Un-pooling 2D

Warning

Still has a bug, if argmax has coincident indexes. Don’t know if this is a desired behaivour or not.

class ComplexUnPooling2D

This class was inspired to recreate the CV-FCN model of [CIT2019-CAO]

Performs UnPooling as explained here.

There are 2 main ways to use this function

  • Using the expected output shape or
  • Using the upsampling_factor parameter.

The second options is the only way to deal with partially known output, for example (None, None, 3) to deal with variable size iamges.

../_images/max_unpool_explain.png

Usage example with desired_output_shape

from cvnn.layers import ComplexUnPooling2D, complex_input, ComplexMaxPooling2DWithArgmax
import tensorflow as tf
import numpy as np

def get_img():
    img_r = np.array([[
        [0, 1, 2],
        [0, 2, 2],
        [0, 5, 7]
    ], [
        [0, 7, 5],
        [3, 7, 9],
        [4, 5, 3]
    ]]).astype(np.float32)
    img_i = np.array([[
        [0, 4, 5],
        [3, 7, 9],
        [4, 5, 3]
    ], [
        [0, 4, 5],
        [3, 2, 2],
        [4, 8, 9]
    ]]).astype(np.float32)
    img = img_r + 1j * img_i
    img = np.reshape(img, (2, 3, 3, 1))
    return img

x = get_img()       # Gets an image just for the example
inputs = complex_input(shape=x.shape[1:])
# Apply max-pooling and gets also argmax
max_pool_o, max_arg = ComplexMaxPooling2DWithArgmax(strides=1, data_format="channels_last", name="argmax")(inputs)
# Applies the Unpooling
outputs = ComplexUnPooling2D(x.shape[1:])([max_pool_o, max_arg])

model = tf.keras.Model(inputs=inputs, outputs=outputs, name="pooling_model")
model.summary()
model(x)

Usage example with upsampling_factor

inputs = complex_input(shape=(None, None, 3))  # Input is an unknown size RGB image
max_pool_o, max_arg = ComplexMaxPooling2DWithArgmax(strides=1, data_format="channels_last", name="argmax")(inputs)
unpool = ComplexUnPooling2D(upsampling_factor=2)([input_to_block, pool_argmax])

model = tf.keras.Model(inputs=inputs, outputs=outputs, name="pooling_model")
model.summary()
model(x)
__init__(self, desired_output_shape, name=None, dtype=DEFAULT_COMPLEX_TYPE, dynamic=False, **kwargs)
Parameters:
  • desired_output_shape – tf.TensorShape (or equivalent like tuple or list). The expected output shape without the batch size. Meaning that for a 2D image to be enlarged, this is size 3 of the form HxWxC or CxHxW
  • upsampling_factor – Integer. The factor to which enlarge the image. For example, if upsampling_factor=2, an input image of size 32x32 will be 64x64. This parameter is ignored if desired_output_shape is used or if the output shape is given to the call funcion.
call(self, inputs, **kwargs)
Parameters:inputs –

A tuple of Tensor objects (input, argmax).

  • input A Tensor.
  • argmax A Tensor. The indices in argmax are flattened (Complains directly to TensorFlow)
  • output_shape (Optional) A tf.TensorShape (or equivalent like tuple or list). The expected output shape without the batch size. Meaning that for a 2D image to be enlarged, this is size 3 of the form HxWxC or CxHxW
[CIT2019-CAO]Cao, Wu, Zhang, Liang and Li. “Pixel-Wise PolSAR Image Classification via a Novel Complex-Valued Deep Fully Convolutional Network” DPI: 10.3390/rs11222653, 2019.