Releases: open-edge-platform/dlstreamer
2026.0.0
Release Notes: Deep Learning Streamer (DL Streamer) Pipeline Framework Release 2026.0
Version 2026.0
Key highlights:
- New elements: gvafpsthrottle, g3dradarprocess, g3dlidarparse
- New model support: YOLOv26, YOLO-E, RT-DETR, HuggingFace ViT
- Streamlined integration with Ultralytics and HuggingFace model hubs
- GstAnalytics metadata support: DLStreamer supports GstAnalytics metadata for object detection, classification, tracking and adds custom GstAnalytics extension for keypoints
- gvawatermark overhaul: object bluring, text backgrounds, label filtering, extra fonts, thickness/color options, FPS overlay
- Inference enhancements: batch timeout, OpenCV tensor compression for all devices
- Windows platform: GPU inference via D3D11, gvapython support, CI integration, build/setup improvements
- New Python samples: VLM Alerts, Smart NVR, ONVIF Discovery, face detection/age classification, open-vocabulary detection, RealSense, DL Streamer + DeepStream
- Optimizer: multi-stream optimization, cross-stream batching, device selection, refactored with tests
- Component updates: OpenVINO 2026.0.0, NPU driver 1.30, RealSense SDK 2.57.5
- Library consolidation: merged gvawatermark3d, gvadeskew, gvamotiondetect, gvagenai into gstvideoanalytics
- CI: Zizmor security scanning, Windows CI, Docker image size checks
Deep Learning Streamer (DL Streamer) Pipeline Framework is a streaming media analytics framework, based on GStreamer* multimedia framework, for creating complex media analytics pipelines. It ensures pipeline interoperability and provides optimized media, and inference operations using Intel® Distribution of OpenVINO™ Toolkit Inference Engine backend, across Intel® architecture, CPU, discrete GPU, integrated GPU and NPU.
The complete solution leverages:
-
Open source GStreamer* framework for pipeline management
-
GStreamer* plugins for input and output such as media files and real-time streaming from camera or network
-
Video decode and encode plugins, either CPU optimized plugins or GPU-accelerated plugins based on VAAPI
-
Deep Learning models converted from training frameworks TensorFlow*, Caffe* etc.
-
The following elements in the Pipeline Framework repository:
Element Description gvaattachroi Adds user-defined regions of interest to perform inference on, instead of full frame. gvaaudiodetect Performs audio event detection using AclNet model. gvaaudiotranscribe Performs audio transcription using OpenVino GenAI Whisper model. gvaclassify Performs object classification. Accepts the ROI as an input and outputs classification results with the ROI metadata. gvadetect Performs object detection on a full-frame or region of interest (ROI) using object detection models such as YOLOv4-v11, MobileNet SSD, Faster-RCNN etc. Outputs the ROI for detected objects. gvafpscounter Measures frames per second across multiple streams in a single process. gvafpsthrottle Throttles the frame rate of a pipeline to a specified FPS value. gvagenai Performs inference with Vision Language Models using OpenVINO™ GenAI, accepts video and text prompt as an input, and outputs text description. It can be used to generate text summarization from video. gvainference Runs deep learning inference on a full-frame or ROI using any model with an RGB or BGR input. gvametaaggregate Aggregates inference results from multiple pipeline branches gvametaconvert Converts the metadata structure to the JSON format. gvametapublish Publishes the JSON metadata to MQTT or Kafka message brokers or files. gvamotiondetect Performs lightweight motion detection on NV12 video frames and emits motion regions of interest (ROIs) as analytics metadata. gvapython Provides a callback to execute user-defined Python functions on every frame. Can be used for metadata conversion, inference post-processing, and other tasks. gvarealsense Provides integration with Intel RealSense cameras, enabling video and depth stream capture for use in GStreamer pipelines. gvatrack Performs object tracking using zero-term, or imageless tracking algorithms. Assigns unique object IDs to the tracked objects. gvawatermark Overlays the metadata on the video frame to visualize the inference results. g3dradarprocess Processes 3D radar data for use in analytics pipelines. g3dlidarparse Parses 3D lidar data for use in analytics pipelines.
For the details on supported platforms, please refer to System Requirements.
For installing Pipeline Framework with the prebuilt binaries or Docker* or to build the binaries from the open source, refer to Intel® DL Streamer Pipeline Framework installation guide.
New in this Release
| Title | High-level description |
|---|---|
| 3D elements (g3dradarprocess, g3dlidarparse) | New 3D plugin support with g3dradarprocess element for radar data processing and g3dlidarparse element for lidar data parsing, enabling 3D analytics pipelines. |
| FPS throttle element (gvafpsthrottle) | New element to throttle the frame rate of a pipeline to a specified FPS value. |
| YOLOv26 model support | Added converters and post-processing for YOLOv26 models, including oriented bounding box (OBB) support and INT8 GPU inference. Added YOLOv26 to supported models in samples. |
| RT-DETR model support | Added RT-DETR support implementation for real-time detection transformer models. |
| HuggingFace ViT classifier support | Added HuggingFace Vision Transformer (ViT) classifier config parser for inference. |
| Batch timeout for inference elements | Added batch-timeout parameter to inference elements, allowing control over batching wait time. |
| VLM Alerts sample | New Python sample for VLM-based alerts with displaying results on produced video. |
| Smart NVR sample | New Python sample for Smart NVR with added custom analytics logic (gvaAnalytics) and custom storage (gvaRecorder) elements. |
| ONVIF Camera Discovery sample | New Python sample demonstrating ONVIF camera discovery and DL Streamer pipeline launcher. |
| Face detection & age classification sample | New Python sample for face detection and age classification using HuggingFace models. |
| Open-vocabulary object detection sample | New Python sample with open-vocabulary prompt for object detection. |
| DL Streamer + DeepStream coexistence sample | New sample demonstrating DL Streamer and DeepStream working in one system. |
| Motion detect sample (Windows) | New sample demonstrating DL Streamer gvamotiondetect functionality |
| RealSense element usage sample | New sample demonstrating gvarealsense element usage. |
Updated in this Release
| Title | High-level description |
|---|---|
| gvawatermark enhancements | Major enhancements to the gvawatermark element: display configuration options (thickness, color index), text background support, inclusive/exclusive label filtering, additional font support, average FPS info overlay, and visual documentation. |
| DLS Optimizer enhancements | Optimizer refactored with multi-stream optimization, cross-stream batching, improved FPS reporting, and device selection improvements. |
| gvametaconvert enhancements | Added reference NTP timestamp from RTCP sender meta extraction to gvametaconvert element |
| ROI object construction enhacement | For existing GstAnalyticsODMtd only, creates GstVideoRegionOfInterestMeta until full GstAnalytics migration |
| Latency tracer multi-source/sink support | Extended latency_tracer to support multiple sources and multiple sinks. |
| Detection anomaly converter | Refactored and enhanced anomaly logic in DetectionAnomalyConverter. |
| FP32 precision in BoxesLabelsConverter | Added FP32 precision support in BoxesLabelsConverter label parsing. |
| Bounding box validation | Added extra validation of bounding boxes to improve robustness. |
| OpenCV tensor compression for all devices | Use OpenCV tensor compression for all inference devices, yielding best performance across CPU/GPU/NPU. |
| Model API refactoring | Moved Model API parser to separate files; added conversion from Ultralytics and HuggingFace metadata to Model API. |
| Python samples overview | Added overview section for Python samples; updated READMEs. |
| Tests | Expanded coverage of functional and unit tests. |
| Windows: GPU inference with D3D11 | Added support for GPU inference on Windows using D3D11. |
| Windows: gvapython support | Added Windows support for gvapython element and gstgva Python bindings. |
| Windows: enhanced build & setup | Enhanced Windows build/setup scripts, added remove script, Visual C++ runtime handling, and JSON output for Windows samples. |
| Windows: CI integration | Enabled Windows tests in GitHub Actions workflow, model downloads on Windows. |
| Library consolidation | Merged gvawatermark3d, gvadeskew, gvamotiondetect, and gvagenai into the gstvideoanalytics library. |
| OpenVINO update | Update to OpenVINO 2026.0.0. |
| NPU driver update | Update to NPU driver version 1.30. |
| RealSense update | Update to Intel RealSense SDK 2.57.5. |
| Model download script improvements | Simplified YOLO model download script, enhanced ... |
2025.2.0
Deep Learning Streamer (DL Streamer) Pipeline Framework Release 2025.2.0
Intel® Deep Learning Streamer (Intel® DL Streamer) Pipeline Framework is a streaming media analytics framework, based on GStreamer* multimedia framework, for creating complex media analytics pipelines. It ensures pipeline interoperability and provides optimized media, and inference operations using Intel® Distribution of OpenVINO™ Toolkit Inference Engine backend, across Intel® architecture, CPU, discrete GPU, integrated GPU and NPU.
