lambda-net/main.py at master · Zhengqi-code/lambda-net · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
import scipy.io as sio
import h5py
import numpy as np
import matplotlib.pyplot as plt
import time
import operation
from datetime import datetime
import os
import tensorflow as tf
import scipy.io as scipyio
import deeper_u_net
import math
from scipy import interpolate
import math
import discriminator

os.environ["CUDA_VISIBLE_DEVICES"]="0 "

subdir = datetime.strftime(datetime.now(), '%Y%m%d-%H%M%S')

model_dir = os.path.join('models/', subdir)
if not os.path.isdir(model_dir):  # Create the model directory if it doesn't exist
    os.makedirs(model_dir)

result_dir = os.path.join('result/', subdir)
if not os.path.isdir(result_dir):  # Create the result directory if it doesn't exist
    os.makedirs(result_dir)

phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')

batch_size=10
masks=sio.loadmat("mask.mat")['data']


training_data_whole = operation.load_hscnn()
testing_data=sio.loadmat('testing_data/test.mat')['data'] #(10,256,256,24)
testing_data=operation.normalize_0_to_1(testing_data)
measurement_test=np.sum(testing_data*masks,axis=3)[:,:,:,np.newaxis] # 20,256,256,1
test_input=np.concatenate((measurement_test,np.tile(masks[np.newaxis,:,:,:],[batch_size,1,1,1])),3) # 20,256,256,25

images_input=tf.placeholder(tf.float32,shape=[batch_size,256,256,25])
ground_truth=tf.placeholder(tf.float32,shape=[batch_size,256,256,24])


output6,output12,output_g=deeper_u_net.inference(images_input,0.8,phase_train_placeholder)


L2_loss24=tf.reduce_mean(tf.square(tf.subtract(ground_truth,output_g)))
L2_loss12=tf.reduce_mean(tf.square(tf.subtract(ground_truth[:,:,:,::2],output12)))
L2_loss6=tf.reduce_mean(tf.square(tf.subtract(ground_truth[:,:,:,::4],output6)))
L2_loss=0.1*L2_loss6+0.1*L2_loss12+L2_loss24


real_AB = tf.concat([images_input[:,:,:,0:1], ground_truth], 3)
fake_AB = tf.concat([images_input[:,:,:,0:1], output_g], 3)
D_logits = discriminator.inference(real_AB,1.0,phase_train_placeholder,0.0,reuse=False)
D_logits_ = discriminator.inference(fake_AB,1.0,phase_train_placeholder, 0.0,reuse=True)

d_loss_real=tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_logits,labels=tf.ones_like(D_logits)))
d_loss_fake=tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_logits_,labels=tf.zeros_like(D_logits_)))
d_loss=d_loss_real+d_loss_fake


g_loss=tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_logits_,labels=tf.ones_like(D_logits_)))+100*L2_loss


t_vars=tf.trainable_variables()
g_vars=[var for var in t_vars if 'generator' in var.name]
d_vars=[var for var in t_vars if 'discriminator' in var.name]


global_step=tf.Variable(0,trainable=False)

d_optim=tf.train.AdamOptimizer(0.00015,beta1=0.5).minimize(d_loss,global_step=global_step,var_list=d_vars)
g_optim=tf.train.AdamOptimizer(0.00015,beta1=0.5).minimize(g_loss,global_step=global_step,var_list=g_vars)

# Create a saver
saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=3)
#
#
sess=tf.InteractiveSession()
sess.run(tf.global_variables_initializer())
sess.run(tf.local_variables_initializer())
#
# ckpt = tf.train.get_checkpoint_state('models/20190830-815')
# saver.restore(sess, ckpt.model_checkpoint_path)

last_round_psnr=[0 for i in range(batch_size)]
for epoch in range(1):
    for i in range(100000000):
        if i % 200 == 0:
            recon_test = sess.run(output_g, feed_dict={images_input: test_input,phase_train_placeholder: False})


            for kk in range(batch_size):
                psnr = 0.0
                for w in range(24):
                    single_mse=np.mean((recon_test[kk,:,:,w] - testing_data[kk,:,:,w]) ** 2)
                    psnr+=20 * math.log10(1 / math.sqrt(single_mse))
                with open(os.path.join(result_dir, 'psnr.txt'), 'at') as f:
                    f.write('PSNR for scene' + str(kk + 1) + ': %.5f\n' % (psnr/24.0))
                if psnr/24.0 > last_round_psnr[kk]:
                    sio.savemat(result_dir + '/test_result'+str(kk+1)+'.mat', {'result': recon_test[kk,:,:,:]})
                    last_round_psnr[kk] = psnr/24.0

        training_data_batch = operation.shuffle_crop(training_data_whole, batch_size=batch_size)  # (10, 256, 256, 24)
        training_data_batch =operation.normalize_0_to_1(training_data_batch)
        measurement_train   = np.sum(training_data_batch * masks, axis=3)[:, :, :, np.newaxis]

        train_input = np.concatenate((measurement_train, np.tile(masks[np.newaxis, :, :, :], [batch_size, 1, 1, 1])), 3)

        out,train_err,_, step = sess.run([output_g,L2_loss,g_optim, global_step], feed_dict={images_input: train_input,
                                                                                ground_truth: training_data_batch,
                                                                                phase_train_placeholder: True})

        _ = sess.run( d_optim, feed_dict={images_input: train_input, phase_train_placeholder: True, ground_truth: training_data_batch})


        print 'training_mse:',train_err
        with open(os.path.join(result_dir, 'psnr.txt'), 'at') as f:
            f.write('train_mse: %.5f\n' % (train_err))

        checkpoint_path = os.path.join(model_dir, 'model-%s.ckpt' % subdir)
        saver.save(sess, checkpoint_path, global_step=i, write_meta_graph=False)