open-edge-platform · rohitkatakol · Mar 11, 2026 · Mar 10, 2026 · Mar 10, 2026 · Mar 10, 2026
@@ -4,7 +4,7 @@ HOST_IP=
 REST_SERVER_PORT=8080
 
 # DL Streamer Pipeline Server
-DLSTREAMER_PIPELINE_SERVER_IMAGE=intel/dlstreamer-pipeline-server:2026.0.0-ubuntu24-rc1
+DLSTREAMER_PIPELINE_SERVER_IMAGE=intel/dlstreamer-pipeline-server:2026.0.0-ubuntu24-rc2
 RTSP_CAMERA_IP=
 
 # MinIO service & client

@@ -4,7 +4,7 @@ HOST_IP=
 REST_SERVER_PORT=8080
 
 # DL Streamer Pipeline Server
-DLSTREAMER_PIPELINE_SERVER_IMAGE=intel/dlstreamer-pipeline-server:2026.0.0-ubuntu24-rc1
+DLSTREAMER_PIPELINE_SERVER_IMAGE=intel/dlstreamer-pipeline-server:2026.0.0-ubuntu24-rc2
 RTSP_CAMERA_IP=
 
 # MinIO service & client

@@ -4,7 +4,7 @@ HOST_IP=
 REST_SERVER_PORT=8080
 
 # DL Streamer Pipeline Server
-DLSTREAMER_PIPELINE_SERVER_IMAGE=intel/dlstreamer-pipeline-server:2026.0.0-ubuntu24-rc1
+DLSTREAMER_PIPELINE_SERVER_IMAGE=intel/dlstreamer-pipeline-server:2026.0.0-ubuntu24-rc2
 RTSP_CAMERA_IP=
 
 # MinIO service & client

@@ -4,7 +4,7 @@ HOST_IP=
 REST_SERVER_PORT=8080
 
 # DL Streamer Pipeline Server
-DLSTREAMER_PIPELINE_SERVER_IMAGE=intel/dlstreamer-pipeline-server:2026.0.0-ubuntu24-rc1
+DLSTREAMER_PIPELINE_SERVER_IMAGE=intel/dlstreamer-pipeline-server:2026.0.0-ubuntu24-rc2
 RTSP_CAMERA_IP=
 
 # MinIO service & client

@@ -4,7 +4,7 @@ HOST_IP=
 REST_SERVER_PORT=8080
 
 # DL Streamer Pipeline Server
-DLSTREAMER_PIPELINE_SERVER_IMAGE=intel/dlstreamer-pipeline-server:2026.0.0-ubuntu24-rc1
+DLSTREAMER_PIPELINE_SERVER_IMAGE=intel/dlstreamer-pipeline-server:2026.0.0-ubuntu24-rc2
 RTSP_CAMERA_IP=
 
 # MinIO service & client

@@ -3,7 +3,7 @@
 # Download artifacts for a specific sample application
 #   by calling respective app's setup.sh script
 SCRIPT_DIR=$(dirname $(readlink -f "$0"))
-MODEL_URL="https://github.com/open-edge-platform/edge-ai-resources/raw/1d40dfe1791d44e8cf6e8472c28c034e40fa508d/models/INT8/pallet_defect_detection.zip"
+MODEL_URL="https://github.com/open-edge-platform/edge-ai-resources/raw/06bb0d621cb14a1791672552a538beddddcc4066/models/INT8/pallet_defect_detection.zip"
 VIDEO_URL="https://github.com/open-edge-platform/edge-ai-resources/raw/c13b8dbf23d514c2667d39b66615bd1400cb889d/videos/warehouse.avi"
 