The complete solution leverages:
-
Open source GStreamer* framework for pipeline management
-
GStreamer* plugins for input and output such as media files and real-time streaming from camera or network
-
Video decode and encode plugins, either CPU optimized plugins or GPU-accelerated plugins based on VAAPI
-
Deep Learning models converted from training frameworks TensorFlow*, Caffe* etc.
-
The following elements in the Pipeline Framework repository:
Element Description gvaattachroi Adds user-defined regions of interest to perform inference on, instead of full frame. gvaaudiodetect Performs audio event detection using AclNet model. gvaaudiotranscribe Performs audio transcription using OpenVino GenAI Whisper model. gvaclassify Performs object classification. Accepts the ROI as an input and outputs classification results with the ROI metadata. gvadetect Performs object detection on a full-frame or region of interest (ROI) using object detection models such as YOLOv4-v11, MobileNet SSD, Faster-RCNN etc. Outputs the ROI for detected objects. gvafpscounter Measures frames per second across multiple streams in a single process. gvagenai Performs inference with Vision Language Models using OpenVINO™ GenAI, accepts video and text prompt as an input, and outputs text description. It can be used to generate text summarization from video. gvainference Runs deep learning inference on a full-frame or ROI using any model with an RGB or BGR input. gvametaaggregate Aggregates inference results from multiple pipeline branches gvametaconvert Converts the metadata structure to the JSON format. gvametapublish Publishes the JSON metadata to MQTT or Kafka message brokers or files. gvamotiondetect Performs lightweight motion detection on NV12 video frames and emits motion regions of interest (ROIs) as analytics metadata. gvapython Provides a callback to execute user-defined Python functions on every frame. Can be used for metadata conversion, inference post-processing, and other tasks. gvarealsense Provides integration with Intel RealSense cameras, enabling video and depth stream capture for use in GStreamer pipelines. gvatrack Performs object tracking using zero-term, or imageless tracking algorithms. Assigns unique object IDs to the tracked objects. gvawatermark Overlays the metadata on the video frame to visualize the inference results.
For the details on supported platforms, please refer to System Requirements.
For installing Pipeline Framework with the prebuilt binaries or Docker* or to build the binaries from the open source, refer to Intel® DL Streamer Pipeline Framework installation guide.
New in this Release
| Title | High-level description |
|---|---|
| Motion detection (gvamotiondetect) | Performs lightweight motion detection on NV12 video frames and emits motion regions of interest (ROIs) as analytics metadata. |
| Audio transcription (gvaaudiotranscribe) | Transcribes audio content with OpenVino GenAI Whisper model. |
| Gvagenai element added | Performs inference with Vision Language Models using OpenVINO™ GenAI, accepts video and text prompt as an input, and outputs text description. Models supported: MiniCPM-V, Gemma3, Phi-4-multimodal-instruct. |
| Deep SORT | Preview version of Deep SORT tracking algorithm in gvatrack element. |
| gvawatermark element support on GPU | Gvawatermark implementation extended about GPU support (CPU default). |
| Pipeline optimizer support | 1st version of DL Streamer optimizer implementation added allowing end user finding the most FPS optimized pipeline. |
| GstAnalytics metadata support | Enabled GstAnalytics metadata support. |
| OpenVINO custom operations | Add support for OpenVINO custom operations. |
| D3D11 preprocessing enabled | Windows support extended about D3D11 preprocessing implementation. |
| UX, Stability && Performance fixes | • memory management fixes • automatically select pre-process-backend=va-surface-sharing for GPU • adjusting caps negotiations and preproc backend selection • removing deleted element from all shared reference lists. • using OpenCV preproc to convert sparse tensors to contiguous tensors • creation of new VADisplay ctx per each inference instance • remove need for dual va+opencv image pre-processing |
| Intel Core Ultra Panther Lake CPU/GPU support | Readiness for supporting Intel Core Ultra Panther Lake CPU/GPU. |
| OpenVINO update | Update to 2025.3 version. |
| GStreamer update | Update to 1.26.6 version. |
| GPU drivers update | Update to 25.40 version (for Ubuntu24) |
| NPU drivers update | Update to 1.23 version. |
Fixed Issues
| # | Issue Description |
|---|---|
| 1 | Fixed issue with segmentation fault and early exit for testing scenarios with mixed GPU/CPU device combinations. |
| 2 | Updated documentation for latency tracer. |
| 3 | Fixed issue where NPU inference required inefficient CPU color processing. |
| 4 | Fixed memory management for elements: gvawatermark, gvametaconvert, gvaclassify. |
| 5 | Improved model-proc check logic for va backend. |
| 6 | Fixed keypoints metadata processing issue for gvawatermark. |
| 7 | Fixed issue with missed gvarealsense element in dlstreamer image. |
| 8 | Fixed issue for scenario when vacompositor scale-method option didn't take affect. |
| 9 | Fixed documentation bug in the installation guide. |
| 10 | Fixed issue with same name for many python modules used by gvapython. |
| 11 | Fixed issue with draw_face_attributes sample (cpp) on TGL Ubuntu 24. |
| 12 | Fixed wrong pose estimation on ARL GPU with yolo11s-pose. |
| 13 | Fixed inconsistent timestamp for vehicle_pedestrian_tracking sample on ARL. |
| 14 | Fixed missing element 'qsvh264dec' in Ubuntu24 docker images. |
Known Issues
| Issue | Issue Description |
|---|---|
| Preview Architecture 2.0 Samples | Preview Arch 2.0 samples have known issues with inference results. |
| Sporadic hang on vehicle_pedestrian_tracking_20_cpu sample | Using Tiger Lake CPU to run this sample may lead to sporadic hang at 99.9% of video processing. Rerun the sample as W/A or use GPU instead. |
Additional Information
System Requirements
Please refer to DL Streamer documentation.
Installation Notes
There are several installation options for Pipeline Framework:
- Install Pipeline Framework from pre-built Debian packages
- Build Docker image from docker file and run Docker image
- Build Pipeline Framework from source code
For more detailed instructions please refer to DL Streamer Pipeline Framework installation guide.
Samples
The samples folder in DL Streamer Pipeline Framework repository contains command line, C++ and Python examples.
Legal Information
Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software, or service activation. Learn more at intel.com, or from the OEM or retailer.
No computer system can be absolutely secure. Intel does not assume any liability for lost or stolen data or systems or any damages resulting from such losses.
You may not use or facilitate the use of this document in connection with any infringement or other legal analysis concerning Intel products described herein. You agree to grant Intel a non-exclusive, royalty-free license to any patent claim thereafter drafted which includes subject matter disclosed herein.
No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document.
Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.
This document contains information on products, services and/or processes in devel...
2025.1.2
Deep Learning Streamer (DL Streamer) Pipeline Framework Release 2025.1.2
Intel® Deep Learning Streamer (Intel® DL Streamer) Pipeline Framework is a streaming media analytics framework, based on GStreamer* multimedia framework, for creating complex media analytics pipelines. It ensures pipeline interoperability and provides optimized media, and inference operations using Intel® Distribution of OpenVINO™ Toolkit Inference Engine backend, across Intel® architecture, CPU, discrete GPU, integrated GPU and NPU.
The complete solution leverages:
-
Open source GStreamer* framework for pipeline management
-
GStreamer* plugins for input and output such as media files and real-time streaming from camera or network
-
Video decode and encode plugins, either CPU optimized plugins or GPU-accelerated plugins based on VAAPI
-
Deep Learning models converted from training frameworks TensorFlow*, Caffe* etc.
