Skip to content

2025.1.2

Choose a tag to compare

View all tags
@nszczygl9 nszczygl9 released this 19 Dec 11:28
· 168 commits to master since this release
v2025.1.2
1e78fc8

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:

  1. Install Pipeline Framework from pre-built Debian packages
  2. Build Docker image from docker file and run Docker image
  3. 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 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.

Assets 3
Loading