Zora-Object-Detection

This application is a simulation of server-side object detection into the prototype with Zora robot, using models loaded in the TensorFlow ModelServer.

Prerequisites

Before to use the application, you need to download the following packages:

pip install tensorflow
pip install grpcio
pip install numpy
pip install scipy
pip install object-detection
pip install hdfs
pip install tensorflow-serving-api

To install the TensorFlow ModelServer, you need to run the following commands:

echo "deb [arch=amd64] http://storage.googleapis.com/tensorflow-serving-apt stable tensorflow-model-server tensorflow-model-server-universal" | sudo tee /etc/apt/sources.list.d/tensorflow-serving.list && \
curl https://storage.googleapis.com/tensorflow-serving-apt/tensorflow-serving.release.pub.gpg | sudo apt-key add -

sudo apt-get update
sudo apt-get install tensorflow-model-server

Usage

First of all, you need to start the Model Server with the following command:

tensorflow_model_server --port=9000 --model_config_file='PATH_TO_BE_CONFIGURED/model_server.config'

Change PATH_TO_BE_CONFIGURED with the absolute path in your PC and change also the paths of models to load in the server, into the file model_server.config.

Then start the object detection of a chosen image (for example harry_meghan.jpg into Images_test directory) with the command:

python test.py --image_path=Images_test/harry_meghan.jpg

This script saves the predicted images with bounding boxes into Images_bbx directory and returns the results that have been sent to Zora into the prototype: a string to be pronounced by robot with the labels of detected objects (case of greater certainty), an array with the labels of possible objects (case of uncertainty) and an empty array if there aren't detected objects.

If Hadoop is running, the application saves a log file and the predicted images into HDFS.