-
The following elements in the Pipeline Framework repository:
Element Description gvadetect Performs object detection on a full-frame or region of interest (ROI) using object detection models such as YOLOv4-v11, MobileNet SSD, Faster-RCNN etc. Outputs the ROI for detected objects. gvaclassify Performs object classification. Accepts the ROI as an input and outputs classification results with the ROI metadata. gvainference Runs deep learning inference on a full-frame or ROI using any model with an RGB or BGR input. gvatrack Performs object tracking using zero-term, or imageless tracking algorithms. Assigns unique object IDs to the tracked objects. gvaaudiodetect Performs audio event detection using AclNet model. gvagenai Performs inference with Vision Language Models using OpenVINO™ GenAI, accepts video and text prompt as an input, and outputs text description. It can be used to generate text summarization from video. gvaattachroi Adds user-defined regions of interest to perform inference on, instead of full frame. gvafpscounter Measures frames per second across multiple streams in a single process. gvametaaggregate Aggregates inference results from multiple pipeline branches gvametaconvert Converts the metadata structure to the JSON format. gvametapublish Publishes the JSON metadata to MQTT or Kafka message brokers or files. gvapython Provides a callback to execute user-defined Python functions on every frame. Can be used for metadata conversion, inference post-processing, and other tasks. gvarealsense Provides integration with Intel RealSense cameras, enabling video and depth stream capture for use in GStreamer pipelines. gvawatermark Overlays the metadata on the video frame to visualize the inference results.
For the details on supported platforms, please refer to System Requirements.
For installing Pipeline Framework with the prebuilt binaries or Docker* or to build the binaries from the open source, refer to Intel® DL Streamer Pipeline Framework installation guide.
New in this Release
| Title | High-level description |
|---|---|
| Custom model post-processing | End user can now create a custom post-processing library (.so); sample added as reference. |
| Latency mode support | Default scheduling policy for DL Streamer is throughput. With this change user can add scheduling-policy=latency for scenarios that prioritize latency requirements over throughput. |
| Visual Embeddings enabled | New models enabled to convert input video into feature embeddings, validated with Clip-ViT-Base-B16/Clip-ViT-Base-B32 models; sample added as reference. |
| VLM models support | new gstgenai element added to convert video into text (with VLM models), validated with miniCPM2.6, available in advanced installation option when building from sources; sample added as reference. |
| INT8 automatic quantization support for Yolo models | Performance improvement, automatic INT8 quantization for Yolo models |
| MS Windows 11 support | Native support for Windows 11 |
| New Linux distribution (Azure Linux derivative) | New distribution added, DL Streamer can be now installed on Edge Microvisor Toolkit. |
| License plate recognition use case support | Added support for models that allow to recognize license plates; sample added as reference. |
| Deep Scenario model support | Commercial 3D model support |
| Anomaly model support | Added support for anomaly model, sample added as reference, sample added as reference. |
| RealSense element support | New gvarealsense element implementation providing basic integration with Intel RealSense cameras, enabling video and depth stream capture for use in GStreamer pipelines. |
| OpenVINO 2025.3 version support | Support of recent OpenVINO version added. |
| GStreamer 1.26.6 version support | Support of recent GStreamer version added. |
| NPU 1.19 version driver support | Support of recent NPU driver version added. |
| Docker image size reduction | Reduction for all images, e.g., Ubuntu 24 Release image size reduced to 1.6GB from 2.6GB |
Known Issues
| Issue | Issue Description |
|---|---|
VAAPI memory with decodebin |
If you are using decodebin in conjunction with vaapi-surface-sharing preprocessing backend you should set caps filter using ""video/x-raw(memory:VASurface)"" after decodebin to avoid issues with pipeline initialization |
Artifacts on sycl_meta_overlay |
Running inference results visualization on GPU via sycl_meta_overlay may produce some partially drawn bounding boxes and labels |
| Preview Architecture 2.0 Samples | Preview Arch 2.0 samples have known issues with inference results. |
Sporadic hang on vehicle_pedestrian_tracking_20_cpu sample |
Using Tiger Lake CPU to run this sample may lead to sporadic hang at 99.9% of video processing. Rerun the sample as W/A or use GPU instead. |
| Simplified installation process for option 2 via script | In certain configurations, users may encounter visible errors |
| Error when using legacy YoloV5 models: Dynamic resize: Model width dimension shall be static | To avoid the issue, modify samples/download_public_models.sh by inserting the following snippet at lines 273 and 280: |
| python3 - <<EOF ""${MODEL_NAME}"" import sys, os from openvino.runtime import Core from openvino.runtime import save_model model_name = sys.argv[1] core = Core() os.rename(f""{model_name}_openvino_model"", f""{model_name}_openvino_modelD"") model = core.read_model(f""{model_name}_openvino_modelD/{model_name}.xml"") model.reshape([-1, 3, 640, 640]) |
Additional Information
System Requirements
Please refer to DL Streamer documentation.
Installation Notes
There are several installation options for Pipeline Framework:
- Install Pipeline Framework from pre-built Debian packages
- Build Docker image from docker file and run Docker image
- Build Pipeline Framework from source code
For more detailed instructions please refer to DL Streamer Pipeline Framework installation guide.
Samples
The samples folder in DL Streamer Pipeline Framework repository contains command line, C++ and Python examples.
Legal Information
Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software, or service activation. Learn more at intel.com, or from the OEM or retailer.
No computer system can be absolutely secure. Intel does not assume any liability for lost or stolen data or systems or any damages resulting from such losses.
You may not use or facilitate the use of this document in connection with any infringement or other legal analysis concerning Intel products described herein. You agree to grant Intel a non-exclusive, royalty-free license to any patent claim thereafter drafted which includes subject matter disclosed herein.
No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document.
Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.
This document contains information on products, services and/or processes in development. All information provided here is subject to change without notice. Contact your Intel representative to obtain the latest forecast, schedule, specifications and roadmaps.
The products and services described may contain defects or errors which may cause deviations from published specifications. Current characterized errata are available on requ...
2025.0.1.3
Intel® Deep Learning Streamer Pipeline Framework Release 2025.0.1.3
Intel® Deep Learning Streamer (Intel® DL Streamer) Pipeline Framework is a streaming media analytics framework, based on GStreamer* multimedia framework, for creating complex media analytics pipelines. It ensures pipeline interoperability and provides optimized media, and inference operations using Intel® Distribution of OpenVINO™ Toolkit Inference Engine backend, across Intel® architecture, CPU, discrete GPU, integrated GPU and NPU.
This release includes Intel® DL Streamer Pipeline Framework elements to enable video and audio analytics capabilities, (e.g., object detection, classification, audio event detection), and other elements to build end-to-end optimized pipeline in GStreamer* framework.
The complete solution leverages:
- Open source GStreamer* framework for pipeline management
- GStreamer* plugins for input and output such as media files and real-time streaming from camera or network
- Video decode and encode plugins, either CPU optimized plugins or GPU-accelerated plugins based on VAAPI
- Deep Learning models converted from training frameworks TensorFlow*, Caffe* etc. from Open Model Zoo (OMZ)
- The following elements in the Pipeline Framework repository:
| Element | Description |
|---|---|
| gvadetect | Performs object detection on a full-frame or region of interest (ROI) using object detection models such as YOLOv4-v11, MobileNet SSD, Faster-RCNN etc. Outputs the ROI for detected objects. |
| gvaclassify | Performs object classification. Accepts the ROI as an input and outputs classification results with the ROI metadata. |
| gvainference | Runs deep learning inference on a full-frame or ROI using any model with an RGB or BGR input. |
| gvatrack | Performs object tracking using zero-term, or imageless tracking algorithms. Assigns unique object IDs to the tracked objects. |
| gvaaudiodetect | Performs audio event detection using AclNet model. |
| gvaattachroi | Adds user-defined regions of interest to perform inference on, instead of full frame. |
| gvafpscounter | Measures frames per second across multiple streams in a single process. |
| gvametaaggregate | Aggregates inference results from multiple pipeline branches |
| gvametaconvert | Converts the metadata structure to the JSON format. |
| gvametapublish | Publishes the JSON metadata to MQTT or Kafka message brokers or files. |
| gvapython | Provides a callback to execute user-defined Python functions on every frame. Can be used for metadata conversion, inference post-processing, and other tasks. |
| gvawatermark | Overlays the metadata on the video frame to visualize the inference results. |
For the details of supported platforms, please refer to System Requirements section.
