How to classify an image?

[yangxue0827]

I want to classify an image, but the result is not correct. What is wrong with the operation?

input_image = tf.placeholder(tf.float32,  shape=(None, 3, 224, 224), name='input_placeholder')
outputs = SE_ResNet(input_image, 1000, is_training=False, data_format='channels_first')

saver = tf.train.Saver()

with tf.Session() as sess:
    init = tf.global_variables_initializer()
    sess.run(init)

    saver.restore(sess, "./seresnet50/se_resnet50.ckpt")
    img = cv2.imread('./person.jpg')
    img = cv2.resize(img, (224, 224))
    img = np.array(img, np.float32)
    img -= np.array([[[104., 117., 123.]]])
    img = np.transpose(img, (2, 0, 1))

    predict = sess.run(outputs, feed_dict={input_image: np.expand_dims(img, axis=0)})
    print(predict)
print(np.argmax(predict))

@HiKapok

[HiKapok]

@yangxue0827 what do you mean by uncorrect?

[yangxue0827]

The input image is panda, but the network prediction is 168 ('redbone'). Is the above code handled incorrectly? @HiKapok

[HiKapok]

I don't know, I just checked the output of logits to match the Caffe models