LensVLM: Compressing long context as images, expanding only relevant pages

Unlike previous methods that process full documents, LensVLM introduces a novel approach by compressing long contexts as images and expanding only relevant pages.