The Intel® Distribution of OpenVINO™ toolkit – an open suite of tools that software developers can use to create optimized visual-inference and deep-learning applications based on neural networks (NNs) – has won a coveted Vision Product of the Year Award in the “Best Developer Tools” category at the Embedded Vision Summit (EVS), held this month in Santa Clara, California. Jeff Bier, Founder of the Embedded Vision Alliance, said “The Vision Product of the Year Awards recognize the companies that are providing impactful, innovative technologies that system and application developers can rely on as they incorporate visual intelligence into their products.”
The OpenVINO toolkit provides tailored acceleration for deep-learning workloads running on a wide range of Intel® processors. The software toolkit includes an inference engine and model optimizer that enable and optimize heterogeneous, deep-learning processing and asynchronous execution across the many types of Intel processors including CPUs, CPUs with integrated graphics, FPGAs, and Intel® Movidius™ vision processing units (VPUs). A set of simple, unified APIs in the toolkit provide inference applications with unified access to heterogeneous hardware and allows execution of NN layers across these heterogeneous hardware targets.
The OpenVINO toolkit’s model optimizer can import trained models from several popular frameworks, including Caffe, Tensorflow, MxNet, ONNX, and Kaldi. It converts these models into a unified, intermediate-representation format that accelerates performance using several techniques including node merging, horizontal fusion, elimination of batch normalization, and quantization to optimize NN topologies.
The toolkit streamlines deep-learning inference application development and deployment using either standard or custom NN layers without incurring framework overhead. The toolkit also includes more than 30 pre-trained NN models including:
- Face detection (standard and enhanced)
- Face re-identifcation
- Head position
- Human detection (eye-level and high-angle detection)
- People, vehicle, and bike detection
- License-plate detection (small and front-facing)
- Vehicle metadata
- Pedestrian and vehicle detection
- Retail environment
- Person and action detection for smart classroom
- Emotion recognition
- Person identification from diﬀerent videos
- Advanced roadside identification
- Vehicle attributes
- Landmarks regression
- Crossroad object detection
- Semantic segmentation
- Facial landmarks
- Human pose estimation
- Single image super resolution
- Gaze estimation
- Action recognition encoder and decoder
- Text recognition
- Instance segmentation networks
The toolkit provides additional tailored acceleration in the form of optimized OpenCV and OpenVX functions that can be used for the development of traditional computer-vision applications.
To download a Product Brief about the Intel Distribution of OpenVINO toolkit, click here.
To read the EVS press release about the Vision Product of the Year Awards, click here.