Contrastive pretraining has advanced image-text understanding, yet image, video, and audio encoders are still trained independently, with cross-modal alignme…
机构:阿里
来源:arXiv 2611.05406 | AI4Papers 论文推荐平台