For installing Pipeline Framework with the prebuilt binaries or Docker* or to build the binaries from the open source, please refer to Intel® DL Streamer Pipeline Framework installation guide
New in this Release
| Title | High-level description |
|---|---|
| Installation process | Enhanced installation scripts for the 'installation on host' option |
| Post installation steps | Added a selection option for the YOLO model and device to the hello_dlstreamer.sh script |
| Download models | Improved download_public_models.sh script |
| Documentation updates | Improved installation processes descriptions and tutorial refresh |
Known Issues
| Issue | Issue Description |
|---|---|
VAAPI memory with decodebin |
If you are using decodebin in conjunction with vaapi-surface-sharing preprocessing backend you should set caps filter using "video/x-raw(memory:VASurface)" after decodebin to avoid issues with pipeline initialization |
Artifacts on sycl_meta_overlay |
Running inference results visualization on GPU via sycl_meta_overlay may produce some partially drawn bounding boxes and labels |
| Preview Architecture 2.0 Samples | Preview Arch 2.0 samples have known issues with inference results |
| Sporadic hang on vehicle_pedestrian_tracking_20_cpu sample | Using Tiger Lake CPU to run this sample may lead to sporadic hang at 99.9% of video processing, rerun the sample as W/A or use GPU instead |
System Requirements
Please refer to Intel® DL Streamer documentation.
Installation Notes
There are several installation options for Pipeline Framework:
- Install Pipeline Framework from pre-built Debian packages
- Build Docker image from docker file and run Docker image
- Build Pipeline Framework from source code
For more detailed instructions please refer to Intel® DL Streamer Pipeline Framework installation guide.
Samples
The samples folder in Intel® DL Streamer Pipeline Framework repository contains command line, C++ and Python examples.
Legal Information
Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software, or service activation. Learn more at intel.com, or from the OEM or retailer.
No computer system can be absolutely secure. Intel does not assume any liability for lost or stolen data or systems or any damages resulting from such losses.
You may not use or facilitate the use of this document in connection with any infringement or other legal analysis concerning Intel products described herein. You agree to grant Intel a non-exclusive, royalty-free license to any patent claim thereafter drafted which includes subject matter disclosed herein.
No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document.
Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.
This document contains information on products, services and/or processes in development. All information provided here is subject to change without notice. Contact your Intel representative to obtain the latest forecast, schedule, specifications and roadmaps.
The products and services described may contain defects or errors which may cause deviations from published specifications. Current characterized errata are available on request.
Intel, the Intel logo, and Xeon are trademarks of Intel Corporation in the U.S. and/or other countries.
FFmpeg is an open source project licensed under LGPL and GPL. See https://www.ffmpeg.org/legal.html. You are solely responsible for determining if your use of FFmpeg requires any additional licenses. Intel is not responsible for obtaining any such licenses, nor liable for any licensing fees due, in connection with your use of FFmpeg.
GStreamer is an open source framework licensed under LGPL. See https://gstreamer.freedesktop.org/documentation/frequently-asked-questions/licensing.html. You are solely responsible for determining if your use of GStreamer requires any additional licenses. Intel is not responsible for obtaining any such licenses, nor liable for any licensing fees due, in connection with your use of GStreamer.
*Other names and brands may be claimed as the property of others.
© 2025 Intel Corporation.
2025.0.1.2
Intel® Deep Learning Streamer Pipeline Framework Release 2025.0.1.2
Intel® Deep Learning Streamer (Intel® DL Streamer) Pipeline Framework is a streaming media analytics framework, based on GStreamer* multimedia framework, for creating complex media analytics pipelines. It ensures pipeline interoperability and provides optimized media, and inference operations using Intel® Distribution of OpenVINO™ Toolkit Inference Engine backend, across Intel® architecture, CPU, discrete GPU, integrated GPU and NPU.
This release includes Intel® DL Streamer Pipeline Framework elements to enable video and audio analytics capabilities, (e.g., object detection, classification, audio event detection), and other elements to build end-to-end optimized pipeline in GStreamer* framework.
The complete solution leverages:
- Open source GStreamer* framework for pipeline management
- GStreamer* plugins for input and output such as media files and real-time streaming from camera or network
- Video decode and encode plugins, either CPU optimized plugins or GPU-accelerated plugins based on VAAPI
- Deep Learning models converted from training frameworks TensorFlow*, Caffe* etc. from Open Model Zoo (OMZ)
- The following elements in the Pipeline Framework repository:
| Element | Description |
|---|---|
| gvadetect | Performs object detection on a full-frame or region of interest (ROI) using object detection models such as YOLOv4-v11, MobileNet SSD, Faster-RCNN etc. Outputs the ROI for detected objects. |
| gvaclassify | Performs object classification. Accepts the ROI as an input and outputs classification results with the ROI metadata. |
| gvainference | Runs deep learning inference on a full-frame or ROI using any model with an RGB or BGR input. |
| gvatrack | Performs object tracking using zero-term, or imageless tracking algorithms. Assigns unique object IDs to the tracked objects. |
| gvaaudiodetect | Performs audio event detection using AclNet model. |
| gvaattachroi | Adds user-defined regions of interest to perform inference on, instead of full frame. |
| gvafpscounter | Measures frames per second across multiple streams in a single process. |
| gvametaaggregate | Aggregates inference results from multiple pipeline branches |
| gvametaconvert | Converts the metadata structure to the JSON format. |
| gvametapublish | Publishes the JSON metadata to MQTT or Kafka message brokers or files. |
| gvapython | Provides a callback to execute user-defined Python functions on every frame. Can be used for metadata conversion, inference post-processing, and other tasks. |
| gvawatermark | Overlays the metadata on the video frame to visualize the inference results. |
For the details of supported platforms, please refer to System Requirements section.
For installing Pipeline Framework with the prebuilt binaries or Docker* or to build the binaries from the open source, please refer to Intel® DL Streamer Pipeline Framework installation guide
New in this Release
| Title | High-level description |
|---|---|
| Geti Models 2.7 version | Support for Geti Classification/Detection Models in 2.7 version |
| GStreamer plugins | Support for gst-rswebrtc-plugins |
| Documentation updates | Documentation updates - "queue" element |
Known Issues
| Issue | Issue Description |
|---|---|
VAAPI memory with decodebin |
If you are using decodebin in conjunction with vaapi-surface-sharing preprocessing backend you should set caps filter using "video/x-raw(memory:VASurface)" after decodebin to avoid issues with pipeline initialization |
Artifacts on sycl_meta_overlay |
Running inference results visualization on GPU via sycl_meta_overlay may produce some partially drawn bounding boxes and labels |
| Preview Architecture 2.0 Samples | Preview Arch 2.0 samples have known issues with inference results |
| Sporadic hang on vehicle_pedestrian_tracking_20_cpu sample | Using Tiger Lake CPU to run this sample may lead to sporadic hang at 99.9% of video processing, rerun the sample as W/A or use GPU instead |
| Simplified installation process for option 2 via script | In certain configurations, users may encounter visible errors |
| Error when using legacy YoloV5 models: Dynamic resize: Model width dimension shall be static | To avoid the issue, modify samples/download_public_models.sh by inserting the following snippet at lines 273 and 280: python3 - <<EOF "${MODEL_NAME}" import sys, os from openvino.runtime import Core from openvino.runtime import save_model model_name = sys.argv[1] core = Core() os.rename(f"{model_name}_openvino_model", f"{model_name}_openvino_modelD") model = core.read_model(f"{model_name}_openvino_modelD/{model_name}.xml") model.reshape([-1, 3, 640, 640]) |
System Requirements
Please refer to Intel® DL Streamer documentation.
Installation Notes
There are several installation options for Pipeline Framework:
- Install Pipeline Framework from pre-built Debian packages
- Build Docker image from docker file and run Docker image
- Build Pipeline Framework from source code
For more detailed instructions please refer to Intel® DL Streamer Pipeline Framework installation guide.
Samples
The samples folder in Intel® DL Streamer Pipeline Framework repository contains command line, C++ and Python examples.
Legal Information
Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software, or service activation. Learn more at intel.com, or from the OEM or retailer.
No computer system can be absolutely secure. Intel does not assume any liability for lost or stolen data or systems or any damages resulting from such losses.
