using tensorflow, got correct prediction

inference using tensorflow

import tensorflow as tf import numpy as np import cv2

x_image = cv2.imread('characters/test_A.jpg', cv2.IMREAD_GRAYSCALE) x_image = cv2.resize(x_image, dsize=(40, 16)) x_image = x_image / 255. x_image = x_image.astype(np.float32)

inputBlob = np.reshape(x_image, [-1, 40, 16, 1])

with open('Model/frozen_modelx.pb', 'rb') as f: out_graph_def = tf.GraphDef() out_graph_def.ParseFromString(f.read()) tf.import_graph_def(out_graph_def, name="")

with tf.Session() as sess:
    data = sess.graph.get_tensor_by_name("input/x_input:0")
    prediction = sess.graph.get_tensor_by_name("softmax/prediction:0")

    sess.run(tf.global_variables_initializer())
    x_image_out = sess.run(prediction, feed_dict={data: inputBlob})

    print(x_image_out)
    print(np.argmax(x_image_out, 1))

[[1.6884116e-13 3.1422300e-07 7.3862933e-28 6.9958590e-16 6.0640429e-07 7.8472853e-12 2.0985551e-07 4.8857016e-17 3.7977627e-19 3.5136532e-21 9.9999881e-01 1.1159207e-21 9.9667357e-32 1.7269920e-31 7.0009942e-21 5.0873650e-25 1.1708389e-13 2.7525333e-24 1.1056422e-19 2.8127509e-08 2.8245915e-14 2.0870798e-23 1.0649909e-11 0.0000000e+00 3.8043613e-11 1.7576119e-15 7.1381208e-14 7.3414647e-08 2.4327422e-20 9.1099174e-17 3.2708306e-19 1.3278245e-14 3.0282866e-22 6.0491925e-24]] [10]

but, using dnn module got wrong prediction

inference using opencv

import numpy as np

import cv2

x_image = cv2.imread('characters/test_A.jpg', cv2.IMREAD_GRAYSCALE)

x_image = cv2.resize(x_image, dsize=(16, 40))

x_image = x_image.astype(np.float32)

x_image = x_image / 255.

read model and inference

inputBlob=cv2.dnn.blobFromImage(x_image)

net = cv2.dnn.readNetFromTensorflow('model/frozen_modelx.pb')

net.setInput(inputBlob)

result = net.forward()

print(result)

print(np.argmax(result, 1))

[[1.3639329e-02 3.8205376e-01 4.1754208e-10 1.0312299e-08 4.3897840e-01 2.1898718e-06 1.0198972e-06 1.3872443e-04 1.2519218e-08 3.5530825e-06 4.7783997e-02 2.8187374e-08 5.5049820e-20 3.5439733e-09 1.4154615e-12 2.4284447e-15 2.9786484e-06 2.5385313e-07 1.6377919e-08 1.9443828e-06 5.2226778e-10 3.8744084e-04 1.8799990e-04 1.3955707e-15 1.1665202e-01 4.2024007e-07 1.0303499e-10 3.1994839e-06 6.9113579e-05 4.2022770e-05 2.7376987e-05 2.3721248e-05 5.4692578e-07 1.5755222e-10]] [4]

Can someone help me to indicate where should be modified! Thank you

using tensorflow, got correct prediction

name="")
with tf.Session() as sess:
 data = sess.graph.get_tensor_by_name("input/x_input:0")
 prediction = sess.graph.get_tensor_by_name("softmax/prediction:0")
 sess.run(tf.global_variables_initializer())
  x_image_out = sess.run(prediction, feed_dict={data: inputBlob})
 print(x_image_out)
  print(np.argmax(x_image_out, 1))

but, using dnn module got wrong prediction

Can someone help me to indicate where should be modified! Thank you

using tensorflow, got correct prediction

# inference using tensorflow
import tensorflow as tf
import numpy as np
import cv2

x_image = cv2.imread('characters/test_A.jpg', cv2.IMREAD_GRAYSCALE)
x_image = cv2.resize(x_image, dsize=(40, 16))
x_image = x_image / 255.
x_image = x_image.astype(np.float32)


inputBlob = np.reshape(x_image, [-1, 40, 16, 1])

with open('Model/frozen_modelx.pb', 'rb') as f:
    out_graph_def = tf.GraphDef()
    out_graph_def.ParseFromString(f.read())
    tf.import_graph_def(out_graph_def, name="")

    with tf.Session() as sess:
        data = sess.graph.get_tensor_by_name("input/x_input:0")
        prediction = sess.graph.get_tensor_by_name("softmax/prediction:0")

        sess.run(tf.global_variables_initializer())
        x_image_out = sess.run(prediction, feed_dict={data: inputBlob})

        print(x_image_out)
        print(np.argmax(x_image_out, 1))


[[1.6884116e-13 3.1422300e-07 7.3862933e-28 6.9958590e-16 6.0640429e-07
  7.8472853e-12 2.0985551e-07 4.8857016e-17 3.7977627e-19 3.5136532e-21
  9.9999881e-01 1.1159207e-21 9.9667357e-32 1.7269920e-31 7.0009942e-21
  5.0873650e-25 1.1708389e-13 2.7525333e-24 1.1056422e-19 2.8127509e-08
  2.8245915e-14 2.0870798e-23 1.0649909e-11 0.0000000e+00 3.8043613e-11
  1.7576119e-15 7.1381208e-14 7.3414647e-08 2.4327422e-20 9.1099174e-17
  3.2708306e-19 1.3278245e-14 3.0282866e-22 6.0491925e-24]]
[10]

but, using dnn module got wrong prediction

# inference using opencv:  
import numpy as np
import cv2

x_image = cv2.imread('characters/test_A.jpg', cv2.IMREAD_GRAYSCALE)
x_image = cv2.resize(x_image, dsize=(16, 40))
x_image = x_image.astype(np.float32)
x_image = x_image / 255.

# read model and inference
inputBlob=cv2.dnn.blobFromImage(x_image)
net = cv2.dnn.readNetFromTensorflow('model/frozen_modelx.pb')
net.setInput(inputBlob)
result = net.forward()

print(result)
print(np.argmax(result, 1))


[[1.3639329e-02 3.8205376e-01 4.1754208e-10 1.0312299e-08 4.3897840e-01
  2.1898718e-06 1.0198972e-06 1.3872443e-04 1.2519218e-08 3.5530825e-06
  4.7783997e-02 2.8187374e-08 5.5049820e-20 3.5439733e-09 1.4154615e-12
  2.4284447e-15 2.9786484e-06 2.5385313e-07 1.6377919e-08 1.9443828e-06
  5.2226778e-10 3.8744084e-04 1.8799990e-04 1.3955707e-15 1.1665202e-01
  4.2024007e-07 1.0303499e-10 3.1994839e-06 6.9113579e-05 4.2022770e-05
  2.7376987e-05 2.3721248e-05 5.4692578e-07 1.5755222e-10]]
[4]

Can someone help me to indicate where should be modified! Thank you