-
Notifications
You must be signed in to change notification settings - Fork 9
/
run_cycgan_mm.py
executable file
·189 lines (139 loc) · 6.39 KB
/
run_cycgan_mm.py
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
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
# Copyright 2013-2018 Lawrence Livermore National Security, LLC and other
# @authors: :Rushil Anirudh, Jayaraman J. Thiagarajan, Timo Bremer
#
# SPDX-License-Identifier: MIT
import tensorflow as tf
import numpy as np
np.random.seed(4321)
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
import os
from modelsv2 import *
from utils import *
import shutil
import cPickle as pkl
import argparse
from sklearn.preprocessing import MinMaxScaler, scale
import wae_metric.model_AVB as wae
from wae_metric.utils import special_normalize
from wae_metric.run_WAE import LATENT_SPACE_DIM, load_dataset
IMAGE_SIZE = 64
batch_size = 64
def run(**kwargs):
fdir = kwargs.get('fdir','./outs')
modeldir = kwargs.get('modeldir','./pretrained_model')
ae_path = kwargs.get('ae_dir','./wae_metric/pretrained_model')
datapath = kwargs.get('datapath','./data/')
visdir = './tensorboard_plots'
if not os.path.exists(fdir):
os.makedirs(fdir)
if not os.path.exists(modeldir):
os.makedirs(modeldir)
if not os.path.exists(visdir):
os.makedirs(visdir)
jag_inp, jag_sca, jag_img = load_dataset(datapath)
tr_id = np.random.choice(jag_sca.shape[0],int(jag_sca.shape[0]*0.95),replace=False)
te_id = list(set(range(jag_sca.shape[0])) - set(tr_id))
X_train = jag_inp[tr_id,:]
y_sca_train = jag_sca[tr_id,:]
y_img_train = jag_img[tr_id,:]
np.random.shuffle(te_id)
X_test = jag_inp[te_id,:]
y_sca_test = jag_sca[te_id,:]
y_img_test = jag_img[te_id,:]
y_img_test_mb = y_img_test[-16:,:]
y_img_test_mb = y_img_test_mb.reshape(16,64,64,4)
for k in range(4):
fig = plot(y_img_test_mb[:,:,:,k],immax=np.max(y_img_test_mb[:,:,:,k].reshape(-1,4096),axis=1),
immin=np.min(y_img_test_mb[:,:,:,k].reshape(-1,4096),axis=1))
plt.savefig('{}/gt_img_{}_{}.png'
.format(fdir,str(k).zfill(3),str(k)), bbox_inches='tight')
plt.close()
print("Dataset dimensions: ",X_test.shape,y_sca_test.shape,y_img_test.shape)
dim_x = X_train.shape[1]
dim_y_sca = y_sca_train.shape[1]
dim_y_img = y_img_train.shape[1]
dim_y_img_latent = LATENT_SPACE_DIM #latent space
### Metric params
''' TEST mini-batch '''
x_test_mb = X_test[-100:,:]
y_sca_test_mb = y_sca_test[-100:,:]
y_img_test_mb = y_img_test[-100:,:]
y_sca = tf.placeholder(tf.float32, shape=[None, dim_y_sca])
y_img = tf.placeholder(tf.float32, shape=[None, dim_y_img])
x = tf.placeholder(tf.float32, shape=[None, dim_x])
train_mode = tf.placeholder(tf.bool,name='train_mode')
y_mm = tf.concat([y_img,y_sca],axis=1)
'''**** Encode the img, scalars ground truth --> latent vector for loss computation ****'''
y_latent_img = wae.gen_encoder_FCN(y_mm, dim_y_img_latent,train_mode=False)
'''**** Train cycleGAN input params <--> latent space of (images, scalars) ****'''
cycGAN_params = {'input_params':x,
'outputs':y_latent_img,
'param_dim':dim_x,
'output_dim':dim_y_img_latent,
'L_adv':1e-2,
'L_cyc':1e-1,
'L_rec':1}
JagNet_MM = cycModel_MM(**cycGAN_params)
JagNet_MM.run(train_mode)
'''**** Decode the prediction from latent vector --> img, scalars ****'''
y_img_out = wae.var_decoder_FCN(JagNet_MM.output_fake, dim_y_img+dim_y_sca,train_mode=False)
img_loss = tf.reduce_mean(tf.square(y_img_out[:,:16384] - y_img))
sca_loss = tf.reduce_mean(tf.square(y_img_out[:,16384:] - y_sca))
fwd_img_summary = tf.summary.scalar(name='Image Loss', tensor=img_loss)
fwd_sca_summary = tf.summary.scalar(name='Scalar Loss', tensor=sca_loss)
merged = tf.summary.merge_all()
t_vars = tf.global_variables()
m_vars = [var for var in t_vars if 'wae' in var.name]
metric_saver = tf.train.Saver(m_vars)
saver = tf.train.Saver(list(set(t_vars)-set(m_vars)))
sess = tf.Session()
sess.run(tf.global_variables_initializer())
ckpt = tf.train.get_checkpoint_state(modeldir)
ckpt_metric = tf.train.get_checkpoint_state(ae_path)
if ckpt_metric and ckpt_metric.model_checkpoint_path:
metric_saver.restore(sess, ckpt_metric.model_checkpoint_path)
print("************ Image Metric Restored! **************")
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
print("************ Model restored! **************")
writer = tf.summary.FileWriter(visdir+'/{}'.format(modeldir), sess.graph)
i = 0
for it in range(100000):
randid = np.random.choice(X_train.shape[0],batch_size,replace=False)
x_mb = X_train[randid,:]
y_img_mb = y_img_train[randid,:]
y_sca_mb = y_sca_train[randid,:]
fd = {x: x_mb, y_sca: y_sca_mb,y_img:y_img_mb,train_mode:True}
for _ in range(10):
_,dloss = sess.run([JagNet_MM.D_solver,JagNet_MM.loss_disc],feed_dict=fd)
gloss0,gloss1,gadv = sess.run([JagNet_MM.loss_gen0,
JagNet_MM.loss_gen1,
JagNet_MM.loss_adv],
feed_dict=fd)
for _ in range(1):
_ = sess.run([JagNet_MM.G0_solver],feed_dict=fd)
if it % 100 == 0:
print('Fidelity -- Iter: {}; Forward: {:.4f}; Inverse: {:.4f}'
.format(it, gloss0, gloss1))
print('Adversarial -- Disc: {:.4f}; Gen: {:.4f}\n'.format(dloss,gadv))
if it % 500 == 0:
nTest=16
x_test_mb = X_test[-nTest:,:]
summary_val = sess.run(merged,feed_dict={x:X_test,y_sca:y_sca_test,y_img:y_img_test,train_mode:False})
writer.add_summary(summary_val, it)
samples,samples_x = sess.run([y_img_out,JagNet_MM.input_cyc],
feed_dict={x: x_test_mb,train_mode:False})
data_dict= {}
data_dict['samples'] = samples
data_dict['samples_x'] = samples_x
data_dict['y_sca'] = y_sca_test
data_dict['y_img'] = y_img_test
data_dict['x'] = x_test_mb
test_imgs_plot(fdir,it,data_dict)
save_path = saver.save(sess, "./"+modeldir+"/model_"+str(it)+".ckpt")
if __name__=='__main__':
run()
### srun -n 1 -N 1 -p pbatch -A lbpm --time=3:00:00 --pty /bin/sh