Result-Set Viewer

Introduction

Machine learning allows computers to learn a model for a given task, such as face recognition, with a high degree of accuracy, using data. However, after these models are generated, they are often treated as black boxes by developers and the limitations of a model are often unknown to end-users. To address these issues this paper introduces the Face Recognition Model Analyzer (FRMA) ontology and a semantically enabled Result-Set Viewer. Together these resources describe image features relevant to face recognition and allow users to explore how well a face recognition model does at classifying images that contain an image feature.

Installing for Linux

Requirements

1. java

   sudo apt update

   sudo apt install openjdk-8-jdk

2. python 2

   sudo apt install python2.7

3. pip2

   sudo apt install python-pip

4. python Tkinter

   sudo apt install python-tk

5. git

   sudo apt install git

6. curl

   sudo apt install curl

7. wget

   sudo apt install wget

Download Code and Data

1. Download Result-Viewer

   git clone https://github.com/FRMA-Ontology/resultset-viewer.git project_folder

2. Download Blazegraph

   curl -L http://sourceforge.net/projects/bigdata/files/bigdata/2.1.4/blazegraph.jar/ > project_folder/lib/blazegraph/blazegraph.jar

3. Download LFW Images

   curl -L http://vis-www.cs.umass.edu/lfw/lfw-deepfunneled.tgz > project_folder/lib/lfw-deepfunneled.tgz

   tar zxvf project_folder/lib/lfw-deepfunneled.tgz --one-top-level=lfw --strip-components 1 -C project_folder/lib

4. Download LFW Features

   wget --load-cookies /tmp/cookies.txt "https://docs.google.com/uc?export=download&confirm=$(wget --quiet --save-cookies /tmp/cookies.txt --keep-session-cookies --no-check-certificate 'https://docs.google.com/uc?export=download&id=1QLpFBM6N4CNZnyeDsDAjOmK2fpcEKCqN' -O- | sed -rn 's/.*confirm=([0-9A-Za-z_]+).*/\1\n/p')&id=1QLpFBM6N4CNZnyeDsDAjOmK2fpcEKCqN" -O lfw_tags.rdf && rm -rf /tmp/cookies.txt

   mv lfw_tags.rdf project_folder/lib/ontologies/lfw_tags.rdf

Configure Enviornment

cd project_folder/

pip2 install -r requirements.txt

Start blazegraph:

cd project_folder/lib/blazegraph

java -server -Xmx4g -Dbigdata.propertyFile=RWStore.properties -jar blazegraph.jar

Start GUI

cd project_folder/

python main.py

Usage

Ontology Load

On first startup the triplestore needs to be intialized by running clicking:

File -> Clear Datasets -> Yes

This will load the ontologies and lfw labels into blazegraph and clear the previous datasets.

Loading Dataset

After that datasets can be loaded in by clicking:

File -> New Dataset

Fill out this dialog box with the ResultSetName, Algorithm, and a ResultSetFile path then click done. Two ResultSet files are included with the viewer project_folder/data/dlib_result.txt and project_folder/data/facenet_result.txt. ResultSet files are tab separated files where each row is one model comparison and is of the following form:

PersonNameImage1  ImageNumber1   PersonNameImage2  ImageNumber2   ClassifiedCorrectly

This dialog box will generate an RDF representation of the results and load it into blazegraph.

Select Dataset

After loading a dataset it can be selected in the viewer by clicking:

File -> Select ResultSet -> Choose ResultSet from drop down -> select done

This will load the ResultSet into the viewer so a user can begin to explore the model. This process can take some time to load in for large dataset, but once loaded performance will improve.