VIOLET : End-to-End Video-Language Transformers with Masked Visual-tokenModeling

VIOLET proposes an end-to-end Video-Language Transformer that introduces Masked Visual-token Modeling (MVM) as a pre-training objective, enabling the model to learn joint video-text representations without relying on pre-extracted sparse features. The approach unifies video-language tasks under a single transformer architecture, achieving strong performance on downstream tasks such as video question answering and text-to-video retrieval.