Inverse-LLaVA

Researchers introduced Inverse-LLaVA, a multimodal learning architecture that projects text into visual space instead of aligning visual features with text, eliminating the need for alignment pretrain…