I am pleased to release ROS2 OpenVINO Toolkit V0.6.0 version.
In this version, some new features are supported:
- Support OpenVINO toolkit 2019R1.1 (see more from Release Notes).
- Support Multiple pipelines. One or more pipelines can be configured and launched in parallel.
- Support Yolo V2 model for object detection.
- Support inference result filtering.
From this version, NCS2 (Intel Neural Compute Stick 2) is directly supported by dldt repo( the opensource version of OpenVINO Toolkit), NCS2 device can be easily used with whichever version of OpenVINO toolkit is adopted as the backbone framework.
Multiple pipeline feature supports one or more pipelines run on the same system in parallel. For example, end user can launch Object Detection pipeline and People re-identification Pipeline at the same time when fulfilling a person-sensitive authentication and semantic scene reconstruction.
Inference result filtering supports to filter inference result in the early stage of the inference process. End user can directly enable filtering feature by editing pipeline configuration files, without changing source code, as long as the corresponding filtering function is implemented already. To demonstrate this feature, vehicle detection (a pipeline of Object Detection ->[Vehicle Attributes Recognition + License Number Recognition] -> Topic Output) was re-implemented in this version.
Meanwhile, source code was refactored to better support different models (say, SSD models and yolov2 model) in parallel for object detection inference.
Please see more from ROS2 OpenVINO Toolkit repo. We are still in progress of backporting these features to ROS version, you can expect to see these features soon in ROS repo.