Nearest Neighbors Comparisons

Semantic Collection Network web application screenshot

The Nearest Neighbors Comparisons are in-browser visualizations to better understand the concept of “similarity” when pre-trained deep learning models such as ResNet50 and VGG19 are used for feature extraction in image analysis. These models are commonly used to measure similarity between images, although their underlying representations are not always easy to interpret. The tool aims to offer means to evaluate the aesthetic similarity between images by comparing different methods and modelsnext to each other. It facilitates the comparison of distance metrics such as cosine similarity and Euclidean distance to perform nearest neighbor searches. With the help of this visualization tool, researchers can gain a visual understanding of the aesthetic similarity of the displayed images.