Credits :: Tensorflow's Issue No 2169 and all the people invloved there.

Unpooling in Tensorflow

Unpooling of tensor that has been pooled using max_pool_with_argmax

What is unpooling ?

It's a tensor operation which tries to regenerate the tensor which has been maxpooled .

This figure below illustrates it perfectly

More on max pooling here ::Max Pooling

The original paper on unpooling :: Unpooling

Some more resources one may find useful::

1. Broadcasting in Numpy
2. Scatter in tensorflow
3. Cocatanation in Tensorflow
4. Max Pool with Arg Max in Tensorflow

Example Usage ::

1. Declare a tensor

original_tensor = tf.random_uniform([1,4,4,3],maxval=100,dtype='float32',seed=2)

2. Use the max pool with arg max function to pool the tensor

pooled_tensor,max_indices=tf.nn.max_pool_with_argmax(original_tensor, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')

3. Now armed with pooled_tensor and max_indices you can call the unpool_with_argamx function !

unpooled_tensor=unpool_with_with_argmax(pooled_tensor,max_indices)

4. It's done !

Lets have a look at it

1. Original Tensor

---

2. Pooled Tensor

---

3. Max Indices

---

4. And finally the Unpooled Tensor