 err() {

@@ -5,7 +5,7 @@
                 "name": "worker_safety_gear_detection_mlops",
                 "source": "gstreamer",
                 "queue_maxsize": 50,
-                "pipeline": "multifilesrc loop=TRUE location=/home/pipeline-server/resources/videos/Safety_Full_Hat_and_Vest.avi name=source ! h264parse ! decodebin3 ! gvadetect name=detection ! gvametaconvert add-empty-results=true name=metaconvert ! queue ! gvafpscounter ! appsink name=destination",
+                "pipeline": "multifilesrc loop=TRUE location=/home/pipeline-server/resources/videos/Safety_Full_Hat_and_Vest.avi name=source ! h264parse ! decodebin3 ! gvadetect name=detection threshold=0.4 ! gvametaconvert add-empty-results=true name=metaconvert ! queue ! gvafpscounter ! appsink name=destination",
                 "parameters": {
                     "type": "object",
                     "properties": {
@@ -23,7 +23,7 @@
                 "name": "worker_safety_gear_detection_mqtt",
                 "source": "gstreamer",
                 "queue_maxsize": 50,
-                "pipeline": "{auto_source} name=source ! decodebin3 ! gvadetect name=detection model-instance-id=inst0 ! gvametaconvert add-empty-results=true name=metaconvert ! queue ! gvawatermark displ-cfg=font-scale=1.0,draw-txt-bg=false ! queue ! gvafpscounter ! queue ! jpegenc ! appsink name=destination",
+                "pipeline": "{auto_source} name=source ! decodebin3 ! gvadetect name=detection model-instance-id=inst0 threshold=0.4 ! gvametaconvert add-empty-results=true name=metaconvert ! queue ! gvawatermark displ-cfg=font-scale=1.0,draw-txt-bg=false ! queue ! gvafpscounter ! queue ! jpegenc ! appsink name=destination",
                 "parameters": {
                     "type": "object",
                     "properties": {
@@ -41,7 +41,7 @@
                 "name": "worker_safety_gear_detection_s3write",
                 "source": "gstreamer",
                 "queue_maxsize": 50,
-                "pipeline": "{auto_source} name=source ! decodebin3 ! gvadetect name=detection model-instance-id=inst0 ! gvametaconvert add-empty-results=true name=metaconvert ! queue ! gvafpscounter ! queue ! jpegenc ! appsink name=destination",
+                "pipeline": "{auto_source} name=source ! decodebin3 ! gvadetect name=detection model-instance-id=inst0 threshold=0.4 ! gvametaconvert add-empty-results=true name=metaconvert ! queue ! gvafpscounter ! queue ! jpegenc ! appsink name=destination",
                 "parameters": {
                     "type": "object",
                     "properties": {
@@ -64,7 +64,7 @@
                 "name": "worker_safety_gear_detection_opcua",
                 "source": "gstreamer",
                 "queue_maxsize": 50,
-                "pipeline": "{auto_source} name=source ! decodebin3 ! gvadetect name=detection model-instance-id=inst0 ! gvametaconvert add-empty-results=true name=metaconvert ! queue ! gvawatermark displ-cfg=font-scale=1.0,draw-txt-bg=false ! queue ! gvafpscounter ! queue ! jpegenc ! appsink name=destination",
+                "pipeline": "{auto_source} name=source ! decodebin3 ! gvadetect name=detection model-instance-id=inst0 threshold=0.4 ! gvametaconvert add-empty-results=true name=metaconvert ! queue ! gvawatermark displ-cfg=font-scale=1.0,draw-txt-bg=false ! queue ! gvafpscounter ! queue ! jpegenc ! appsink name=destination",
                 "parameters": {
                     "type": "object",
                     "properties": {
@@ -86,7 +86,7 @@
                 "name": "worker_safety_gear_detection_gpu",
                 "source": "gstreamer",
                 "queue_maxsize": 50,
-                "pipeline": "{auto_source} name=source ! decodebin3 ! gvadetect device=GPU model-instance-id=instgpu0 inference-region=full-frame inference-interval=1 batch-size=8 nireq=2 ie-config=\"GPU_THROUGHPUT_STREAMS=2\" threshold=0.7 name=detection ! queue ! gvametaconvert add-empty-results=true name=metaconvert ! queue ! gvafpscounter ! appsink name=destination",
+                "pipeline": "{auto_source} name=source ! decodebin3 ! gvadetect device=GPU model-instance-id=instgpu0 inference-region=full-frame inference-interval=1 batch-size=8 nireq=2 ie-config=\"GPU_THROUGHPUT_STREAMS=2\" threshold=0.4 name=detection ! queue ! gvametaconvert add-empty-results=true name=metaconvert ! queue ! gvafpscounter ! appsink name=destination",
                 "parameters": {
                     "type": "object",
                     "properties": {
@@ -104,7 +104,7 @@
                 "name": "worker_safety_gear_detection_npu",
                 "source": "gstreamer",
                 "queue_maxsize": 50,
-                "pipeline": "{auto_source} name=source ! decodebin3 ! gvadetect device=NPU model-instance-id=instnpu0 inference-region=full-frame inference-interval=1 batch-size=1 nireq=4 threshold=0.7 name=detection ! queue ! gvametaconvert add-empty-results=true name=metaconvert ! queue ! gvafpscounter ! appsink name=destination",
+                "pipeline": "{auto_source} name=source ! decodebin3 ! gvadetect device=NPU model-instance-id=instnpu0 inference-region=full-frame inference-interval=1 batch-size=1 nireq=4 threshold=0.4 name=detection ! queue ! gvametaconvert add-empty-results=true name=metaconvert ! queue ! gvafpscounter ! appsink name=destination",
                 "parameters": {
                     "type": "object",
                     "properties": {
@@ -122,7 +122,7 @@
                 "name": "worker_safety_gear_detection",
                 "source": "gstreamer",
                 "queue_maxsize": 50,
-                "pipeline": "{auto_source} name=source ! decodebin3 ! gvadetect device=CPU model-instance-id=inst0 name=detection ! queue ! gvametaconvert add-empty-results=true name=metaconvert ! queue ! gvafpscounter ! appsink name=destination",
+                "pipeline": "{auto_source} name=source ! decodebin3 ! gvadetect device=CPU model-instance-id=inst0 threshold=0.4 name=detection ! queue ! gvametaconvert add-empty-results=true name=metaconvert ! queue ! gvafpscounter ! appsink name=destination",
                 "parameters": {
                     "type": "object",
                     "properties": {

