使用我自己的数据的Tensorflow矩阵大小错误

Question

I'm begginner of tensorflow. I wanted to use with my own medical raw image data and make simple autoencorder but I fail.I guess marix size is wrong. This is perhaps a noob question but I can't figure it out.

my image data size is 512*512*1 and dtype is int16.

---

error

Traceback (most recent call last):
File "/Users/tk/Desktop/tensorflow_test/test.py", line 17, in <module>
decoded = tf.sigmoid(tf.matmul(w_dec,encoded) + b_dec)
File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/math_ops.py", line 814, in binary_op_wrapper
return func(x, y, name=name)
File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/gen_math_ops.py", line 73, in add
result = _op_def_lib.apply_op("Add", x=x, y=y, name=name)
File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/op_def_library.py", line 759, in apply_op
op_def=op_def)
File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2242, in create_op
set_shapes_for_outputs(ret)
File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1617, in set_shapes_for_outputs
shapes = shape_func(op)
File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1568, in call_with_requiring
return call_cpp_shape_fn(op, require_shape_fn=True)
File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/common_shapes.py", line 610, in call_cpp_shape_fn
debug_python_shape_fn, require_shape_fn)
File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/common_shapes.py", line 675, in _call_cpp_shape_fn_impl
raise ValueError(err.message)
ValueError: Dimensions must be equal, but are 625 and 262144 for 'add_1' (op: 'Add') with input shapes: [262144,625], [262144].

---

my code

import tensorflow as tf
import numpy as np

#load_image
temp = np.fromfile('/Users/learn.raw',np.int16)
input_image = tf.to_float(temp)

# Variables
x = tf.placeholder("float",None)
w_enc = tf.Variable(tf.random_normal([625, 262144], mean=0.0, stddev=0.05))
w_dec = tf.Variable(tf.random_normal([262144, 625], mean=0.0, stddev=0.05))
b_enc = tf.Variable(tf.zeros([625]))
b_dec = tf.Variable(tf.zeros([262144]))

#model
encoded = tf.sigmoid(tf.matmul(w_enc,x) + b_enc)
decoded = tf.sigmoid(tf.matmul(w_dec,encoded) + b_dec)


# Cost Function
cross_entropy = -1. * x * tf.log(decoded) - (1. - x) * tf.log(1. - decoded)
loss = tf.reduce_mean(cross_entropy)
train_step = tf.train.AdagradOptimizer(0.1).minimize(loss)

# Train
init = tf.initialize_all_variables()

with tf.Session() as sess:
    sess.run(init)
    print('Training...')
    train_step.run(feed_dict = {x: input_image.eval()})

Answer 1

The dimensions of x, b_enc, d_dec were wrong. Please see the working version below. In general, it is useful first to implement the algorithm in numpy and then port to tensorflow if you begin with TF.

import tensorflow as tf
import numpy as np

tf.reset_default_graph()

#load_image
image_size = 16
k = 64
temp = np.zeros((image_size, image_size))


# Variables
x_placeholder = tf.placeholder("float", (image_size, image_size))

x = tf.reshape(x_placeholder, [image_size * image_size, 1])
w_enc = tf.Variable(tf.random_normal([k, image_size * image_size], mean=0.0, stddev=0.05))
w_dec = tf.Variable(tf.random_normal([image_size * image_size, k], mean=0.0, stddev=0.05))
b_enc = tf.Variable(tf.zeros([k, 1]))
b_dec = tf.Variable(tf.zeros([image_size * image_size, 1]))

#model
encoded = tf.sigmoid(tf.matmul(w_enc, x) + b_enc)
decoded = tf.sigmoid(tf.matmul(w_dec,encoded) + b_dec)


# Cost Function
cross_entropy = -1. * x * tf.log(decoded) - (1. - x) * tf.log(1. - decoded)
loss = tf.reduce_mean(cross_entropy)
train_step = tf.train.AdagradOptimizer(0.1).minimize(loss)

# Train
init = tf.global_variables_initializer()

with tf.Session() as sess:
    sess.run(init)
    print('Training...')
    for _ in xrange(10):
      loss_val, _ = sess.run([loss, train_step], feed_dict = {x_placeholder: temp})
      print loss_val
    print('Done!')