Added learning fft model example

Author: michaelmendoza

examples/fft/data_loader.py

@@ -0,0 +1,58 @@
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+import os
+
+import numpy as np
+import matplotlib.pyplot as plt
+import matplotlib.image as mpimg
+
+class DataLoader:
+
+    def __init__(self):
+        self.load()
+
+    def get(self):
+        input = np.reshape(self.input, (-1, self.WIDTH * self.HEIGHT * 2))
+        output = np.reshape(self.output, (-1, self.WIDTH * self.HEIGHT * 2))
+        return input, output
+
+    def load(self):
+
+        filepath = '.'
+        filename = 'shepp256.png'
+        img = mpimg.imread(os.path.join(filepath, filename))
+        img = np.array(img)
+
+        self.WIDTH = img.shape[0]
+        self.HEIGHT = img.shape[1]
+        self.CHANNELS = 2
+
+        kSpace = np.fft.ifftshift(np.fft.fft2(img))
+        inverse = np.fft.ifft2(kSpace)
+
+        self.input = np.dstack((np.abs(kSpace), np.angle(kSpace)))
+        self.output = np.dstack((np.abs(inverse), np.angle(inverse)))
+
+    def show(self):
+
+        input = self.input[:,:,0] * np.exp(1j*self.input[:,:,1])
+        output = self.output[:,:,0] * np.exp(1j*self.output[:,:,1])
+
+        plt.subplot(2, 1, 1)
+        plt.imshow(np.abs(input), cmap='gray')
+
+        plt.subplot(2, 1, 2)
+        plt.imshow(np.abs(output), cmap='gray')
+        plt.show()
+
+    def info(self):
+        print(self.input.dtype)
+
+if __name__ == '__main__':
+    data = DataLoader()
+    data.info()
+    data.show()
+    x, y = data.get()
+    print(x.shape, y.shape)

examples/fft/fft.py

@@ -0,0 +1,69 @@
+
+'''
+Tensorflow Code for a fourier transform network
+'''
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import tensorflow as tf 
+import numpy as np
+import matplotlib
+import matplotlib.pyplot as plt
+import os
+os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
+
+# Import Dataset
+from data_loader import DataLoader
+data = DataLoader()
+print('Data Loaded')
+
+# Training Parameters
+learning_rate = 0.0001
+num_steps = 10000
+batch_size = 32
+display_step = 100
+
+# Network Parameters 
+WIDTH = data.WIDTH; HEIGHT = data.HEIGHT; CHANNELS = data.CHANNELS
+NUM_INPUTS = WIDTH * HEIGHT * CHANNELS
+NUM_OUTPUTS =  WIDTH * HEIGHT * CHANNELS
+
+# Network Varibles and placeholders
+X = tf.placeholder(tf.float64, [None, NUM_INPUTS])  # Input
+Y = tf.placeholder(tf.float64, [None, NUM_OUTPUTS]) # Truth Data - Output
+
+# Network Architecture
+def simple_net(x): 
+    he_init = tf.contrib.layers.variance_scaling_initializer()
+    fc1 = tf.layers.dense(x, 128, activation=tf.nn.relu, kernel_initializer=he_init, name='fc1')   
+    fc2 = tf.layers.dense(fc1, NUM_OUTPUTS, activation=None, kernel_initializer=he_init, name='fc2') 
+    return fc2
+
+# Define loss and optimizer
+prediction = simple_net(X) #unet(X)
+loss = tf.reduce_mean(tf.square(prediction - Y))
+optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)
+trainer = optimizer.minimize(loss)
+
+# Initalize varibles, and run network 
+init = tf.global_variables_initializer()
+sess = tf.Session()
+sess.run(init)
+
+print ('Start Training: BatchSize:', batch_size,' LearningRate:', learning_rate)
+
+for step in range(num_steps):
+    x, y = data.get()
+    sess.run(trainer, feed_dict={X:x, Y:y})
+
+    if(step % display_step == 0):
+        _loss = sess.run(loss, feed_dict={ X:x, Y:y })    
+        print("Step: " + str(step) + " Loss: " + str(_loss)) 
+
+x, y = data.get()
+img = sess.run(prediction, feed_dict={X:x})
+img = np.reshape(img, (data.WIDTH, data.HEIGHT, data.CHANNELS))
+plt.imshow(img[:,:,0], cmap="gray")
+plt.show()

examples/fft/fft_example.py

@@ -0,0 +1,26 @@
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+import os
+
+import numpy as np
+import matplotlib.pyplot as plt
+import matplotlib.image as mpimg
+
+
+filepath = '.'
+filename = 'shepp256.png'
+img = mpimg.imread(os.path.join(filepath, filename))
+img = np.array(img)
+
+kSpace = np.fft.ifftshift(np.fft.fft2(img))
+inverse = np.fft.ifft2(kSpace)
+
+plt.subplot(2, 1, 1)
+plt.imshow(np.abs(kSpace), cmap='gray')
+
+plt.subplot(2, 1, 2)
+plt.imshow(np.abs(inverse), cmap='gray')
+plt.show()
+