saving example

In [4]:

    

import os
os.environ['TF_CPP_MIN_LOG_LEVEL']='2'
import numpy as np
np.set_printoptions(threshold=np.nan)
import tensorflow as tf
import time
import pandas as pd
import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
import progressbar
import iqa_tools


    

In [5]:

    

sess = tf.Session()
new_saver = tf.train.import_meta_graph('ssimNET0.meta')
new_saver.restore(sess, tf.train.latest_checkpoint('./'))


    

    




INFO:tensorflow:Restoring parameters from ./ssimNET4900

In [6]:

    

graph = tf.get_default_graph()
op_restore = graph.get_tensor_by_name('BiasAdd_4:0')
x = graph.get_tensor_by_name('Placeholder:0')


    

In [7]:

    

data_path = '/home/dirty_mike/Dropbox/github/image_quality_analysis/data/sample_data/'
train_features, train_target, test_features, test_target = iqa_tools.load_data(local=True, path=data_path)


    

In [8]:

    

train_features.shape


    

    
Out[8]:





(500, 96, 96, 4)

In [12]:

    

pred = sess.run(op_restore, feed_dict={x: test_features})


    

In [13]:

    

pred.shape


    

    
Out[13]:





(140, 86, 86, 1)

In [20]:

    

plt.figure(figsize = (16,12))
gs1 = gridspec.GridSpec(3, 4)
gs1.update(wspace=0, hspace=0.03)

for j in range(3):
    index = np.random.randint(pred.shape[0])
    ax1, ax2, ax3, ax4 = plt.subplot(gs1[4*j]), plt.subplot(gs1[4*j+1]), plt.subplot(gs1[4*j+2]), plt.subplot(gs1[4*j+3])
    for ax in [ax1, ax2, ax3, ax4]:
        ax.set_xticklabels([])
        ax.set_yticklabels([])
        ax.get_xaxis().set_visible(False)
        ax.get_yaxis().set_visible(False)
    if j == 0:
        ax1.set_title('original', size=20)
        ax2.set_title('reconstructed', size=20)
        ax3.set_title('ssim', size=20)
        ax4.set_title('ssim net prediction', size=20)

    ax1.imshow(test_features[index,:,:,0], cmap='gray')
    ax2.imshow(test_features[index,:,:,1], cmap='gray')
    ax3.imshow(test_target[index,:,:,0], cmap='plasma')
    ax4.imshow(pred[index,:,:,0], cmap='plasma')
plt.savefig('prediction.png')
plt.show()


    

    






<matplotlib.figure.Figure at 0x7f93e006e470>






    

In [26]:

    

weights1 = sess.run(graph.get_tensor_by_name('weights1:0'))
weights1.shape


    

    
Out[26]:





(11, 11, 4, 100)

In [44]:

    

plt.figure(figsize = (16,12))
gs1 = gridspec.GridSpec(3, 4)
gs1.update(wspace=0, hspace=0.03)

for j in range(3):
    index = np.random.randint(100)
    ax1, ax2, ax3, ax4 = plt.subplot(gs1[4*j]), plt.subplot(gs1[4*j+1]), plt.subplot(gs1[4*j+2]), plt.subplot(gs1[4*j+3])
    for ax in [ax1, ax2, ax3, ax4]:
        ax.set_xticklabels([])
        ax.set_yticklabels([])
        ax.get_xaxis().set_visible(False)
        ax.get_yaxis().set_visible(False)
    if j == 0:
        ax1.set_title('original', size=20)
        ax2.set_title('recon', size=20)
        ax3.set_title('original sq', size=20)
        ax4.set_title('recon sq', size=20)

    ax1.imshow(weights1[:,:,0,index], cmap='gray')
    ax2.imshow(weights1[:,:,1,index], cmap='gray')
    ax3.imshow(weights1[:,:,2,index], cmap='gray')
    ax4.imshow(weights1[:,:,3,index], cmap='gray')
plt.savefig('trained_filters.png')
plt.show()