Object detection server side application sample program written in Java. It uses the TensorFlow Java API with a trained YOLOv2 model. The server application is implemented with Spring Framework and it is built by Gradle.

It provides a web user interface to upload images and detect objects. Please have a look at the following screenshots:

Step1: upload your image

Step2: display the recognized objects

Preconditions:

• Java JDK 1.8 or greater;
• TensorFlow 1.6 or grater;
• Git version control system;

Strongly recommended to install:

• nVidia CUDA Toolkit 8.0 or higher version;
• nVidia cuDNN GPU accelerated deep learning framework;

Download the frozen graph and the label file

Before compiling the source code you have to place the frozen graph and the label file into the ./graph/YOLO directory. Download one of my graphs from my google drive. There are two graphs: tiny-yolo-voc.pb and yolo-voc.pb. The tiny-yolo.pb has a lower size, however it is less accurate than the yolo-voc.pb. Modify the application.yml configuration file if it is necessary. Here you can increase the file upload limit also.

Compile with Gradle

Compile the code by typing ./gradlew clean build in the terminal window.
Run it with the command ./gradlew bootRun

Open the http://localhot:8080 and you should see the webpage.
If you want to understand better how the image recognition part works, have a look at my previous project here: Tensorflow Java API example application

Deployed to Google cloud: http://35.231.120.117:8080/

Deployed to Heroku with a tiny-yolo model: https://still-crag-64816.herokuapp.com/