@@ -3,7 +3,7 @@
 # Download artifacts for a specific sample application
 #   by calling respective app's setup.sh script
 SCRIPT_DIR=$(dirname $(readlink -f "$0"))
-MODEL_URL="https://github.com/open-edge-platform/edge-ai-resources/raw/1d40dfe1791d44e8cf6e8472c28c034e40fa508d/models/INT8/worker-safety-gear-detection.zip"
+MODEL_URL="https://github.com/open-edge-platform/edge-ai-resources/raw/06bb0d621cb14a1791672552a538beddddcc4066/models/INT8/worker-safety-gear-detection.zip"
 VIDEO_URL="https://github.com/open-edge-platform/edge-ai-resources/raw/edd25f37c324a9ef73df1642354b2ba5fa7b7df5/videos/Safety_Full_Hat_and_Vest.avi"
 
 err() {

@@ -58,6 +58,7 @@
    MTX_WEBRTCICESERVERS2_0_USERNAME=<username>  # WebRTC credentials e.g. intel1234
    MTX_WEBRTCICESERVERS2_0_PASSWORD=<password>
    ```
+    > **Note:** For GPU/NPU based pipelines, set `privileged_access_required: true` in the `helm/values_<SAMPLE_APP>.yaml` file to enable access to host hardware devices.
 
 4. Install pre-requisites for all instances
 
@@ -806,7 +807,7 @@ Applications can take advantage of S3 publish feature from DL Streamer Pipeline
    >NOTE- For sake of simplicity, we assume that the new model has already been downloaded by Model Download microservice. The following curl command is only a simulation that just downloads the model. In production, however, they will be downloaded by the Model Download service.
 
