deconv

deconv.py

#  import tensorflow as tf
#  # part1
#  #case 2
#  input = tf.Variable(tf.random_normal([3,97,97,10]))
#  filter = tf.Variable(tf.random_normal([3,3,10,20]))
#  x = tf.nn.conv2d(input, filter, strides=[1, 4, 4, 1], padding='SAME')
#  #  y = tf.nn.conv2d_transpose(x, filter, output_shape=[1,5,5,3], strides=[1,2,2,1],padding="SAME")

#  with tf.Session() as sess:
    #  sess.run(tf.initialize_all_variables())
    #  res = (sess.run(x))
    #  print (res.shape)

# part2
import tensorflow as tf
from tensorflow.contrib import slim
import numpy as np

inputs = tf.placeholder(tf.float32, shape=[None, None, None, 3])

conv1 = slim.conv2d(inputs, num_outputs=20, kernel_size=3, stride=4)

de_weight = tf.get_variable('de_weight', shape=[3, 3, 3, 20])

deconv1 = tf.nn.conv2d_transpose(conv1, filter=de_weight, output_shape=tf.shape(inputs),
                                 strides=[1, 4, 4, 1], padding='SAME')

loss = deconv1 - inputs
train_op = tf.train.GradientDescentOptimizer(0.001).minimize(loss)

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

    for i in range(10):
        data_in = np.random.normal(size=[3, 97, 97, 3])
        aa, los_ = sess.run([train_op, loss], feed_dict={inputs: data_in})
        print(aa.shape)