Making an Image Classifier Part 2

If you’re reading this, I assume you have completed part 1 and now have your own, working, retrained Inception model. Now it’s time to build a simplistic GUI to avoid running the label_image.py script through your terminal and so that you can send the program to your less tech literate friends to show off without them having to know how to run a script through their terminal.

Setting up your environment

To start with, you must have a window that has a QPushButton element in it and two QLabel elements. The QLabel elements will allow you to display text and the images that you would like to view, and classify. As for the QPushButton, this will trigger an event that allows you to browse your files and pick an image to classify. So withou further ado, here’s the code:

import sys
from PyQt5.QtGui import *
from PyQt5.QtCore import *
from PyQt5.QtWidgets import *

class Identifier(QMainWindow):
  def __init__(self):
    QMainWindow.__init__(self)
    self.initUI()
  def initUI(self):
    self.line = QLabel(self)
    self.line = QLabel(self)
    self.line.move(5, 550)
    self.line.resize(400, 40)
    self.line.show()

    browse = QPushButton("Browse...", self)
    browse.move(400, 555)
    browse.resize(100, 40)
    browse.clicked.connect(self.openImage)

    self.label = QLabel(self)
    self.label.resize(500, 500)
    self.label.show()

    self.setGeometry(300, 300, 500, 600)
    self.setWindowTitle("Image Identifier")
    self.setFixedSize(500, 600)
    self.show()

def main():
  app = QApplication(sys.argv)
  main = Identifier()
  main.show()
  sys.exit(app.exec_())

if __name__ == "__main__":
  main()

This piece of code is simple enough, the window is initialised, the elements are placed in the correct positions and the window is made into a suitable size. One thing to take into account, is when the browse button is pressed, it connects to a subroutine called openImage. We will create this subroutine in just a second. Outside of the class, all that is happening is the window is being run when the user runs the class. SIMPLE!

Now what we want to do is create the openImage subroutine. This subroutine is responsible for letting the user choose an image and feeding the image to label_image.py as well as loading the image to your window. The first thing we need to do is be able to pick the image, this can be done through the following line of code:

def openImage(self):
  fName = QFileDialog.getOpenFileName(self, 'Open Image', '', 'Image Files (*.jpg *.jpeg)')

This allows the user to only open *.jpg and *.jpeg images from their computer. The variable fName will be a list, however we only want the first item so we then change fName to the first item in the list by writing this line of code:

fName = fName[0]
self.label.setPixmap(QPixmap(filePath).scaled(self.width(), self.height(), Qt.KeepAspectRatio))

The second line in this section loads the image to the top label and scales it so that it adapts to the windows size as well as keeping the images aspect ratio.

Editing label_image.py

Before we go any further, we must edit some parts of the label_image.py script in order to allow us to pass values to our UI script. This is simple enough, first we must place all the code in the label_image.py script into a subroutine which we will call label. We also want to pass the file path so we will mention that when defining the subroutine in the brackets. We will then make the variable imagePath equal to filePath. We will then get rid of the final for loop in the program and replace it with this:

human_string = label_lines[top_k[0]]
score = predictions[0][top_k[0]]
score *= 100
return (human_string, score)

This code takes the top prediction and converts its score to a percentage. It then returns these values so that they can be used in our UI script. Your label_image.py script must now look like this:

import os, sys

import tensorflow as tf
def label(filePath):
    os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

    # change this as you see fit
    image_path = filePath

    # Read in the image_data
    image_data = tf.gfile.FastGFile(image_path, 'rb').read()

    # Loads label file, strips off carriage return
    label_lines = [line.rstrip() for line
                       in tf.gfile.GFile("tensor/retrained_labels.txt")]

    # Unpersists graph from file
    with tf.gfile.FastGFile("tensor/retrained_graph.pb", 'rb') as f:
        graph_def = tf.GraphDef()
        graph_def.ParseFromString(f.read())
        tf.import_graph_def(graph_def, name='')

    with tf.Session() as sess:
        # Feed the image_data as input to the graph and get first prediction
        softmax_tensor = sess.graph.get_tensor_by_name('final_result:0')

        predictions = sess.run(softmax_tensor, \
                 {'DecodeJpeg/contents:0': image_data})

        # Sort to show labels of first prediction in order of confidence
        top_k = predictions[0].argsort()[-len(predictions[0]):][::-1]

        human_string = label_lines[top_k[0]]
        score = predictions[0][top_k[0]]
        score *= 100
        return (human_string, score)

Final Stretch

In order to retrieve the returned values in our UI script, we must first import the label_image. To do this, we must add this line to the top of the program:

from tensor.label_image import *

Then, in the openImage subroutine we must add these lines of code:

human_string, score = label(filePath)
self.line.setText(human_string + " " + str(round(score, 2)) + "%")

The first line simply receives the two variables and then the next line puts the variables into a string to be displayed in the bottom label. This subroutine should now look like this:

def openImage(self):
  fName = QFileDialog.getOpenFileName(self, 'Open Image', '', 'Image Files (*.jpg *.jpeg)')
  filePath = fName[0]
  human_string, score = label(filePath)
  self.line.setText(human_string + " " + str(round(score, 2)) + "%")
  self.label.setPixmap(QPixmap(filePath).scaled(self.width(), self.height(), Qt.KeepAspectRatio))

And that’s it, you now have ultimate bragging rights. I hope you enjoyed this tutorial series and as always, all the files you need are at https://github.com/royabouhamad/TensorFlow-Image-Identifier and you can contact me through any of my above listed cotact methods. Thanks for reading!