You may not use or facilitate the use of this document in connection with any infringement or other legal analysis concerning Intel products described herein. You agree to grant Intel a non-exclusive, royalty-free license to any patent claim thereafter drafted which includes subject matter disclosed herein.
No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document.
Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.
This document contains information on products, services and/or processes in development. All information provided here is subject to change without notice. Contact your Intel representative to obtain the latest forecast, schedule, specifications and roadmaps.
The products and services described may contain defects or errors which may cause deviations from published specifications. Current characterized errata are available on request.
Intel, the Intel logo, and Xeon are trademarks of Intel Corporation in the U.S. and/or other countries.
FFmpeg is an open source project licensed under LGPL and GPL. See https://www.ffmpeg.org/legal.html. You are solely responsible for determining if your use of FFmpeg requires any additional licenses. Intel is not responsible for obtaining any such licenses, nor liable for any licensing fees due, in connection with your use of FFmpeg.
GStreamer is an open source framework licensed under LGPL. See https://gstreamer.freedesktop.org/documentation/frequently-asked-questions/licensing.html. You are solely responsible for determining if your use of GStreamer requires any additional licenses. Intel is not responsible for obtaining any such licenses, nor liable for any licensing fees due, in connection with your use of GStreamer.
*Other names and brands may be claimed as the property of others.
© 2025 Intel Corporation.
2025.0.1
Intel® Deep Learning Streamer Pipeline Framework Release 2025.0.1
Intel® Deep Learning Streamer (Intel® DL Streamer) Pipeline Framework is a streaming media analytics framework, based on GStreamer* multimedia framework, for creating complex media analytics pipelines. It ensures pipeline interoperability and provides optimized media, and inference operations using Intel® Distribution of OpenVINO™ Toolkit Inference Engine backend, across Intel® architecture, CPU, discrete GPU, integrated GPU and NPU.
This release includes Intel® DL Streamer Pipeline Framework elements to enable video and audio analytics capabilities, (e.g., object detection, classification, audio event detection), and other elements to build end-to-end optimized pipeline in GStreamer* framework.
The complete solution leverages:
- Open source GStreamer* framework for pipeline management
- GStreamer* plugins for input and output such as media files and real-time streaming from camera or network
- Video decode and encode plugins, either CPU optimized plugins or GPU-accelerated plugins based on VAAPI
- Deep Learning models converted from training frameworks TensorFlow*, Caffe* etc. from Open Model Zoo (OMZ)
- The following elements in the Pipeline Framework repository:
| Element | Description |
|---|---|
| gvadetect | Performs object detection on a full-frame or region of interest (ROI) using object detection models such as YOLOv4-v11, MobileNet SSD, Faster-RCNN etc. Outputs the ROI for detected objects. |
| gvaclassify | Performs object classification. Accepts the ROI as an input and outputs classification results with the ROI metadata. |
| gvainference | Runs deep learning inference on a full-frame or ROI using any model with an RGB or BGR input. |
| gvatrack | Performs object tracking using zero-term, or imageless tracking algorithms. Assigns unique object IDs to the tracked objects. |
| gvaaudiodetect | Performs audio event detection using AclNet model. |
| gvaattachroi | Adds user-defined regions of interest to perform inference on, instead of full frame. |
| gvafpscounter | Measures frames per second across multiple streams in a single process. |
| gvametaaggregate | Aggregates inference results from multiple pipeline branches |
| gvametaconvert | Converts the metadata structure to the JSON format. |
| gvametapublish | Publishes the JSON metadata to MQTT or Kafka message brokers or files. |
| gvapython | Provides a callback to execute user-defined Python functions on every frame. Can be used for metadata conversion, inference post-processing, and other tasks. |
| gvawatermark | Overlays the metadata on the video frame to visualize the inference results. |
For the details of supported platforms, please refer to System Requirements section.
For installing Pipeline Framework with the prebuilt binaries or Docker* or to build the binaries from the open source, please refer to Intel® DL Streamer Pipeline Framework installation guide
New in this Release
| Title | High-level description |
|---|---|
| LVM support | Support for Large Vision Models |
| LVM support | Sample demonstrating image embedding extraction with Visual Transformer (LVM) |
| OpenVINO 2025.0 support | Update to the latest version of OpenVINO |
| GStreamer 1.24.12 support | Update GStreamer to 1.24.12 version |
| Updated NPU driver | Updated NPU driver to 1.13.0 version. |
| Documentation updates | Documentation how to convert from DeepStream to Deep Learning Steamer |
Known Issues
| Issue | Issue Description |
|---|---|
VAAPI memory with decodebin |
If you are using decodebin in conjunction with vaapi-surface-sharing preprocessing backend you should set caps filter using "video/x-raw(memory:VASurface)" after decodebin to avoid issues with pipeline initialization |
Artifacts on sycl_meta_overlay |
Running inference results visualization on GPU via sycl_meta_overlay may produce some partially drawn bounding boxes and labels |
| Preview Architecture 2.0 Samples | Preview Arch 2.0 samples have known issues with inference results |
| Sporadic hang on vehicle_pedestrian_tracking_20_cpu sample | Using Tiger Lake CPU to run this sample may lead to sporadic hang at 99.9% of video processing, rerun the sample as W/A or use GPU instead |
| Simplified installation process for option 2 via script | In certain configurations, users may encounter visible errors |
| Error when using legacy YoloV5 models: Dynamic resize: Model width dimension shall be static | To avoid the issue, modify samples/download_public_models.sh by inserting the following snippet at lines 273 and 280: python3 - <<EOF "${MODEL_NAME}" import sys, os from openvino.runtime import Core from openvino.runtime import save_model model_name = sys.argv[1] core = Core() os.rename(f"{model_name}_openvino_model", f"{model_name}_openvino_modelD") model = core.read_model(f"{model_name}_openvino_modelD/{model_name}.xml") model.reshape([-1, 3, 640, 640]) |
System Requirements
Please refer to Intel® DL Streamer documentation.
Installation Notes
There are several installation options for Pipeline Framework:
- Install Pipeline Framework from pre-built Debian packages
- Build Docker image from docker file and run Docker image
- Build Pipeline Framework from source code
For more detailed instructions please refer to Intel® DL Streamer Pipeline Framework installation guide.
Samples
The samples folder in Intel® DL Streamer Pipeline Framework repository contains command line, C++ and Python examples.
Legal Information
Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software, or service activation. Learn more at intel.com, or from the OEM or retailer.
No computer system can be absolutely secure. Intel does not assume any liability for lost or stolen data or systems or any damages resulting from such losses.
You may not use or facilitate the use of this document in connection with any infringement or other legal analysis concerning Intel products described herein. You agree to grant Intel a non-exclusive, royalty-free license to any patent claim thereafter drafted which includes subject matter disclosed herein.
No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document.
Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.
This document contains information on products, services and/or processes in development. All information provided here is subject to change without notice. Contact your Intel representative to obtain the latest forecast, schedule, specifications and roadmaps.
The products and services described may contain defects or errors which may cause deviations from published specifications. Current characterized errata are available on request.
Intel, the Intel logo, and Xeon are trademarks of Intel Corporation in the U.S. and/or other countries.
FFmpeg is an open source project licensed under LGPL and GPL. See https://www.ffmpeg.org/legal.html. You are solely responsible for determining if your use of FFmpeg requires any additional licenses. Intel is not responsible for obtaining any such licenses, nor liable for any licensing fees due, in connection with your use of FFmpeg.
GStreamer is an open source framework licensed under LGPL. See https://gstreamer.freedesktop.org/documentation/frequently-asked-questions/licensing.html. You are solely responsible for determining if your use of GStreamer requires any additional licenses. Intel is not responsible for obtaining any such licenses, nor liable for any licensing fees due, in connection with your use of GStreamer.
*Other names and brands may be claimed as the property of others.
© 2025 Intel Corporation.
2025.0.0
Intel® Deep Learning Streamer Pipeline Framework Release 2025.0.0
Intel® Deep Learning Streamer (Intel® DL Streamer) Pipeline Framework is a streaming media analytics framework, based on GStreamer* multimedia framework, for creating complex media analytics pipelines. It ensures pipeline interoperability and provides optimized media, and inference operations using Intel® Distribution of OpenVINO™ Toolkit Inference Engine backend, across Intel® architecture, CPU, discrete GPU, integrated GPU and NPU.
