卷积神经网络-如何在同一张图像上使用两个cnn模型

Question

I want to use two differently trained CNN (convolutional neural network) modules on the same image. I trained the two modules, 1 for detection and one for classification. Now, I want to use these two modules on the same image. The code is in python using the keras and tensorflow libraries. Two different CNN on the same image

Answer 1

In tensorflow you need to explicitly specify the computational graph for both of your models.

# build two separate graphs `g1` and `g2`

tf.reset_default_graph()
with tf.Session(graph=g1) as session:
    result = sess.run(detect, feed_dict={x: test_x})
    print(result)

tf.reset_default_graph()
with tf.Session(graph=g2) as session:
    result = sess.run(recognize, feed_dict={x: test_x})
    print(result)

There are also some caveats when multiple graphs are built in one application, see this question .