metermodel_predict.py

import tensorflow as tf
import numpy as np
import cv2
import imutils
from PIL import Image
import os
import matplotlib.pyplot as plt



MODEL_NAME = 'meter_model'
PATH_TO_CKPT = MODEL_NAME + '/frozen_meter_model.pb'

detection_graph = tf.Graph()
with detection_graph.as_default():
	od_graph_def = tf.GraphDef()
	with tf.gfile.GFile(PATH_TO_CKPT, 'rb') as fid:
		serialized_graph = fid.read()
		od_graph_def.ParseFromString(serialized_graph)
		tf.import_graph_def(od_graph_def, name='')

	#l = [n.name for n in detection_graph.as_graph_def().node]
	#print(l)
	sess = tf.Session(graph=detection_graph)


def load_image_into_numpy_array(image):
	(im_width, im_height) = image.size
  	return np.array(image.getdata()).reshape(
    	(im_height, im_width, 1)).astype(np.uint8)

def predict(image):

		gray = cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)

		#image = imutils.rotate(gray,90)

		# plt.imshow(image)
		# plt.show()
		im = Image.fromarray(gray)
		#gray = imutils.rotate(gray,-90)
		#im = im.convert('L')
		#im = Image.fromarray(gray)
		
		im = im.resize((96,96))
		im = im.rotate(90)

		

		# plt.imshow(im)
		# plt.show()


		image_np = load_image_into_numpy_array(im)
		image_np_expanded = np.expand_dims(image_np, axis=0)

		x_tensor = detection_graph.get_tensor_by_name('Reshape:0')

		output = detection_graph.get_tensor_by_name('output/dense/BiasAdd:0')

		out =   sess.run([output],feed_dict={x_tensor:image_np_expanded})

		pred = np.argmax(out[0],axis=1)

		if pred[0] == 1:
			return 'digital'
		else:
			return 'analog'


def main():

	files = [f for f in os.listdir('/home/stone/Documents/MeterImages/digital') if f.endswith('.jpg') or f.endswith('.jpeg')]
   	analog_count = 0
   	digital_count = 0
   	for file in files:
		print(file)
		image = cv2.imread('/home/stone/Documents/MeterImages/digital/'+file)

		# plt.imshow(image)
		# plt.show()
		#image = Image.open('/home/stone/Documents/MeterImages/digital/'+file)
		output = predict(image)
		print(output)

		if output == 'analog':
			analog_count += 1
		else:
			digital_count += 1
		print("digital count %d analog count %d" %(digital_count,analog_count))
		accuracy = digital_count/(float(len(files)))
		print("accuracy %f " %(accuracy))


if __name__ == '__main__':
    main()