectors from naturalistic images using deep convolutional neural networks (DCNNs), enabling quantitative comparison of images and concept prototypes for studies of visual mental representations.

Key Features:

• High-Dimensional Vector Spaces: Represents images and concept prototypes as high-dimensional feature vectors for nuanced analysis and comparison.
• Deep Convolutional Neural Networks (DCNNs): Derives feature vectors from naturalistic images using DCNNs trained on image classification tasks.
• Vision-Based Similarity Scores: Computes similarity scores between pairs of images or concept prototypes based on their derived vectors.
• Behavioral Data Correlation: Validates computed similarities against human behavioral data including offline visual and typicality judgments and online discrimination and priming tasks.
• Experimental Material Control: Enables precise control and manipulation of image-based experimental materials, including combinations of images with words.
• Vision-Based Word-Pair Metric: Provides a novel vision-based similarity metric applicable to word-pair analyses.

Scientific Applications:

• Modeling Visual Mental Representations: Provides a quantitative, data-driven representation for testing theories of visual mental representations.
• Experimental Design and Stimulus Control: Supports selection and manipulation of image stimuli and image–word combinations in psychological experiments.
• Behavioral Prediction and Validation: Facilitates prediction and validation of human responses in visual and typicality judgments and in discrimination and priming paradigms.

Methodology:

ViSpa employs deep convolutional neural networks (DCNNs) to extract feature vectors from visual stimuli, constructs high-dimensional vector spaces from those vectors to represent images and concept prototypes, and computes vision-based similarity scores from the resulting vector representations.