This release includes Intel® DL Streamer Pipeline Framework elements to enable video and audio analytics capabilities, (e.g., object detection, classification, audio event detection), and other elements to build end-to-end optimized pipeline in GStreamer* framework.
The complete solution leverages:
- Open source GStreamer* framework for pipeline management
- GStreamer* plugins for input and output such as media files and real-time streaming from camera or network
- Video decode and encode plugins, either CPU optimized plugins or GPU-accelerated plugins based on VAAPI
- Deep Learning models converted from training frameworks TensorFlow*, Caffe* etc. from Open Model Zoo (OMZ)
- The following elements in the Pipeline Framework repository:
| Element | Description |
|---|---|
| gvadetect | Performs object detection on a full-frame or region of interest (ROI) using object detection models such as YOLOv4-v11, MobileNet SSD, Faster-RCNN etc. Outputs the ROI for detected objects. |
| gvaclassify | Performs object classification. Accepts the ROI as an input and outputs classification results with the ROI metadata. |
| gvainference | Runs deep learning inference on a full-frame or ROI using any model with an RGB or BGR input. |
| gvatrack | Performs object tracking using zero-term, or imageless tracking algorithms. Assigns unique object IDs to the tracked objects. |
| gvaaudiodetect | Performs audio event detection using AclNet model. |
| gvaattachroi | Adds user-defined regions of interest to perform inference on, instead of full frame. |
| gvafpscounter | Measures frames per second across multiple streams in a single process. |
| gvametaaggregate | Aggregates inference results from multiple pipeline branches |
| gvametaconvert | Converts the metadata structure to the JSON format. |
| gvametapublish | Publishes the JSON metadata to MQTT or Kafka message brokers or files. |
| gvapython | Provides a callback to execute user-defined Python functions on every frame. Can be used for metadata conversion, inference post-processing, and other tasks. |
| gvawatermark | Overlays the metadata on the video frame to visualize the inference results. |
For the details of supported platforms, please refer to System Requirements section.
For installing Pipeline Framework with the prebuilt binaries or Docker* or to build the binaries from the open source, please refer to Intel® DL Streamer Pipeline Framework installation guide
New in this Release
| Title | High-level description |
|---|---|
| Enhanced support of Intel® Core™ Ultra Processors (Series 2) (formerly codenamed Lunar Lake); enabled va-surface-sharing pre-process backend. | Validated with Ubuntu 24.04, 6.12.3-061203-generic and the latest Intel® Graphics Compute Runtime for oneAPI Level Zero and OpenCL™ Driver v24.52.32224.5 |
| [preview] Enabled Intel® Arc™ B-Series Graphics [products formerly Battlemage] | Validated with Ubuntu 24.04, 6.12.3-061203-generic and the latest Intel® Graphics Compute Runtime for oneAPI Level Zero and OpenCL™ Driver v24.52.32224.5 + the latest public Intel Graphics Media Driver version + pre-rerelease Intel® Graphics Memory Management Library version |
| OpenVINO 2024.6 support | Update to the latest version of OpenVINO |
| Updated NPU driver | Updated NPU driver to 1.10.1 version. |
| Bug fixing | Running multiple gstreamer pipeline objects in the same process on dGPU leads to error; DLStreamer docker image build is failing (2024.2.2 and 2024.3.0 versions); Fixed installation scripts: minor fixes of GPU, NPU installation section; Updated documentation: cleanup, added missed parts, added DLS system requirements |
Known Issues
| Issue | Issue Description |
|---|---|
VAAPI memory with decodebin |
If you are using decodebin in conjunction with vaapi-surface-sharing preprocessing backend you should set caps filter using "video/x-raw(memory:VASurface)" after decodebin to avoid issues with pipeline initialization |
Artifacts on sycl_meta_overlay |
Running inference results visualization on GPU via sycl_meta_overlay may produce some partially drawn bounding boxes and labels |
| Preview Architecture 2.0 Samples | Preview Arch 2.0 samples have known issues with inference results |
| Sporadic hang on vehicle_pedestrian_tracking_20_cpu sample | Using Tiger Lake CPU to run this sample may lead to sporadic hang at 99.9% of video processing, rerun the sample as W/A or use GPU instead |
System Requirements
Please refer to Intel® DL Streamer documentation.
Installation Notes
There are several installation options for Pipeline Framework:
- Install Pipeline Framework from pre-built Debian packages
- Build Docker image from docker file and run Docker image
- Build Pipeline Framework from source code
For more detailed instructions please refer to Intel® DL Streamer Pipeline Framework installation guide.
Samples
The samples folder in Intel® DL Streamer Pipeline Framework repository contains command line, C++ and Python examples.
Legal Information
Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software, or service activation. Learn more at intel.com, or from the OEM or retailer.
No computer system can be absolutely secure. Intel does not assume any liability for lost or stolen data or systems or any damages resulting from such losses.
You may not use or facilitate the use of this document in connection with any infringement or other legal analysis concerning Intel products described herein. You agree to grant Intel a non-exclusive, royalty-free license to any patent claim thereafter drafted which includes subject matter disclosed herein.
No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document.
Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.
This document contains information on products, services and/or processes in development. All information provided here is subject to change without notice. Contact your Intel representative to obtain the latest forecast, schedule, specifications and roadmaps.
The products and services described may contain defects or errors which may cause deviations from published specifications. Current characterized errata are available on request.
Intel, the Intel logo, and Xeon are trademarks of Intel Corporation in the U.S. and/or other countries.
FFmpeg is an open source project licensed under LGPL and GPL. See https://www.ffmpeg.org/legal.html. You are solely responsible for determining if your use of FFmpeg requires any additional licenses. Intel is not responsible for obtaining any such licenses, nor liable for any licensing fees due, in connection with your use of FFmpeg.
GStreamer is an open source framework licensed under LGPL. See https://gstreamer.freedesktop.org/documentation/frequently-asked-questions/licensing.html. You are solely responsible for determining if your use of GStreamer requires any additional licenses. Intel is not responsible for obtaining any such licenses, nor liable for any licensing fees due, in connection with your use of GStreamer.
*Other names and brands may be claimed as the property of others.
© 2025 Intel Corporation.
2024.3.0
Intel® Deep Learning Streamer Pipeline Framework Release 2024.3.0
Intel® Deep Learning Streamer (Intel® DL Streamer) Pipeline Framework is a streaming media analytics framework, based on GStreamer* multimedia framework, for creating complex media analytics pipelines. It ensures pipeline interoperability and provides optimized media, and inference operations using Intel® Distribution of OpenVINO™ Toolkit Inference Engine backend, across Intel® architecture, CPU, discrete GPU, integrated GPU and NPU.
This release includes Intel® DL Streamer Pipeline Framework elements to enable video and audio analytics capabilities, (e.g., object detection, classification, audio event detection), and other elements to build end-to-end optimized pipeline in GStreamer* framework.
The complete solution leverages:
- Open source GStreamer* framework for pipeline management
- GStreamer* plugins for input and output such as media files and real-time streaming from camera or network
- Video decode and encode plugins, either CPU optimized plugins or GPU-accelerated plugins based on VAAPI
- Deep Learning models converted from training frameworks TensorFlow*, Caffe* etc. from Open Model Zoo (OMZ)
- The following elements in the Pipeline Framework repository:
| Element | Description |
|---|---|
| gvadetect | Performs object detection on a full-frame or region of interest (ROI) using object detection models such as YOLOv4-v11, MobileNet SSD, Faster-RCNN etc. Outputs the ROI for detected objects. |
| gvaclassify | Performs object classification. Accepts the ROI as an input and outputs classification results with the ROI metadata. |
| gvainference | Runs deep learning inference on a full-frame or ROI using any model with an RGB or BGR input. |
| gvaaudiodetect | Performs audio event detection using AclNet model. |
| gvatrack | Performs object tracking using zero-term, or imageless tracking algorithms. Assigns unique object IDs to the tracked objects. |
| gvametaaggregate | Aggregates inference results from multiple pipeline branches |
| gvametaconvert | Converts the metadata structure to the JSON format. |
| gvametapublish | Publishes the JSON metadata to MQTT or Kafka message brokers or files. |
| gvapython | Provides a callback to execute user-defined Python functions on every frame. Can be used for metadata conversion, inference post-processing, and other tasks. |
| gvawatermark | Overlays the metadata on the video frame to visualize the inference results. |
| gvafpscounter | Measures frames per second across multiple streams in a single process |
For the details of supported platforms, please refer to System Requirements section.