    ```sh
-   export MODEL_URL='https://github.com/open-edge-platform/edge-ai-resources/raw/1d40dfe1791d44e8cf6e8472c28c034e40fa508d/models/INT8/pallet_defect_detection.zip'
+   export MODEL_URL='https://github.com/open-edge-platform/edge-ai-resources/raw/06bb0d621cb14a1791672552a538beddddcc4066/models/INT8/pallet_defect_detection.zip'
 
    curl -L "$MODEL_URL" -o "$(basename $MODEL_URL)"
 

@@ -23,6 +23,7 @@
    ```sh
    cp helm/values_pallet-defect-detection.yaml helm/values.yaml
    ```
+      > **Note:** For GPU/NPU based pipelines, set `privileged_access_required: true` in the `helm/values.yaml` file to enable access to host hardware devices.
 
 3. Optional: Pull the helm chart and replace the existing helm folder with it
 
@@ -32,13 +33,13 @@
    - Download helm chart with the following command
 
      ```bash
-     helm pull oci://registry-1.docker.io/intel/pallet-defect-detection-reference-implementation --version 2.6.0-rc1
+     helm pull oci://registry-1.docker.io/intel/pallet-defect-detection-reference-implementation --version 2.6.0-rc2
      ```
 
    - Unzip the package using the following command
 
      ```bash
-     tar -xvf pallet-defect-detection-reference-implementation-2.6.0-rc1.tgz
+     tar -xvf pallet-defect-detection-reference-implementation-2.6.0-rc2.tgz
      ```
 
    - Replace the helm directory
@@ -408,7 +409,7 @@ Applications can take advantage of S3 publish feature from DL Streamer Pipeline
    >NOTE- For sake of simplicity, we assume that the new model has already been downloaded by Model Download microservice. The following curl command is only a simulation that just downloads the model. In production, however, they will be downloaded by the Model Download service.
 
    ```sh
-   export MODEL_URL='https://github.com/open-edge-platform/edge-ai-resources/raw/1d40dfe1791d44e8cf6e8472c28c034e40fa508d/models/INT8/pallet_defect_detection.zip'
+   export MODEL_URL='https://github.com/open-edge-platform/edge-ai-resources/raw/06bb0d621cb14a1791672552a538beddddcc4066/models/INT8/pallet_defect_detection.zip'
 
    curl -L "$MODEL_URL" -o "$(basename $MODEL_URL)"
 

@@ -6,6 +6,7 @@ This section collects guides for Pallet Defect Detection sample application.
 - [Manage pipelines](./how-to-guides/manage-pipelines.md)
 - [Run multiple AI pipelines](./how-to-guides/run-multiple-ai-pipelines.md)
 - [Use GPU For Inference](./how-to-guides/use-gpu-for-inference.md)
+- [Use NPU For Inference](./how-to-guides/use-npu-for-inference.md)
 - [Use Your AI Model and Video](./how-to-guides/use-your-ai-model-and-video.md)
 - [Change the Input Video Source](./how-to-guides/change-input-video-source.md)
 - [Scale Video Resolution](./how-to-guides/scale-video-resolution.md)
@@ -27,6 +28,7 @@ This section collects guides for Pallet Defect Detection sample application.
 ./how-to-guides/manage-pipelines
 ./how-to-guides/run-multiple-ai-pipelines
 ./how-to-guides/use-gpu-for-inference
+./how-to-guides/use-npu-for-inference
 ./how-to-guides/use-your-ai-model-and-video
 ./how-to-guides/change-input-video-source
 ./how-to-guides/scale-video-resolution

@@ -24,7 +24,7 @@ cd edge-ai-libraries/microservices/dlstreamer-pipeline-server
 Create a Docker file named `BalluffDockerfile` inside your `dlstreamer-pipeline-server` directory with the following content.
 
 ```dockerfile
-FROM intel/dlstreamer-pipeline-server:2026.0.0-ubuntu24-rc1
+FROM intel/dlstreamer-pipeline-server:2026.0.0-ubuntu24-rc2
 
 USER root
 

@@ -22,7 +22,7 @@ cd edge-ai-libraries/microservices/dlstreamer-pipeline-server
 Create a Docker file named `BaslerDockerfile` inside your `dlstreamer-pipeline-server` directory with the following content.
 
