skills.rest
atomicrajat
atomicrajatCOMMUNITY
Explore all skills by this author →
GitHub

vss-deploy-detection-tracking-3d

Deploys the RTVI-CV-3D MV3DT microservice for multi-camera 3D detection and tracking with BEV Fusion.

2|Updated Aug 20, 2026
Search Tags:#docker compose#deepstream#mv3dt#camera calibration#nvidia vss#bev fusion#multi-camera tracking
One-click install
npx skills add https://github.com/atomicrajat/industry_safety_monitoring_system --skill vss-deploy-detection-tracking-3d-atomicrajat
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: vss-deploy-detection-tracking-3d
Source: https://github.com/atomicrajat/industry_safety_monitoring_system/tree/main/.claude/skills/vss-deploy-detection-tracking-3d
Command: npx skills add https://github.com/atomicrajat/industry_safety_monitoring_system --skill vss-deploy-detection-tracking-3d-atomicrajat

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Setting up multi-camera 3D detection and tracking with NVIDIA's Video Search and Summarization (VSS) warehouse blueprint requires coordinating DeepStream perception, BEV Fusion, camera calibration, MQTT/Kafka brokers, and VST sensor registration across dozens of containers — a process with many failure points around calibration paths, stream counts, GPU limits, and permissions. ## Core Features & Use Cases - Guided MV3DT deployment: Routes through four questions (profile size, data source, calibration coverage, detector/dataset slug) to deploy the RTVI-CV-3D stack on the bundled sample dataset, custom video files, or live RTSP streams. - Calibration chaining: Hands off to the AutoMagicCalib (AMC) skill when calibration is missing, runs VGGT refinement, exports MV3DT calibration files, and lands them at the correct mount path. - Camera configuration and verification: Normalizes camera names, syncs NUM_STREAMS with calibration and GPU stream caps, verifies perception FPS and BEV fusion health, and supports teardown and troubleshooting. - Use Case: An engineer with four calibrated warehouse cameras on an RTX PRO 6000 wants bounding-box overlays in the VST video wall without the full agent/LLM/VLM stack — this skill deploys the extended MV3DT profile end to end. ## Quick Start Ask the agent to deploy RTVI-CV-3D as MV3DT on your video files or RTSP streams, and it will route through calibration, camera configuration, deployment, and verification.

Frequently Asked Questions about vss-deploy-detection-tracking-3d

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

FAQPage Schema
How do I deploy multi-camera 3D detection and tracking with NVIDIA VSS?▼

Use this skill to deploy the RTVI-CV-3D microservice in MV3DT mode, which combines per-camera DeepStream perception with BEV Fusion over calibrated cameras. It supports the bundled sample dataset, custom MP4 videos, or live RTSP streams via docker compose.

How do I calibrate cameras for MV3DT BEV Fusion?▼

When calibration is missing, the skill chains to the vss-generate-video-calibration skill to run AutoMagicCalib end to end via its REST API, optionally with VGGT refinement. The exported calibration.json and camInfo files are landed at the warehouse-mv3dt-app calibration mount path.

Which GPUs support the MV3DT stack and how many streams?▼

Supported hardware profiles include RTX PRO 6000 Blackwell (18 streams), H100 (13), L40S (7), and IGX Thor or DGX Spark (4 each). The configurator enforces the per-GPU cap at deploy time by trimming the stream count.

Why does MV3DT perception show Active sources 0 after deployment?▼

This usually means stale VST sensor state from a prior deploy, a wrong dataset slug, missing calibration, or camera names that do not match the Camera, Camera_01 convention. Reconcile VST sensors or reset with docker compose down -v, then redeploy.

What is the difference between minimal and extended MV3DT profiles?▼

The extended profile (default) adds Elasticsearch, Logstash, Kibana, the video analytics API, and calibration import, which are required for bounding-box overlays in the VST video wall. Minimal runs only the MV3DT core with metadata on Kafka or Redis and no overlays.

When should I use the full warehouse blueprint instead of MV3DT only?▼

Use the vss-deploy-profile skill when you need the complete warehouse blueprint with the agent, LLM, and VLM stack, or Sparse4D with MODE=3d. This skill is strictly for the MV3DT microservice without those layers.

Related Skills

Explore More Software Engineering Skills→
NVIDIA-AI-BlueprintsNVIDIA-AI-Blueprints
1.8k

vss-generate-video-calibration

Run AutoMagicCalib camera calibration on MP4 files, RTSP streams, or sample datasets via REST API.

Official
Advanced
NVIDIA-AI-BlueprintsNVIDIA-AI-Blueprints
1.8k

vss-setup-behavior-analytics

Deploy the vss-behavior-analytics container standalone with chosen entrypoint, config, and calibration.

Official
Advanced
NVIDIA-AI-BlueprintsNVIDIA-AI-Blueprints
1.8k

vios

Manage video sensors and streams via VIOS REST API calls.

Official
Intermediate
NVIDIA-AI-BlueprintsNVIDIA-AI-Blueprints
1.8k

vss-deploy-detection-tracking-3d

Deploy the RTVI-CV-3D MV3DT microservice for multi-camera 3D detection and tracking.

Official
Advanced