NVIDIA Metropolis VSS
A video search and summarization reference architecture blueprint developed by NVIDIA. It combines VLMs, NLMs, embedding models, and computer vision models to transform surveillance footage into an intelligent system capable of conversational querying, search, and alerting.
Timeline (last 90 days)
NVIDIA Metropolis VSS (Video Search and Summarization blueprint) is composed of four types of models working together: a vision-language model, an NLM responsible for reasoning and responses, an embedding model, and a computer vision model for fast detection and tracking.
In terms of VSS performance, deep retrieval can return results within 5 seconds, alerts can be triggered within 1 second, and event summarization can run at approximately 60x speed.
VSS is essentially a reference architecture rather than an out-of-the-box finished product; it is highly modular, supports replacement of proprietary detection models, VLMs, embeddings, and databases, and can run on edge, on-premises, cloud, or hybrid deployments.
All Facts (3)
NVIDIA Metropolis VSS (Video Search and Summarization blueprint) is composed of four types of models working together: a vision-language model, an NLM responsible for reasoning and responses, an embedding model, and a computer vision model for fast detection and tracking.
50%UnverifiedIn terms of VSS performance, deep retrieval can return results within 5 seconds, alerts can be triggered within 1 second, and event summarization can run at approximately 60x speed.
50%UnverifiedVSS is essentially a reference architecture rather than an out-of-the-box finished product; it is highly modular, supports replacement of proprietary detection models, VLMs, embeddings, and databases, and can run on edge, on-premises, cloud, or hybrid deployments.
50%