OpenVX 1.3 Adds Feature Sets for Optimized AI & Vision Processing
October 22, 2019
Khronos has also released an open source implementation of OpenVX 1.3 for the Raspberry Pi 3 Model B, code samples, and conformance test suite.
The Khronos Group has ratified the OpenVX 1.3 specification, a royalty-free open standard for accelerating machine learning inferencing and computer vision workloads. As part of the announcement, Khronos has also released an open source implementation of OpenVX 1.3 for the Raspberry Pi 3 Model B, code samples, and conformance test suite.
OpenVX 1.3 defines multiple feature sets for common use cases, including:
- Graph Infrastructure (baseline for other feature sets)
- Default Vision
- Enhanced Vision
- Neural Network Inferencing (including tensor objects)
- NNEF Kernel import (including tensor objects)
- Binary Images,
- Safety Critical (reduced features to enable easier safety certification)
The new specification can be found on the OpenVX registry.
The OpenVX 1.3 implementation for the Raspberry Pi 3 Model B was ported to the Raspbian operating system by MulticoreWare, and uses tiling and chaining techniques to provide memory optimizations. This allows the kernel to leverage the aforementioned instruction sets, merge threads into higher-order operations, and automatically parallelize them across multicore CPUs and GPUs.
Sample OpenVX 1.3 implementations are available on GitHub, while core API specifications, headers, extensions, and documentation can be found at khronos.org/openvx.