StableMaterials: Enhancing Diversity in Material Generation via Semi-Supervised Learning

Clip-based Nearest Neighbor search.

We demonstrate ability of StableMaterials to generate new materials by evaluating a CLIP nearest neighbor search on the training data. This analysis confirms that the model efficiently generalizes beyond mere replication of its training set.

Input
(StableMaterials generation)
Nearest Neigbor #1 Nearest Neigbor #2 Nearest Neigbor #3 Nearest Neigbor #4