realtime-image-recognition

Real-time Image Recognition

This demonstration application uses the webcam video stream of the laptop computer as a source and recognizes the images in each frame. The recognition done with a model pre-tranied with CIFAR-10 dataset. The model is read from the filesystem and loaded into the Jet processors which are making the recognition. The recognitions along with recognition scores are aggregated in 1 second windows and frame with the maximum score will be shown in a GUI application.

This application shows how you can implement a custom streaming source and sink and how to make streaming aggregations in Hazelast Jet.

Package Level Structure

The main class (RealTimeImageRecognition) with the main method, webcam data source (WebcamSource) and serialiable version of the buffered image(SerializableBufferedImage) classes resides in the default package.

Prerequisites

• Git Large File Storage: Installation Guide Some of the demo applications includes machine learning models in their use cases. Since some models' size exceeds GitHub's 100MB file storage limit this repository uses Git LFS.

• Java Development Kit 8+: Installation Guide

• Apache Maven: Installation Guide

• To make sure you've installed Git Large File Storage and checked out the repository correctly run the following commands:

$ git lfs pull
$ git lfs ls-files
038574163b - realtime-image-recognition/likevgg_cifar10/model.net

You should see an output like above which shows our model is ready to use.

Building the Application

To build and package the application, run:

mvn clean package

Running the Application

After building the application, run the application with:

mvn exec:java

You should be able to see that your webcam input is taken and after a few seconds, outputted to the second GUI with the recognized tag, like below.

Since this is a demonstration application the model is used here is a model pre-trained with CIFAR-10.

The size of the model is around 100MB and contains just 10 categories.