Releases: dlstreamer/pipeline-server
v0.3.1.1-alpha
Release Notes (v0.3.1.1)
Video Analytics Serving is a python package and microservice for deploying hardware optimized media analytics pipelines. It supports
pipelines defined in GStreamer* or FFmpeg* media frameworks and provides APIs to discover, start, stop, customize and monitor pipeline execution. Video Analytics Serving is based on DL Streamer and FFmpeg Video Analytics.
New and Changed in Release v0.3.1.1
This is a patch for v0.3.1 that fixes the default image build (issue #25) and addresses some documentation issues.
Known Issues
Known issues can be found as GitHub issues. If you encounter defects in functionality, please submit an issue.
| Description | Issue |
|---|---|
| Pre-built base images do not support audio detection | #31 |
Tested Base Images
All validation is done in docker environment. Host built configurations are not supported.
| Base Image | Framework | Link | Default |
|---|---|---|---|
| DL Streamer Audio Preview 2020.4 | GStreamer | GitHub | Y |
| FFmpeg Video Analytics v4.2 | FFmpeg | GitHub | Y |
| OpenVINO 2020.4 ubuntu18_data_dev | GStreamer | Docker Hub | N |
| Open Visual Cloud 20.4 xeone3-ubuntu1804-analytics-gst | GStreamer | Docker Hub | N |
| Open Visual Cloud 20.4 xeone3-ubuntu1804-analytics-ffmpeg | FFmpeg | Docker Hub | N |
* Other names and brands may be claimed as the property of others.
Release Notes (v0.3.1)
Release Notes (v0.3.1)
Video Analytics Serving is a python package and micro-service for deploying hardware optimized media analytics pipelines. It supports pipelines defined in GStreamer* or FFMpeg* media frameworks and provides APIs to discover, start, stop, customize and monitor pipeline execution. Video Analytics Serving is based on DL Streamer and FFmpeg Video Analytics.
New and Changed in Release v0.3.1
| Title | High-level description |
|---|---|
| Improvements to docker build | 1. Intel(R) Distribution of OpenVINO™ Toolkit Docker image, which includes DL Streamer, can now be used as a base image for a video analytics serving container. Please see the below list of tested base images for the images that can be used as a base image. |
| 2. Provided an ability to use DL Streamer local repository on a developer's system for building a docker image. | |
| Audio event detection pipeline | Added support for audio event detection pipeline with DL Streamer and provided a sample pipeline definition. |
| Debug support | Enabled GStreamer bus message logging via a new parameter in the pipeline definition file. |
| Extended documentation | Added docker documentation with build and run script reference guides. Also added reference documentation for pipeline definition files. |
Known Issues
N/A - If you encounter defects in functionality, please submit as a GitHub issue.
Tested Base Images
All validation is done in docker environment. Running and installing outside of Docker is not tested.
- OpenVINO 2020.4 ubuntu18_data_dev (dockerhub)
- Open Visual Cloud 20.4 xeone3-ubuntu1804-analytics-gst (dockerhub)
- Open Visual Cloud 20.4 xeone3-ubuntu1804-analytics-ffmpeg (dockerhub)
* Other names and brands may be claimed as the property of others.
Release Notes (v0.3-alpha)
Release Notes (v0.3)
What’s New in This Release:
| Title | High-level description |
|---|---|
| Consolidated VA Serving source code to primary repository | VA Serving was originally released as a component embedded into the Intel Ad-Insertion Demo before being released independently. Intel® Smart-Cities and Ad-Insertion both had local copies of VA Serving which were being managed independently with their own set of bug fixes and feature enhancements. With VA Serving 0.3 we have consolidated to a “one Source” Media Analytics Pipeline Serving library. |
| Improved Docker files, Build and Run Scripts | Complete refactor of build stages to cleanly separate:
|
| Container runs as non-root user | Dockerfile creates a non-root privilege user for increased security. |
| Removed dead code and auto-generated files | We initially generated the skeleton for VA Serving using openapi. This led to many files being generated that were not actually used (though we use the specification with a library to define the APIs). |
| VA Serving Core Library | VA Serving now provides a library interface allowing developers to embed pipeline serving capabilities directly into their applications. This allows applications complete freedom in how they receive and process incoming requests and removes the need to use REST APIs. VA Serving still maintains and supports the service layer that uses the VA Serving Core Library underneath. |
| Model Caching between pipeline instances | DL Streamer elements which are given a specific ID (which indicates that they share inference requests between multiple pipelines) are now cached between pipeline invocations to avoid unnecessary start up and shut down costs. |
| FFmpeg v0.5 Parameter Support + Segment Recording | Complete refactor of FFmpeg parameter handling to bring closer in line to the modularity of GStreamer. Parameters can now be specified on specific ‘filters’ as opposed to only simple text substitution. Along with this change we now support recording segments with FFMpeg using normalized timestamps in line with GStreamer segment recording capabilities. |
| Code brought into Pep 8 Compliance | All files and naming were updated to be Pep8 compliant to provide a better base for development. |
| Removed static methods where possible | Following singleton pattern to improve design and provide a better base for development. |
Known Issues
N/A
If you encounter defects in functionality, please submit as a GitHub issue.
