deepstream-sop

Build and operate GPU-accelerated SOP video inference services with DeepStream and Triton.

49|14|Updated Aug 27, 2025
One-click install
npx skills add https://github.com/NVIDIA/sop-monitoring-blueprints --skill deepstream-sop
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: deepstream-sop
Source: https://github.com/NVIDIA/sop-monitoring-blueprints/tree/main/agentic/ds-sop-skills/deepstream-sop
Command: npx skills add https://github.com/NVIDIA/sop-monitoring-blueprints --skill deepstream-sop

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This skill helps developers build, deploy, debug, evaluate, and optimize GPU-accelerated video services that determine whether operators follow a defined manufacturing procedure in the correct sequence.

Core Features & Use Cases

  • End-to-End SOP Inference: Generate and customize a FastAPI microservice using DeepStream event-boundary detection, Triton-backed DDM models, and Cosmos Reason VLM classification.
  • Production Input and Output Support: Handle video files, RTSP streams, and Basler cameras through an OpenAI-compatible chat-completions API with SSE streaming, optional Kafka messaging, and Prometheus metrics.
  • Deployment and Validation: Configure Docker and Compose deployments, camera emulation, model loading, API verification, test suites, and file or live-stream latency measurements.
  • Use Case: Use the skill to generate a factory work-cell monitoring service that segments an assembly video, classifies each action, detects missing or out-of-order steps, and reports compliance results.

Quick Start

Ask the deepstream-sop skill to generate or evaluate a DeepStream SOP inference microservice for your industrial video workflow.

Frequently Asked Questions about deepstream-sop

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I build a GPU-accelerated SOP video monitoring service with DeepStream and Triton?

Build a GPU-accelerated SOP video monitoring service by generating a FastAPI microservice that uses DeepStream for event-boundary detection and Triton for model inference. This setup detects procedure boundaries, classifies actions, and validates step order for industrial video workflows.

What is the best way to validate step order in manufacturing video streams?

The best way to validate step order is to use a Cosmos Reason VLM classification pipeline that segments video streams, classifies each action, and detects missing or out-of-order steps. This ensures operators follow defined manufacturing procedures correctly.

Can I use Basler cameras and RTSP streams for real-time video analytics with FastAPI?

Yes, you can handle Basler cameras and RTSP streams for real-time video analytics through an OpenAI-compatible chat-completions API. The generated FastAPI microservice supports SSE streaming and optional Kafka messaging for production output.

How do I benchmark latency for live-stream video inference in Docker deployments?

Benchmark latency for live-stream video inference by configuring Docker and Compose deployments, then running file or live-stream latency measurements. The evaluation workflows include API verification, test suites, and performance measurement to validate inference speed.

Does DeepStream SOP inference support Kafka messaging and Prometheus metrics?

Yes, DeepStream SOP inference supports optional Kafka messaging for output and Prometheus metrics for monitoring. The microservice exposes an OpenAI-compatible chat-completions API with SSE streaming to integrate event data into downstream systems.

Why do I need Triton and DeepStream to detect procedure boundaries in factory video files?

You need Triton and DeepStream to detect procedure boundaries because DeepStream provides GPU-accelerated video stream parsing while Triton serves the DDM models. Together they enable real-time event-boundary detection and Cosmos Reason VLM classification for accurate SOP compliance.