🟩NVIDIA Developer Blog•較早收集於 30m
NVIDIA 發布用於影片搜尋與摘要的 Metropolis Blueprint
💡了解如何利用 NVIDIA 最新的 AI 代理藍圖，將海量影片存檔轉化為可搜尋的數據。
⚡ 30-Second TL;DR
有什麼變化
實現對數百萬個即時與錄製影片串流的即時搜尋功能。
為什麼重要
此工具顯著縮短了安全與營運團隊分析影片素材所需的時間。它使企業能夠在不增加人力成本的情況下，擴展其影片監控能力。
下一步行動
查閱 NVIDIA Metropolis Blueprint 文件，評估您目前的影片處理流程是否能整合這些 AI 代理以實現自動化索引。
誰應關注：Enterprise & Security Teams
關鍵要點
- •實現對數百萬個即時與錄製影片串流的即時搜尋功能。
- •利用 AI 代理與技能將原始影像轉化為具備行動價值的情報。
- •專為解決從海量數據驅動的影片環境中提取洞察的挑戰而設計。
🧠 深度解析
Web-grounded analysis with 7 cited sources.
🔑 增強重點摘要
- •The Metropolis Blueprint for VSS leverages advanced AI models including NVIDIA VILA, NVIDIA Llama Nemotron, NVIDIA NeMo Retriever microservices, and retrieval-augmented generation (RAG) to connect large language models (LLMs) to enterprise data.
- •It significantly accelerates video summarization, capable of condensing an hour-long video into a text summary in less than one minute, representing a 100x speed improvement over real-time viewing.
- •The blueprint supports a wide range of hardware deployments, from single NVIDIA A100 or H100 GPUs for smaller workloads to edge deployments on NVIDIA RTX 6000 PRO and NVIDIA DGX Spark computing platforms.
- •Beyond summarization, the VSS blueprint enables various agent workflows including natural language search across video archives, interactive question and answering, alerts, event review and verification, and object tracking.
- •Early adopters like electronics manufacturing firm Pegatron have reported significant operational improvements, including a 7% reduction in labor costs and a 67% decrease in defect rates by utilizing AI agents built with the VSS blueprint.
🛠️ 技術深入
- Built on the NVIDIA Metropolis platform, a comprehensive developer platform for automating physical processes.
- Utilizes Vision Language Models (VLMs) and Large Language Models (LLMs) such as NVIDIA VILA, NVIDIA Llama Nemotron, and NVIDIA Cosmos Reason.
- Incorporates NVIDIA NeMo Retriever microservices for connecting LLMs to enterprise data and Retrieval-Augmented Generation (RAG) to enhance accuracy and reduce hallucinations.
- Leverages NVIDIA NIM microservices, including `cosmos-reason2-8b` and `nemotron-nano-9b-v2`, for VLMs, LLMs, and advanced AI frameworks.
- Features a scalable video ingestion pipeline to process hundreds of live video streams or burst clips simultaneously.
- Supports deployment on a broad range of NVIDIA GPUs and platforms, including A100, H100, RTX 6000 PRO, DGX Spark, Jetson Thor, B200, H200, L40/L40S, and A6000.
- Provided as customizable agentic workflow examples, including reference code, documentation, and Docker Compose for streamlined deployment.
- Offers multimodal model fusion and audio transcription capabilities in addition to visual understanding.
🔮 前景展望AI analysis grounded in cited sources
The Metropolis Blueprint will significantly accelerate the adoption of AI-powered video analytics across physical industries.
By providing a customizable, end-to-end framework with pre-built components and hardware support, it lowers the barrier for enterprises to deploy complex video AI agents.
Natural language interaction will become the primary interface for managing and extracting insights from vast video data archives.
The blueprint's integration of VLMs and LLMs enables users to query and summarize video content using natural language, moving beyond traditional manual review.
The blueprint will foster a new wave of specialized AI agents tailored for specific industrial and urban challenges.
Its modular design and customizable workflows allow developers to create agents for diverse applications like worker safety, traffic management, and quality control.
⏳ 時間線
2017-05
NVIDIA Metropolis intelligent video analytics platform announced at GTC.
2023
NVIDIA NeMo Retriever microservices first made available, later integrated into VSS Blueprint.
2025-01
NVIDIA releases a YouTube video detailing the Metropolis AI Blueprint for video search and summarization.
2025-03
Superb AI announces integration of NVIDIA AI Blueprint for Video Search and Summarization into its solutions.
2025-05
NVIDIA AI Blueprint for Video Search and Summarization becomes generally available.
2026-01
NVIDIA releases open models and blueprints, including an updated VSS Blueprint, built on Cosmos.
📎 來源 (7)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
AI 週報
閱讀本週精選 AI 大事摘要 →
👉相關動態
AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源： NVIDIA Developer Blog ↗