Release Notes (v0.2.3-alpha)
Release Notes (v0.2.3)
Important note: This is a pre-production release of Video Analytics Serving. It is intended to bootstrap your efforts but has NOT been vetted for a production setting. It does not meet traditional requirements in terms of uptime, scalability, security, etc.
If you encounter defects in functionality beyond those listed below, please submit as a GitHub issue.
Added Features
N/A
Bug Fixes
This patch resolves build time issues with Video Analytics Serving. Two dependencies of the build were hosted on a site that is no longer available, which caused a timeout when downloading these dependencies. This timeout fails the build.
This patch updates the location where those dependencies are available and builds them accordingly.
Release Notes (v0.2.2-alpha)
Release Notes (v0.2.2)
Important note: This is a pre-production release of Video Analytics Serving. It is intended to bootstrap your efforts but has NOT been vetted for a production setting. It does not meet traditional requirements in terms of uptime, scalability, security, etc.
If you encounter defects in functionality beyond those listed below, please submit as a GitHub issue.
Added Features
N/A
Bug Fixes
This patch resolves a critical runtime issue that impacts the ability to run Video Analytics Serving. A nested dependency of the connexion component has changed its interaction pattern in such a way that it triggers a fatal exit of Video Analytics Serving on startup.
This patch updates the version of connexion to load and is required in order to support expected Video Analytics Serving runtime behavior.
Release Notes (v0.2.1-alpha)
Release Notes (v0.2.1)
Important note: This is a pre-production release of Video Analytics Serving. It is intended to bootstrap your efforts but has NOT been vetted for a production setting. It does not meet traditional requirements in terms of uptime, scalability, security, etc.
If you encounter defects in functionality beyond those listed below, please submit as a GitHub issue.
Added Features
N/A
Bug Fixes
This patch resolves an issue witnessed when using gvametaconvert for injecting timestamps needed to calculate the actual time of each inference.
Release Notes (v0.2.0-alpha)
Release Notes (v0.2)
Important note: This is a pre-production release of Video Analytics Serving. It is intended to bootstrap your efforts but has NOT been vetted for a production setting. It does not meet traditional requirements in terms of uptime, scalability, security, etc.
If you encounter defects in functionality beyond those listed below, please submit as a GitHub issue.
Added features
- Converged options and settings and added environment variable default support.
- port, framework, pipeline_dir, model_dir, network_preference, max_running_pipelines, log_level, config_path can all be set as environment variables or command line arguments when running VAS
- Common handling for source, destination, tags and parameters.
- Default schema for source, destination and tags
- Ability to overwrite any section of the default on a per pipeline basis as needed. This is done in the pipeline.json file
- Separated Dockerfiles into base and application stages.
- The base Dockerfile will prepare the environment needed to run the service.
- The application Dockerfile will copy the sources of VAS that are needed to run into the image.
Known Issues
- Out of the box configuration currently limits one pipeline to run at a time. This will cause other pipelines to remain in QUEUED state until the running pipeline completes.
System Requirements
Hardware
- Refer to Intel OpenVINO SDK for inference plugin requirements.
Operating System
- Ubuntu 16.04/18.04
- Other operating systems that are supported via docker container (Not validated, experimental support).
Release Notes (v0.1.0-alpha)
Release Notes (v0.1.0-alpha)
Important note: This is a pre-production release of Video Analytics Serving. Be advised that there are defects in functionality beyond those listed below.
- This initial release contains Video Analytics Serving components and instructions to build and run examples within a docker container.
- Build Docker image and run the container with fully prepared environment
- This leverages GStreamer or FFmpeg to execute predefined pipelines.
- Current samples include pre-defined pipelines for:
- GStreamer:
- Object Detection
- Emotion Recognition
- FFmpeg:
- Object Detection
- Emotion Recognition
- GStreamer:
- The following models are included as samples in support of the above pipelines:
- Object Detection
- Face Detection
- Face Reidentification
- Landmarks Regression
- Emotion Recognition
Known Issues
- Out of the box configuration currently limits one pipeline to run at a time. This will cause other pipelines to remain in QUEUED state until the running pipeline completes.
- Pipelines that run concurrently across multiple Video Analytics Serving containers and emit inferences to MQTT (through metapublish element) must use distinct/unique values for clientid.
- Pipelines with kafka as the method parameter for metapublish element will fail.
System Requirements
Hardware
- Refer to Intel OpenVINO SDK for inference plugin requirements.
Operating System
- Ubuntu 16.04/18.04
- Other operating systems that are supported via docker container (Not validated, experimental support).