RTVI-CV 2D Detection & Tracking Operations
Deploy and operate the RTVI-CV 2D video detection and tracking microservice.
The skill documents prerequisites, credential purposes, user-overridable defaults, and some confirmation and recovery paths for reuse, parallel deployment, and teardown. However, it requires Docker, host networking, GPU access, host-directory mounts, and NGC credentials supplied through environment variables; privilege boundaries, sensitive-data flow, complete confirmation, and rollback coverage remain incomplete, so points are deducted.
Routing tables, helper-script inventory, troubleshooting, resource discovery, cache handling, and abnormal branches are detailed and mostly consistent. Static review cannot reproduce key paths, however, and execution depends on external images, NGC, Docker, and scripts whose implementations are not supplied; failure messaging and edge behavior therefore remain uncertain, warranting a conservative deduction.
The audience, deployment/debug/teardown/API scenarios, trigger phrases, and non-fit boundaries versus VLM, embedding, and analytics are relatively clear, with multiple platform declarations. There is no evidence of Chinese-language interaction support or mainland-China network reachability, and the environment requires a deployed service, GPU, NGC, and substantial tooling, so points are deducted.
The skill provides routing, progressive reference loading, prerequisites, limitations, troubleshooting, script arguments, version 3.2.1, Apache-2.0 licensing, and evaluation material. Author metadata is missing, and explicit maintenance ownership, changelog, dependency-version governance, and complete installation guidance are not clear, so points are deducted.
It covers standalone deployment, REST stream management, health probes, and metrics, with an output contract and examples that could make core tasks directly usable. Static material cannot verify actual results, several required references and scripts are not included in the evidence, and deployment is operationally costly; the score is capped by static calibration and reduced for unverified usability.
The pinned repository revision, version metadata, evaluation manifests, API examples, script inventory, and benchmark report provide some auditability. There is no independently reproducible CI, complete committed test suite, or corroborating evidence covering the key paths, and the benchmark report lacks verifiable execution detail, so the score remains limited.
- Execution may expose or consume NGC credentials and uses Docker host networking, GPU access, and host-directory mounts; confirm privilege scope, credential scope, ports, and data paths before use.
- Teardown and restart flows may affect existing containers and services; obtain explicit confirmation for destructive actions and verify recoverable configuration and assets.
- The required reference documents, script implementations, and actual image behavior are not fully present in the supplied evidence, so the benchmark report should not be treated as independent runtime verification.
- Chinese interaction quality and mainland-China network reachability are unproven; NGC, image, and model downloads may be affected by connectivity, rate limits, licensing, and GPU capacity.
What does this skill do, and when should you use it?
This skill targets the RTVI-CV 2D detection and tracking microservice used in NVIDIA VSS deployments. It guides agents through deployment, startup, debugging, teardown, and REST API operations. It requires a reachable VSS deployment, Docker, curl, jq, and appropriate NGC credentials. It does not cover VLM, embedding, or analytics tasks, which are routed to other vss-* skills.
It resolves platform and use-case deployment defaults, prepares NGC resources, synthesizes a Docker launch command, edits in-container configuration, and starts the application. Its REST workflows add, remove, and discover video streams, probe /live, /ready, and /startup, collect /api/v1/metrics, and support embedding-related calls. Helper scripts also cover logs, GPU checks, engine caches, calibration, and troubleshooting.
- A VSS operator needs to deploy RTVI-CV 2D detection and tracking for a warehouse or smart-city workload.
- A platform engineer needs to launch the perception container on an x86/aarch64 dGPU, SBSA, or Jetson host.
- A video-analytics developer needs to add or remove camera streams from a running RTVI-CV instance.
- An operator needs to check liveness, readiness, FPS, GPU usage, or service metrics.
- An engineer needs to diagnose connection refusals, NGC authentication failures, or model-loading OOM errors.
What are this skill's strengths and limitations?
- Unifies deployment, debugging, teardown, and REST API workflows.
- Documents support for multiple NVIDIA GPU, SBSA, and Jetson platforms.
- Includes stream-management, health-check, metrics, and resource-diagnostic helpers.
- Provides explicit routing for static and dynamic stream workflows.
- Requires an existing reachable VSS deployment and matching microservice environment.
- Needs Docker, curl, jq, NGC credentials, and suitable hardware resources.
- NGC models and NIMs may be constrained by rate limits, GPU memory, and licensing.
- The full workflows depend on references/, scripts/, and configuration files not included in the supplied material.
How do you install this skill?
Install it from the NVIDIA/skills catalog with: npx skills add nvidia/skills --skill vss-deploy-detection-tracking-2d --yes. You can target a client explicitly, for example: npx skills add nvidia/skills --skill vss-deploy-detection-tracking-2d --agent codex. The skill becomes available when the agent next loads skills and encounters a matching task.
How do you use this skill?
Give the agent an intent-specific request such as “deploy rtvi-cv warehouse 2d with 4 streams and display”, “add a stream to rtvi-cv”, or “check rtvi-cv health”. Deployment, teardown, and debugging require the corresponding deploy reference document; API usage assumes an already-running instance reachable at http://<host>:9000. The supplied material does not include the referenced documents or complete script parameters.