sop-monitoring-blueprints

Run `npx skills add NVIDIA/sop-monitoring-blueprints --all -g -y` in your terminal to install all skills in this suite globally.

It watches industrial video feeds, detects whether operators perform assembly steps in the correct order, and flags missing or out-of-sequence actions. It combines a temporal boundary detection model with a vision-language model for action classification.

Annotate your videos with action start and end timestamps, then run the fine-tuning skills to generate QA training data and retrain the DDM and Cosmos-Reason models. The orchestrator skill runs the full augment, train, evaluate, and root-cause-analysis loop automatically.

You need an NVIDIA H100, H200, or A100 GPU with at least 80 GB of VRAM, driver 580 with CUDA 13, Docker, and the NVIDIA Container Toolkit. The deploy skill includes a preflight script that checks and auto-installs missing prerequisites.

Yes. The deploy skill includes an RTSP stream simulator that replays sample video as a live camera feed, so you can validate the full pipeline end to end without hardware.