This page visualizes a large image collection within an interactive WebGL scene. Each image was processed with an Inception Convolutional Neural Network, trained on ImageNet 2012, and projected into a two-dimensional manifold with the UMAP algorithm such that similar images appear proximate to one another.