Enabling Multimodal Generation on CLIP via Vision-Language Knowledge Distillation

32646

Related Entities