These catalog contains sample applications that fulfill specific market needs and examples to optimize, tune and accelerate applications to run efficiently on Intel architectures.
Accelerated Object Detection
Basics
Learn how to accelerate your object detection applications with Asynchronous inference and offloading to multiple types of processing units.
Store Aisle Monitor
Retail
Capture video, generate a heat map, record the number of patrons, then integrate the results. The program can also create output video and save snapshots.
Store Traffic Monitor
Retail
The Store Traffic Monitor builds upon the two previous examples to provide a more advanced use case. This reference sample demonstrates how to perform object detection and inference with multiple videos simultaneously. In this use case, the application monitors the activity of people inside and outside an imaginary storefront and keeps track of product inventory using a pretrained neutral network for detection.
Shopper Gaze Monitor
Retail
Build a solution to analyze customer expressions and reactions to product advertising positioned on retail shelves.
People Counter System
Transportation | Security
Create a smart video application using the Intel® Distribution of OpenVINO™ toolkit. The toolkit uses models and inference to run single class object detection.
Pneumonia Detection
Healthcare
This example showcases a healthcare application by classifying the probability of pneumonia in X-ray images. The application uses the inference engine in the Intel® Distribution of OpenVINO™ toolkit and applies a pretrained neural network using an open source dataset. The inference results are stored in an output file.
Brain Tumor Segmentation
Healthcare
This example implements an inference engine based on the U-Net architecture for brain tumor segmentation in MRI scans. The code demonstrates several approaches. First, a stock TensorFlow implementation is presented. Next, the same implementation is executed with an Intel-optimized TensorFlow backed by MKL-DNN. We also present an alternative implementation that relies on the Intel® Distribution of OpenVINO™ toolkit. This latter implementation allows you to use not only Intel CPUs for inference, but also Intel HD Graphics, Intel Neural Compute Stick 2, HDDL-R, and Intel FPGAs (HDDL-F).
Safety Gear Detection
Industrial | Security
The Safety Gear Detection sample is another demonstration of object detection, this time in an industrial/safety use case. The sample involves presenting a video frame-by-frame to the inference engine (IE), which then uses a trained and optimized neural network – Mobilenet-SSD – to detect people and their safety gear.
Optical Character Recognition
Basics
The Optical Character Recognition (OCR) sample demonstrates the use of the Intel® Distribution of OpenVINO™ toolkit to perform OCR using Long Short-Term Memory (LSTM), which is a Convolutional Recurrent Neural Network (CRNN) architecture for deep learning. The sample recognizes words in a sample JPEG file.
Intruder Detection
Security
The Intruder Detection sample demonstrates how to build an application to alert you when someone enters a restricted area. Here, you will learn how to use models to perform multiclass object detection.
Accelerated Object Detection (C++)
Basics | General
Learn how to accelerate your object detection applications with asynchronous inference and offloading to multiple types of processing units.
New Speech Recognition C++ Sample
General
This sample demonstrates the building of speech recognition application using a Kaldi acoustic and language model to transcribe an audio file. The Intel® Distribution of OpenVINO™ toolkit inference engine (IE), Intel® Speech Decoder, and Intel® Speech Extraction libraries are used to create a simple Speech Library API.
American Sign Language (ASL) Recognition
General
The ASL Recognition sample demonstrates how to use a person detection model to track a person, plus a gesture recognition model to identify up to 100 ASL gestures used by the tracked person. The application uses the Intel® Distribution of OpenVINO™ toolkit inference engine (IE) and renders a bounding box with a text translation of gestures attached to each frame.
Coming soon
Clean Room Worker Safety - ONNX Object Detection Sample
Industrial | Security
The Clean Room Worker Safety sample demonstrates object detection in an industrial/clean room use case. The sample involves presenting a frame-by-frame video to the ONNX Runtime (RT), which uses an ONNX RT Execution Provider for OpenVINO™ toolkit to run inference on various Intel® hardware, such as CPU, iGPU, accelerator cards NCS2, FPGA, and VAD-M. This sample uses a pretrained Tiny Yolo V2 Deep Learning ONNX Model for the detection of safety gear (e.g., bunny suit, safety glasses), robots, and heads, and can be used for hazard detection purposes.
Radar Image Classification
Public Sector/Government
This example showcases the use of Intel® Distribution of OpenVINO™ Toolkit to optimize and deploy a pre-trained ResNet18 model that classifies Synthetic Aperture Radar (SAR) images associated with 10 separate military vehicle classifiers, such as tanks and armored vehicles.
Tiny Yolo V3 Object Detection
General
This sample showcases how to convert a pre-trained DarkNet Tiny YOLO V3 model to TensorFlow format, run accelerated Inference with Tiny YOLO V3 using OpenVINO's Inference engine, and fine tune the number of inference request processed at a time and the number of streams to achieve optimal performance.
Robotic Surgery Segmentation
Healthcare
This sample showcases a Pytorch model based on TeranusNet being converted to ONNX to utilize Intel® Distribution of OpenVINO™ toolkit for accelerated inference to segment robotic tools in an endoscopic surgical video. The PyTorch model won the 2017 MICCAI Robotic Surgery Instrument Segmentation Challenge.
Biomedical Text Mining
Healthcare
This sample showcases the BioBERT transformer-based model designed for biomedical text mining tasks being converted to utilize the Intel® Distribution of OpenVINO™ toolkit and walks through the process of preparing and running with inputs to a BERT model, including tokenization, preparing input masks, and segment ids.
Electrocardiography
Healthcare
This sample showcases a Keras model developed by the Stanford ML group being converted to utilize the Intel® Distribution of OpenVINO™ toolkit to accelerate inference times on streaming electrical activity of the heart gathered from electrodes placed on the skin. The model is used to detect arrhythmias in this PhysioNet 2017 challenge ECG time series data.
Restricted Zone Notifier
Security | Retail
This sample application demonstrates how a smart video IoT solution may be created using Intel® hardware and software tools to perform restricted zone notification. This solution detects any number of people within a video frame and for each person determines whether they have entered into the restricted zone.
Benchmark Sample
Basics | General
The Benchmark sample showcases how you can use AWS S3 bucket to upload deep learning models and download them to a local devcloud instance to benchmark on OpenVINO™ supported hardware.
Seismic Interpretation
Industrial
This sample application uses the Intel® Distribution of OpenVINO™ Toolkit to perform inference on the salt model, based on Convolutional neural networks for automated seismic interpretation. This model runs an analysis on seismic amplitude data to identify salt bodies in the Earth's crust.
Up Next:
Roll up your sleeves with our step-by-step instructions on using computer vision at the edge with these advanced Jupyter* Notebook tutorials.