hello @dkurt, current implement is work well, but a little different from tensorflow.
for example:
Input size: (1, 3, 525, 700)
Tensorflow Output size is: (1, 3, 525, 700)
Our Output size is:(1, 3, 524, 700)
I think i should need to fix this problem, any solution?

@vonchenplus, For which parameters you have this shapes? Please share input resolution, kernel size, strides and paddings.

@dkurt ,
for example:
inp = tf.placeholder(tf.float32, [1, 525, 700, 3], 'input')
pool = tf.layers.max_pooling2d(inp, pool_size=(2, 2), strides=(2, 2))
conv = tf.layers.conv2d(inputs=pool, filters=3, kernel_size=[1, 1], padding='VALID')
unpool = MaxUnPooling2D(pool.op, conv)

layer.input(0) == pool.op.input == [1, 525, 700, 3]
layer,input(1) == pool.op.output == [1, 262, 350, 3]
layer.input(2) == conv == [1, 262, 350, 3]
tensorflow's unpool size is [1, 525, 700, 3], current implement size is [1, 524, 700, 3]

the first input has this shapes.

Can you please share text graph so we can see if there are useful parameters in TF graph. You can do it in Python:

for node in graph_def.node:
  print(node)

hello @dkurt, this is MaxPoolGrad text graph Info, .

name: "MaxPoolGrad"
op: "MaxPoolGrad"
input: "X_content"
input: "max_pooling2d/MaxPool"
input: "conv2d/BiasAdd"
attr {
  key: "T"
  value {
    type: DT_FLOAT
  }
}
attr {
  key: "data_format"
  value {
    s: "NHWC"
  }
}
attr {
  key: "ksize"
  value {
    list {
      i: 1
      i: 2
      i: 2
      i: 1
    }
  }
}
attr {
  key: "padding"
  value {
    s: "VALID"
  }
}
attr {
  key: "strides"
  value {
    list {
      i: 1
      i: 2
      i: 2
      i: 1
    }
  }
}

I don't think there are other parameters with the correct shape, so I connect Maxpool's input as the third input, please have a look, thanks.

As usual avoid using of merge commits in patches.
Use rebase instead (git pull --rebase or git rebase -i updated_upstream_head).

To fix that:

• save you work + git branch -f saved_work
• setup upstream remote by following this GitHub guide
• fetch fresh upstream code: git fetch upstream.
• rebase your commits onto actual 3.4 branch:
  git rebase -i upstream/3.4
  (check list of your commits, save and quit (Esc + "wq" + Enter)
• resolve conflicts if any. To continue run git rebase --continue.
• check commits and diffs (gitk is a nice tool for that: gitk --all --date-order &)
• check files with your saved copy from step 1.
• push rebased commits into source branch of your fork (with --force option)

Note: no needs to re-open PR, apply changes "inplace". Do NOT remove your fork or your branch in GitHub - this would break/close this PR completely.

hello @alalek, I fixed this problem, thinks.

Please add skip for Inference Engine backend tests (you may see other tests for reference):

failed tests: https://pullrequest.opencv.org/buildbot/builders/precommit_custom_linux/builds/2327