For installing Pipeline Framework with the prebuilt binaries or Docker* or to build the binaries from the open source, please refer to Intel® DL Streamer Pipeline Framework installation guide
New in this Release
| Title | High-level description |
|---|---|
| GStreamer 1.24.10 | Updated GStreamer to the 1.24.10 version |
| Documentation for MQTT | Documentation for MQTT updated |
| Added support for numactl | Added support for numactl in the docker image |
| Enabled Intel® Core™ Ultra Processors (Series 2) (formerly codenamed Lunar Lake) | Validated with Ubuntu 24.04, 6.12.3-061203-generic |
Known Issues
| Issue | Issue Description |
|---|---|
VAAPI memory with decodebin |
If you are using decodebin in conjunction with vaapi-surface-sharing preprocessing backend you should set caps filter using "video/x-raw(memory:VASurface)" after decodebin to avoid issues with pipeline initialization |
Artifacts on sycl_meta_overlay |
Running inference results visualization on GPU via sycl_meta_overlay may produce some partially drawn bounding boxes and labels |
| Preview Architecture 2.0 Samples | Preview Arch 2.0 samples have known issues with inference results |
| Sporadic hang on vehicle_pedestrian_tracking_20_cpu sample | Using Tiger Lake CPU to run this sample may lead to sporadic hang at 99.9% of video processing, rerun the sample as W/A or use GPU instead |
| Simplified installation process for option#2 via script | In certain configurations, users may encounter visible errors |
System Requirements
Please refer to Intel® DL Streamer documentation.
Installation Notes
There are several installation options for Pipeline Framework:
- Install Pipeline Framework from pre-built Debian packages
- Build Docker image from docker file and run Docker image
- Build Pipeline Framework from source code
For more detailed instructions please refer to Intel® DL Streamer Pipeline Framework installation guide.
Samples
The samples folder in Intel® DL Streamer Pipeline Framework repository contains command line, C++ and Python examples.
Legal Information
Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software, or service activation. Learn more at intel.com, or from the OEM or retailer.
No computer system can be absolutely secure. Intel does not assume any liability for lost or stolen data or systems or any damages resulting from such losses.
You may not use or facilitate the use of this document in connection with any infringement or other legal analysis concerning Intel products described herein. You agree to grant Intel a non-exclusive, royalty-free license to any patent claim thereafter drafted which includes subject matter disclosed herein.
No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document.
Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.
This document contains information on products, services and/or processes in development. All information provided here is subject to change without notice. Contact your Intel representative to obtain the latest forecast, schedule, specifications and roadmaps.
The products and services described may contain defects or errors which may cause deviations from published specifications. Current characterized errata are available on request.
Intel, the Intel logo, and Xeon are trademarks of Intel Corporation in the U.S. and/or other countries.
FFmpeg is an open source project licensed under LGPL and GPL. See https://www.ffmpeg.org/legal.html. You are solely responsible for determining if your use of FFmpeg requires any additional licenses. Intel is not responsible for obtaining any such licenses, nor liable for any licensing fees due, in connection with your use of FFmpeg.
GStreamer is an open source framework licensed under LGPL. See https://gstreamer.freedesktop.org/documentation/frequently-asked-questions/licensing.html. You are solely responsible for determining if your use of GStreamer requires any additional licenses. Intel is not responsible for obtaining any such licenses, nor liable for any licensing fees due, in connection with your use of GStreamer.
*Other names and brands may be claimed as the property of others.
© 2024 Intel Corporation.
2024.2.2
Intel® Deep Learning Streamer Pipeline Framework Release 2024.2.2
Intel® Deep Learning Streamer (Intel® DL Streamer) Pipeline Framework is a streaming media analytics framework, based on GStreamer* multimedia framework, for creating complex media analytics pipelines. It ensures pipeline interoperability and provides optimized media, and inference operations using Intel® Distribution of OpenVINO™ Toolkit Inference Engine backend, across Intel® architecture, CPU, discrete GPU, integrated GPU and NPU.
This release includes Intel® DL Streamer Pipeline Framework elements to enable video and audio analytics capabilities, (e.g., object detection, classification, audio event detection), and other elements to build end-to-end optimized pipeline in GStreamer* framework.
The complete solution leverages:
- Open source GStreamer* framework for pipeline management
- GStreamer* plugins for input and output such as media files and real-time streaming from camera or network
- Video decode and encode plugins, either CPU optimized plugins or GPU-accelerated plugins based on VAAPI
- Deep Learning models converted from training frameworks TensorFlow*, Caffe* etc. from Open Model Zoo (OMZ)
- The following elements in the Pipeline Framework repository:
| Element | Description |
|---|---|
| gvadetect | Performs object detection on a full-frame or region of interest (ROI) using object detection models such as YOLOv4-v11, MobileNet SSD, Faster-RCNN etc. Outputs the ROI for detected objects. |
| gvaclassify | Performs object classification. Accepts the ROI as an input and outputs classification results with the ROI metadata. |
| gvainference | Runs deep learning inference on a full-frame or ROI using any model with an RGB or BGR input. |
| gvaaudiodetect | Performs audio event detection using AclNet model. |
| gvatrack | Performs object tracking using zero-term, or imageless tracking algorithms. Assigns unique object IDs to the tracked objects. |
| gvametaaggregate | Aggregates inference results from multiple pipeline branches |
| gvametaconvert | Converts the metadata structure to the JSON format. |
| gvametapublish | Publishes the JSON metadata to MQTT or Kafka message brokers or files. |
| gvapython | Provides a callback to execute user-defined Python functions on every frame. Can be used for metadata conversion, inference post-processing, and other tasks. |
| gvawatermark | Overlays the metadata on the video frame to visualize the inference results. |
| gvafpscounter | Measures frames per second across multiple streams in a single process |
For the details of supported platforms, please refer to System Requirements section.
For installing Pipeline Framework with the prebuilt binaries or Docker* or to build the binaries from the open source, please refer to Intel® DL Streamer Pipeline Framework installation guide
New in this Release
| Title | High-level description |
|---|---|
| Installation of Intel® DL Streamer Pipeline Framework from Debian packages using APT repository | Support for apt-get install has been added. |
| Yolo11s-pose support | Added support for Yolo11s-pose model. |
| Change in gvafpscounter element | Reset FPS counters whenever a stream is added/removed. |
| OpenVINO updated | OpenVINO updated to the 2024.5 version. |
| GStreamer 1.24.9 | Updated GStreamer to the 1.24.9 version. |
| NPU 1.10.0 | NPU drivers updated to NPU 1.10.0 version. |
| Bugs fixing | Fixed issue with failing performance tests ; Fixed fuzzy tests ; Enabled debug mode ; Created TLS configuration that allows for secure communication between DL Streamer and MQTT broker; Fixed python error: init_threadstate: thread state already initialized; Fixed problem with DLS compilation / GSTreamer base plugin error.; Fixed issue with sample_test: python_draw_face_attributes on Ubuntu 24.04; Fixed issue with sample_test: gvapython cpu/gpu on Ubuntu 24.04 |
Known Issues
| Issue | Issue Description |
|---|---|
VAAPI memory with decodebin |
If you are using decodebin in conjunction with vaapi-surface-sharing preprocessing backend you should set caps filter using "video/x-raw(memory:VASurface)" after decodebin to avoid issues with pipeline initialization |
Artifacts on sycl_meta_overlay |
Running inference results visualization on GPU via sycl_meta_overlay may produce some partially drawn bounding boxes and labels |
| Preview Architecture 2.0 Samples | Preview Arch 2.0 samples have known issues with inference results |
| Sporadic hang on vehicle_pedestrian_tracking_20_cpu sample | Using Tiger Lake CPU to run this sample may lead to sporadic hang at 99.9% of video processing, rerun the sample as W/A or use GPU instead |
| Simplified installation process for option#2 via script | In certain configurations, users may encounter visible errors |
| Yolov8s model inference does not work on NPU when model converted with OpenVINO 2024.3 and 2024.4 | Model yolov8s converted with OpenVINO Python package in version 2024.3 and later cannot be used to inference on NPU. As a workaround for this model and similar situation, please change the version of installed ‘openvino’ package to 2024.2.0 in download_public_models.sh script (pip install openvino==2024.2.0) |
System Requirements
Please refer to Intel® DL Streamer documentation.
