Today The Khronos Group announces the ratification and public release of the OpenVX™ 1.3 specification, along with code samples and a prototype conformance test suite. OpenVX is a royalty-free open standard for portable, optimized, and power-efficient vision and machine learning inferencing acceleration, vital to embedded and real-time use cases, such as face-, body-, and gesture-tracking, smart video surveillance, advanced driver assistance systems, object and scene reconstruction, augmented reality, visual inspection, robotics, and more. Also available today is an open source implementation of OpenVX 1.3 for Raspberry Pi to make OpenVX widely accessible to developers. The new specification can be found on the OpenVX registry.
To enable deployment flexibility while avoiding fragmentation, OpenVX 1.3 defines a number of feature sets that are targeted at common embedded use cases. Hardware vendors can include one or more complete feature sets in their implementations to meet the needs of their customers and be fully conformant. The flexibility of OpenVX enables deployment on a diverse range of accelerator architectures, and feature sets are expected to dramatically increase the breadth and diversity of available OpenVX implementations. The defined OpenVX 1.3 feature sets include:
- Graph Infrastructure (baseline for other feature sets),
- Default Vision,
- Enhanced Vision (functions introduced in OpenVX 1.2),
- Neural Network Inferencing (including tensor objects),
- NNEF Kernel import (including tensor objects),
- Binary Images,
- Safety Critical (reduced features to enable easier safety certification).
MulticoreWare has worked with Khronos to provide an OpenVX 1.3 implementation for the Raspberry Pi 3 Model B using the Raspbian operating system. This implementation takes advantage of OpenVX’s flexible architecture to include: automatic optimization of memory access patterns via tiling and chaining; the ability to use highly optimized kernels leveraging multimedia instruction sets; automatic parallelization to utilize multiple compute resources such as multicore CPUs and GPUs; automatic merging of common sequences of processing kernels into single, higher-performance kernels.
The Conformance Test Suite for OpenVX 1.3 is in development and is expected to be released before the end of 2019. Sample implementations of OpenVX 1.3 are available on GitHub for any developer to build upon. The OpenVX 1.3 specification and more information is available on the Khronos website or through the OpenVX registry, which contains specifications of the core API, headers, extensions, and related documentation.
Neil Trevett, President of the Khronos Group and Vice President Developer Ecosystems at NVIDIA, will discuss the launch of OpenVX 1.3 and APIs for vision and inferencing at Embedded Vision on Thursday, October 24 in Stuttgart, Germany in his presentation, APIs for Accelerating Vision and Inferencing: an Industry Overview of Options and Trade-offs.