 ```dockerfile
-FROM intel/dlstreamer-pipeline-server:2026.0.0-ubuntu24-rc1
+FROM intel/dlstreamer-pipeline-server:2026.0.0-ubuntu24-rc2
 
 USER root
 

@@ -0,0 +1,98 @@
+# How to use NPU for inference
+
+## Pre-requisites
+
+To take full advantage of hardware acceleration, pipelines can be designed so that different stages—such as decoding and inference—are executed on the most suitable hardware devices.
+
+Low-power accelerators like a Neural Processing Unit (NPU) can offload neural network computation from the CPU or GPU, enabling more efficient resource utilization and improved overall system performance.
+
+DLStreamer and the DLStreamer Pipeline Server support inference on NPU devices, allowing applications built on these frameworks to leverage NPU acceleration for improved efficiency and performance.
+
+Before running inference on an NPU, ensure that:
+- The host system includes a supported NPU device
+- The required NPU drivers are installed and properly configured
+
+For detailed setup instructions, refer to the [documentation](https://docs.openedgeplatform.intel.com/2026.0/edge-ai-libraries/dlstreamer/dev_guide/advanced_install/advanced_install_guide_prerequisites.html#optional-prerequisite-2-install-intel-npu-drivers).
+
+ For containerized application, following additional changes are required.
+
+### Provide NPU access to the container
+
+This can be done by making the following changes to the docker compose file.
+
+```yaml
+services:
+  dlstreamer-pipeline-server:
+    group_add:
+      # render group ID for ubuntu 22.04 host OS
+      - "110"
+      # render group ID for ubuntu 24.04 host OS
+      - "992"
+    devices:
+      # you can add specific devices in case you don't want to provide access to all like below.
+      - "/dev:/dev"
+```
+The changes above adds the container user to the `render` group and provides access to the NPU devices.
+
+### Hardware specific encoder/decoders
+
+Unlike the changes done for the container above, the following requires a modification to the media pipeline itself.
+
+Gstreamer has a variety of hardware specific encoders and decoders elements such as Intel specific VA-API elements that you can benefit from by adding them into your media pipeline. Examples of such elements are `vah264dec`, `vah264enc`, `vajpegdec`, `vajpegdec`, etc.
+
+Additionally, one can also enforce zero-copy of buffers using GStreamer caps (capabilities) to the pipeline by adding `video/x-raw(memory: VAMemory)` for Intel NPUs.
+
+Read DL Streamer [docs](https://dlstreamer.github.io/dev_guide/gpu_device_selection.html) for more details.
+
+### NPU specific element properties
+
+DL Streamer inference elements also provides property such as `device=NPU` and `pre-process-backend=va` which should be used in pipelines with NPU memory. It performs mapping to the system memory and uses VA pre-processor. Read DL Streamer [docs](https://dlstreamer.github.io/dev_guide/model_preparation.html#model-pre-and-post-processing) for more.
+
+## Tutorial on how to use NPU specific pipelines
+
+> Note - This sample application already provides a default `docker-compose.yml` file that includes the necessary NPU access to the containers.
+
+The pipeline `pallet_defect_detection_npu` in `pipeline-server-config.json` contains NPU specific elements and uses NPU backend for inferencing. Follow the steps below to run the pipeline.
+
+### Steps
+
+1. Ensure that the sample application is up and running. If not, follow the steps [here](../get-started.md#set-up-the-application) to setup the application and then bring the services up
+
+    >If you're running multiple instances of app, start the services using `./run.sh up` instead.
+
+    ```sh
+    docker compose up -d
+    ```
+2. Start the pipeline.
+    ```sh
+    ./sample_start.sh -p pallet_defect_detection_npu
+    ```
+
+    This will start the pipeline. The inference stream can be viewed on WebRTC, in a browser, at the following url:
+
+    >If you're running multiple instances of app, ensure to provide `NGINX_HTTPS_PORT` number in the url for the app instance i.e. replace <HOST_IP> with <HOST_IP>:<NGINX_HTTPS_PORT>
+
+    ```bash
+    https://<HOST_IP>/mediamtx/pdd/
+    ```
+
+## Deploying with Helm
+
+### Intel GPU K8S Extension
+
+If you're deploying a NPU based pipeline (example: with VA elements like `vapostproc`, `vah264dec` etc., and/or with `device=NPU` in `gvadetect` in `dlstreamer_pipeline_server_config.json`) with Intel GPU k8s Extension, ensure to set the below details in the file `helm/values.yaml` appropriately in order to utilize the underlying NPU.
+
+```sh
+gpu:
+  enabled: true
+  type: "gpu.intel.com/i915"
+  count: 1
+```
+
+### Without Intel GPU K8S Extension
+
+If you're deploying a NPU based pipeline (example: with VA elements like `vapostproc`, `vah264dec` etc., and/or with `device=NPU` in `gvadetect` in `dlstreamer_pipeline_server_config.json`) without Intel GPU k8s Extension, ensure to set the below details in the file `helm/values.yaml` appropriately in order to utilize the underlying NPU.
+
+```sh
+privileged_access_required: true
+```
@@ -58,6 +58,7 @@
    MTX_WEBRTCICESERVERS2_0_USERNAME=<username>  # WebRTC credentials e.g. intel1234
    MTX_WEBRTCICESERVERS2_0_PASSWORD=<password>
    ```
+    > **Note:** For GPU/NPU based pipelines, set `privileged_access_required: true` in the `helm/values_<SAMPLE_APP>.yaml` file to enable access to host hardware devices.
 
