HERMES shows that real-time video AI fails not from lack of memory, but from storing everything equally. By hierarchically compressing older context while preserving recent detail, it enables faster, more accurate streaming video understanding.
HERMES shows that real-time video AI fails not from lack of memory, but from storing everything equally. By hierarchically compressing older context while preserving recent detail, it enables faster, more accurate streaming video understanding.
Sign in to your account