Cross the Gap: Exposing the Intra-modal Misalignment in CLIP via Modality Inversion

Open in new window