Installation Notes
There are several installation options for Pipeline Framework:
- Install Pipeline Framework from pre-built Debian packages
- Build Docker image from docker file and run Docker image
- Build Pipeline Framework from source code
For more detailed instructions please refer to Intel® DL Streamer Pipeline Framework installation guide.
Samples
The samples folder in Intel® DL Streamer Pipeline Framework repository contains command line, C++ and Python examples.
Legal Information
Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software, or service activation. Learn more at intel.com, or from the OEM or retailer.
No computer system can be absolutely secure. Intel does not assume any liability for lost or stolen data or systems or any damages resulting from such losses.
You may not use or facilitate the use of this document in connection with any infringement or other legal analysis concerning Intel products described herein. You agree to grant Intel a non-exclusive, royalty-free license to any patent claim thereafter drafted which includes subject matter disclosed herein.
No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document.
Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.
This document contains information on products, services and/or processes in development. All information provided here is subject to change without notice. Contact your Intel representative to obtain the latest forecast, schedule, specifications and roadmaps.
The products and services described may contain defects or errors which may cause deviations from published specifications. Current characterized errata are available on request.
Intel, the Intel logo, and Xeon are trademarks of Intel Corporation in the U.S. and/or other countries.
FFmpeg is an open source project licensed under LGPL and GPL. See https://www.ffmpeg.org/legal.html. You are solely responsible for determining if your use of FFmpeg requires any additional licenses. Intel is not responsible for obtaining any such licenses, nor liable for any licensing fees due, in connection with your use of FFmpeg.
GStreamer is an open source framework licensed under LGPL. See https://gstreamer.freedesktop.org/documentation/frequently-asked-questions/licensing.html. You are solely responsible for determining if your use of GStreamer requires any additional licenses. Intel is not responsible for obtaining any such licenses, nor liable for any licensing fees due, in connection with your use of GStreamer.
*Other names and brands may be claimed as the property of others.
© 2024 Intel Corporation.
2024.2.1
Intel® Deep Learning Streamer Pipeline Framework Release 2024.2.1
Intel® Deep Learning Streamer (Intel® DL Streamer) Pipeline Framework is a streaming media analytics framework, based on GStreamer* multimedia framework, for creating complex media analytics pipelines. It ensures pipeline interoperability and provides optimized media, and inference operations using Intel® Distribution of OpenVINO™ Toolkit Inference Engine backend, across Intel® architecture, CPU, discrete GPU, integrated GPU and NPU.
This release includes Intel® DL Streamer Pipeline Framework elements to enable video and audio analytics capabilities, (e.g., object detection, classification, audio event detection), and other elements to build end-to-end optimized pipeline in GStreamer* framework.
The complete solution leverages:
- Open source GStreamer* framework for pipeline management
- GStreamer* plugins for input and output such as media files and real-time streaming from camera or network
- Video decode and encode plugins, either CPU optimized plugins or GPU-accelerated plugins based on VAAPI
- Deep Learning models converted from training frameworks TensorFlow*, Caffe* etc. from Open Model Zoo (OMZ)
- The following elements in the Pipeline Framework repository:
| Element | Description |
|---|---|
| gvadetect | Performs object detection on a full-frame or region of interest (ROI) using object detection models such as YOLOv4, MobileNet SSD, Faster-RCNN etc. Outputs the ROI for detected objects. |
| gvaclassify | Performs object classification. Accepts the ROI as an input and outputs classification results with the ROI metadata. |
| gvainference | Runs deep learning inference on a full-frame or ROI using any model with an RGB or BGR input. |
| gvaaudiodetect | Performs audio event detection using AclNet model. |
| gvatrack | Performs object tracking using zero-term, or imageless tracking algorithms. Assigns unique object IDs to the tracked objects. |
| gvametaaggregate | Aggregates inference results from multiple pipeline branches |
| gvametaconvert | Converts the metadata structure to the JSON format. |
| gvametapublish | Publishes the JSON metadata to MQTT or Kafka message brokers or files. |
| gvapython | Provides a callback to execute user-defined Python functions on every frame. Can be used for metadata conversion, inference post-processing, and other tasks. |
| gvawatermark | Overlays the metadata on the video frame to visualize the inference results. |
| gvafpscounter | Measures frames per second across multiple streams in a single process |
For the details of supported platforms, please refer to System Requirements section.
For installing Pipeline Framework with the prebuilt binaries or Docker* or to build the binaries from the open source, please refer to Intel® DL Streamer Pipeline Framework installation guide
New in this Release
| Title | High-level description |
|---|---|
| Update NPU drivers to version 1.8.0 | Update NPU driver version inside docker images |
| Yolo 11 | Added support for YOLO 11 model (CPU and GPU only) |
| GStreamer | GStreamer updated to the 1.24.8 version |
| Fix Github issue: #440 | gvapython error: Fatal Python error: init_threadstate: thread state already initialized |
Known Issues
| Issue | Issue Description |
|---|---|
VAAPI memory with decodebin |
If you are using decodebin in conjunction with vaapi-surface-sharing preprocessing backend you should set caps filter using "video/x-raw(memory:VASurface)" after decodebin to avoid issues with pipeline initialization |
Artifacts on sycl_meta_overlay |
Running inference results visualization on GPU via sycl_meta_overlay may produce some partially drawn bounding boxes and labels |
| Preview Architecture 2.0 Samples | Preview Arch 2.0 samples have known issues with inference results |
| Sporadic hang on vehicle_pedestrian_tracking_20_cpu sample | Using Tiger Lake CPU to run this sample may lead to sporadic hang at 99.9% of video processing, rerun the sample as W/A or use GPU instead |
| Simplified installation process for option#2 via script | In certain configurations, users may encounter visible errors |
| Yolov8s model inference does not work on NPU when model converted with OpenVINO 2024.3 and 2024.4 | Model yolov8s converted with OpenVINO Python package in version 2024.3 and later cannot be used to inference on NPU. As a workaround for this model and similar situation, please change the version of installed ‘openvino’ package to 2024.2.0 in download_public_models.sh script (pip install openvino==2024.2.0) |
System Requirements
Please refer to Intel® DL Streamer documentation.
Installation Notes
There are several installation options for Pipeline Framework:
- Install Pipeline Framework from pre-built Debian packages
- Build Docker image from docker file and run Docker image
- Build Pipeline Framework from source code
For more detailed instructions please refer to Intel® DL Streamer Pipeline Framework installation guide.
Samples
The samples folder in Intel® DL Streamer Pipeline Framework repository contains command line, C++ and Python examples.
Legal Information
Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software, or service activation. Learn more at intel.com, or from the OEM or retailer.
No computer system can be absolutely secure. Intel does not assume any liability for lost or stolen data or systems or any damages resulting from such losses.
You may not use or facilitate the use of this document in connection with any infringement or other legal analysis concerning Intel products described herein. You agree to grant Intel a non-exclusive, royalty-free license to any patent claim thereafter drafted which includes subject matter disclosed herein.
No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document.
Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.
This document contains information on products, services and/or processes in development. All information provided here is subject to change without notice. Contact your Intel representative to obtain the latest forecast, schedule, specifications and roadmaps.
The products and services described may contain defects or errors which may cause deviations from published specifications. Current characterized errata are available on request.
Intel, the Intel logo, and Xeon are trademarks of Intel Corporation in the U.S. and/or other countries.
FFmpeg is an open source project licensed under LGPL and GPL. See https://www.ffmpeg.org/legal.html. You are solely responsible for determining if your use of FFmpeg requires any additional licenses. Intel is not responsible for obtaining any such licenses, nor liable for any licensing fees due, in connection with your use of FFmpeg.
GStreamer is an open source framework licensed under LGPL. See https://gstreamer.freedesktop.org/documentation/frequently-asked-questions/licensing.html. You are solely responsible for determining if your use of GStreamer requires any additional licenses. Intel is not responsible for obtaining any such licenses, nor liable for any licensing fees due, in connection with your use of GStreamer.
*Other names and brands may be claimed as the property of others.
© 2024 Intel Corporation.