VisCo: Leveraging Large Language Models as Intrinsic Encoders for Visual Token Compression

Part of

Relevance-Gated Hyperdimensional Memory Streaming