Copernicus-Bench

DuoJEPA: Cross-Modal Translation Without Embedding Alignment

Researchers have developed DuoJEPA, a novel method for cross-modal translation that does not require direct alignment of embedding spaces. This approach uses self-supervised learning with a LeJEPA objective and adds tra…