Issue 4746

[guotong1988]

#4746

This code is the first commit to fix issue-4746 . I will commit the unit test code and doc later.
Thank you for your review . Thank you for any advice.I will work on it again for the advice and learn from the reviewers.

I test this code with:

x_image = tf.placeholder(tf.float32,shape=[4,4])
x = tf.reshape(x_image,[1,4,4,1])

ksize_feed = np.array([1,2,2,1])

ksize_init = tf.placeholder(tf.int32, shape=[4])
ksize0 = tf.reshape(ksize_init,[2,2])
ksize1 = tf.reshape(ksize0,[1,2,2,1])

strides=[1,2,2,1]
padding = 'VALID'

y = tf.nn.max_pool(value=x,ksize=ksize1,strides=strides,padding=padding)

x_data = np.array([[4,3,1,8],[7,2,6,3],[2,0,1,1],[3,4,2,5]],
            dtype = np.float32)

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

    x = (sess.run(x,feed_dict={x_image:x_data}))
    y = (sess.run(y,feed_dict={x_image:x_data,ksize_init:ksize_feed}))

    print "The shape of x:\t", x.shape, ",\t and the x.reshpe(4,4) is :"
    print x.reshape(4,4)
    print ""

    print "The shape of y:\t", y.shape, ",\t and the y.reshpe(2,2) is :"
    print y.reshape(2,2)
    print ""

[tensorflow-jenkins]

Can one of the admins verify this patch?

[mention-bot]

@guotong1988, thanks for your PR! By analyzing the history of the files in this pull request, we identified @vrv, @keveman and @zheng-xq to be potential reviewers.

[guotong1988]

also test the code with:

import numpy as np
import tensorflow as tf

x_image = tf.placeholder(tf.float32,shape=[4,4])
x = tf.reshape(x_image,[1,4,4,1])

ksize = np.array([1,2,2,1])

strides=[1,2,2,1]
padding = 'VALID'

y = tf.nn.max_pool(value=x,ksize=ksize,strides=strides,padding=padding)

x_data = np.array([[4,3,1,8],[7,2,6,3],[2,0,1,1],[3,4,2,5]],
            dtype = np.float32)

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

    x = (sess.run(x,feed_dict={x_image:x_data}))
    y = (sess.run(y,feed_dict={x_image:x_data}))

    print "The shape of x:\t", x.shape, ",\t and the x.reshpe(4,4) is :"
    print x.reshape(4,4)
    print ""

    print "The shape of y:\t", y.shape, ",\t and the y.reshpe(2,2) is :"
    print y.reshape(2,2)
    print ""

[drpngx]

Did you try building this? This looks like a syntax error.

[guotong1988]

It's my mistake.I did build it but I make a copy mistake.

[drpngx]

We need to clean these up.

[zheng-xq]

Unfortunately, modifying an existing op like this breaks the backward compatiblity of TensorFlow.

For now, it is a good idea to add a custom op on github outside TensorFlow, so people can try it if they are interested. When it gets popular and used in more models, we can move it to TF contrib or TF core.

https://www.tensorflow.org/versions/r0.11/how_tos/adding_an_op/index.html

Thanks.

[guotong1988]

OK , I will take another issue.

[drpngx]

Thanks for looking into this! I'll close the PR. Feel free to reopen if appropriate.