Helm Deployment | NVIDIA Blueprint for Video Search and Summarization

The Warehouse Blueprint can also be deployed on Kubernetes using Helm, as an alternative to the Docker Compose method used in the Quickstart Guide. This guide provides an overview of the available Helm profiles and how to deploy them.

Overview

Three Warehouse Blueprint Helm charts are available under deploy/helm/industry-profiles/warehouse-operations/:

Profile Type	Chart	Detection Approach
2D Vision AI Profile	`warehouse-2d-app`	2D single-camera detection and tracking, using the RT-DETR model
3D Vision AI Profile	`warehouse-3d-app`	3D multi-camera detection and tracking, using the Sparse4D model
MV3DT Vision AI Profile	`warehouse-mv3dt-app`	3D multi-camera detection and tracking, using RT-DETR combined with MV3DT distributed tracking

Agents are optional settings in the 2D chart, not a separate chart. Auto Calibration is available through Docker Compose only.

Each of these charts wraps the shared microservice charts under deploy/helm/services/ (VIOS, RTVI, analytics, infra, monitoring, and calibration-import) and configures them for its own detection approach, in the same way that the corresponding Docker Compose profile does. See Deployment & Customization for a description of each profile’s architecture.

Prerequisites

Before you begin, ensure all of the prerequisites are met:

A Kubernetes cluster that you can reach with kubectl.
The NVIDIA GPU Operator, installed on the cluster so pods can request nvidia.com/gpu.
A StorageClass available for PVC provisioning. The charts create persistent volumes for VST, Elasticsearch, and related storage.
Helm 3.7+ and kubectl, installed on the machine you will deploy from.
An NGC API key, used for the image pull secret and the NGC model and app-data download job.
A TURN server, required for WebRTC playback in the VST/VIOS web UI whenever the browser is not on the same network as the cluster.

GPU requirements vary by profile. See the GPU requirements section in the chart README for the profile you are deploying, referenced in Chart Documentation below.

Deploy a Profile

This section guides you through the steps to deploy a Warehouse Blueprint profile on Kubernetes.

Download the Repository

Perform this on the machine where you intend to deploy the blueprint.

Git LFS required: The repository uses Git LFS for large files. Install it before cloning or pulling. For example, on Ubuntu/Debian: sudo apt-get install git-lfs. On other systems, see Git LFS installation.

git clone https://github.com/NVIDIA-AI-Blueprints/video-search-and-summarization.git
cd video-search-and-summarization
git checkout "develop"
git lfs install
git lfs pull

Stay at the Repository Root

Run the chart README commands from the video-search-and-summarization repository root; their paths are relative to that directory.

Follow the Chart README

Open the chart README for the profile you are deploying and follow its installation, validation, and uninstall instructions.

Chart Documentation

Profile	Chart Documentation
2D Vision AI Profile	warehouse-2d-app README
3D Vision AI Profile	warehouse-3d-app README
MV3DT Vision AI Profile	warehouse-mv3dt-app README

Each README covers, in order, prerequisites and GPU requirements, required secrets, web UI and ingress configuration, the full values.yaml reference, installation and post-install validation, the alerts stack (2D only), monitoring, stream scaling by GPU, and upgrade and uninstall instructions.

After running helm upgrade --install, wait until service pods are Running/Ready and init Jobs (e.g. the NGC download job) show Completed (check with kubectl get pods -n <NAMESPACE> -w) before opening the UI.

Common Configuration

Ingress, required secrets, and stream scaling are shared across all three profile charts. Alerts is available only on the 2D Vision AI Profile. See the chart README for a profile for the full detail on each.

Ingress

Ingress is disabled by default on all three profiles. To expose the web UIs, either set global.vssIngress.enabled=true and install an ingress controller (see the “Install the ingress controller” section in the chart README for your profile), or install with the bundled values-nodeport.yaml file to expose the UIs on NodePorts instead.

Required Secrets

Every profile requires an ngc-api secret containing the NGC API key and an ngc-docker-reg-secret image pull secret to be created in the release namespace before installing. See the “Required secrets” section in the chart README for your profile for the exact commands.

Stream Scaling

Each profile chart ships with a fixed default NUM_STREAMS and does not cap it to the available GPU automatically, unlike Docker Compose, which caps it per HARDWARE_PROFILE. Before an initial install, or an upgrade where the stream count should match your hardware, generate a values override using deploy/helm/industry-profiles/warehouse-operations/scripts/compute_stream_cap.py. See “Scaling: NUM_STREAMS by GPU” in the chart README for the profile you are deploying.

Alerts

The alerts stack is only available on the 2D Vision AI Profile, and is disabled by default there. Enabling it requires four components to be turned on together: the alert bridge, the agent, the agent UI, and either the in-cluster real-time VLM or an external VLM endpoint. See the “Alerts” section of the warehouse-2d-app README for the values needed to enable it.