 4. Install pre-requisites for all instances
 

@@ -23,17 +23,18 @@
    ```sh
    cp helm/values_pcb-anomaly-detection.yaml helm/values.yaml
    ```
+      > **Note:** For GPU/NPU based pipelines, set `privileged_access_required: true` in the `helm/values.yaml` file to enable access to host hardware devices.
 
 3. Optional: Pull the helm chart and replace the existing helm folder with it
 
    - Note: The helm chart should be downloaded when you are not using the helm chart provided in `edge-ai-suites/manufacturing-ai-suite/industrial-edge-insights-vision/helm`
 
    - Download helm chart with the following command
 
-       `helm pull oci://registry-1.docker.io/intel/pcb-anomaly-detection --version 1.2.0-rc1`
+       `helm pull oci://registry-1.docker.io/intel/pcb-anomaly-detection --version 1.2.0-rc2`
    - unzip the package using the following command
 
-       `tar -xvf pcb-anomaly-detection-1.2.0-rc1.tgz`
+       `tar -xvf pcb-anomaly-detection-1.2.0-rc2.tgz`
    - Replace the helm directory
 
        `rm -rf helm && mv pcb-anomaly-detection helm`

@@ -58,6 +58,7 @@
    MTX_WEBRTCICESERVERS2_0_USERNAME=<username>  # WebRTC credentials e.g. intel1234
    MTX_WEBRTCICESERVERS2_0_PASSWORD=<password>
    ```
+    > **Note:** For GPU/NPU based pipelines, set `privileged_access_required: true` in the `helm/values_<SAMPLE_APP>.yaml` file to enable access to host hardware devices.
 
 4. Install pre-requisites for all instances