-
Notifications
You must be signed in to change notification settings - Fork 20
/
Copy pathHiNet_SynthModel.py
214 lines (143 loc) · 8.17 KB
/
HiNet_SynthModel.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
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Wed Apr 24 14:04:44 2019
@author: tao
"""
import os
import torch
import torch.nn as nn
import numpy as np
import time
import datetime
from torch.utils.data import DataLoader
#from models import *
#from fusion_models import * # revise in 09/03/2019
from dataset import MultiModalityData_load
from funcs.utils import *
import torch.nn as nn
import scipy.io as scio
from torch.autograd import Variable
import torch.autograd as autograd
#import IVD_Net as IVD_Net
import model.syn_model as models
#from config import opt
#from visualize import Visualizer
#testing
#os.environ["CUDA_VISIBLE_DEVICES"] = '5,6'
cuda = True if torch.cuda.is_available() else False
FloatTensor = torch.cuda.FloatTensor if cuda else torch.FloatTensor
LongTensor = torch.cuda.LongTensor if cuda else torch.LongTensor
class LatentSynthModel():
###########################################################################
def __init__(self,opt):
self.opt = opt
self.generator = models.Multi_modal_generator(1,1,32)
self.discrimator = models.Discriminator()
if opt.use_gpu:
self.generator = self.generator.cuda()
self.discrimator = self.discrimator.cuda()
if torch.cuda.device_count() > 1:
self.generator = nn.DataParallel(self.generator,device_ids=self.opt.gpu_id)
self.discrimator = nn.DataParallel(self.discrimator,device_ids=self.opt.gpu_id)
###########################################################################
def train(self):
if not os.path.isdir(self.opt.save_path+'/'+'task_'+str(self.opt.task_id)+'/'):
mkdir_p(self.opt.save_path+'/'+'task_'+str(self.opt.task_id)+'/')
logger = Logger(os.path.join(self.opt.save_path+'/'+'task_'+str(self.opt.task_id)+'/'+'run_log.txt'), title='')
logger.set_names(['Run epoch', 'D Loss', 'G Loss'])
#
self.generator.apply(weights_init_normal)
self.discrimator.apply(weights_init_normal)
print('weights_init_normal')
# Optimizers
optimizer_D = torch.optim.Adam(self.discrimator.parameters(), lr=self.opt.lr,betas=(self.opt.b1, self.opt.b2))
optimizer_G = torch.optim.Adam(self.generator.parameters(),lr=self.opt.lr,betas=(self.opt.b1, self.opt.b2))
# Learning rate update schedulers
lr_scheduler_G = torch.optim.lr_scheduler.LambdaLR(optimizer_G, lr_lambda=LambdaLR(self.opt.epochs, 0, self.opt.decay_epoch).step)
lr_scheduler_D = torch.optim.lr_scheduler.LambdaLR(optimizer_D, lr_lambda=LambdaLR(self.opt.epochs, 0, self.opt.decay_epoch).step)
# Lossesgenerator
criterion_GAN = nn.MSELoss().cuda()
criterion_identity = nn.L1Loss().cuda()
# Load data
train_data = MultiModalityData_load(self.opt,train=True)
train_loader = DataLoader(train_data,batch_size=self.opt.batch_size,shuffle=False)
batches_done = 0
prev_time = time.time()
# ---------------------------- *training * ---------------------------------
for epoch in range(self.opt.epochs):
for ii, inputs in enumerate(train_loader):
print(ii)
# define diferent synthesis tasks
[x1,x2,x3] = model_task(inputs,self.opt.task_id) # train different synthesis task
fake = torch.zeros([inputs[0].shape[1]*inputs[0].shape[0],1,6,6], requires_grad=False) #.cuda()
valid = torch.ones([inputs[0].shape[1]*inputs[0].shape[0],1,6,6], requires_grad=False)#.cuda()
###############################################################
if self.opt.use_gpu:
x1 = x1.cuda()
x2 = x2.cuda()
x3 = x3.cuda()
x_fu = torch.cat([x1,x2],dim=1)
# ----------------------
# Train generator
# ----------------------
optimizer_G.zero_grad()
x_fake,x1_re,x2_re = self.generator(x_fu)
# Identity loss
loss_re3 = criterion_identity(x_fake, x3)
loss_re1 = criterion_identity(x1_re, x1)
loss_re2 = criterion_identity(x2_re, x2)
# gan loss
loss_GAN = criterion_GAN(self.discrimator(x_fake), valid)
# total loss
loss_G = loss_GAN + 100*loss_re3 + 20*loss_re1 + 20*loss_re2
loss_G.backward(retain_graph=True)
optimizer_G.step()
# ----------------------
# Train Discriminators
# ----------------------
optimizer_D.zero_grad()
# Real loss
loss_real = criterion_GAN(self.discrimator(x3), valid)
loss_fake = criterion_GAN(self.discrimator(x_fake), fake)
# Total loss
loss_D = (loss_real + loss_fake) / 2
loss_D.backward(retain_graph=True)
optimizer_D.step()
# time
batches_left = self.opt.epochs * len(train_loader) - batches_done
time_left = datetime.timedelta(seconds=batches_left * (time.time() - prev_time) / self.opt.n_critic)
prev_time = time.time()
#print('Epoch:', epoch, '| D_loss: %.6f' % loss_D.item(),'| G_loss: %.6f' % loss_G.item())
print('\r[Epoch %d/%d]:' % (epoch, self.opt.epochs),'[Batch %d/%d]:' % (ii, len(train_loader)), '| D_loss: %.6f' % loss_D.item(),'| G_loss: %.6f' % loss_G.item(),'ETA: %s' %time_left)
batches_done += 1
# Update learning rates
lr_scheduler_G.step()
lr_scheduler_D.step()
logger.append([epoch, loss_D.item(), loss_G.item()])
# Save model checkpoints
if epoch > 20 and (epoch) % self.opt.checkpoint_interval == 0:
torch.save(self.generator.state_dict(), self.opt.save_path+'/'+'task_'+str(self.opt.task_id)+'/generator_%d.pkl' % (epoch))
torch.save(self.discrimator.state_dict(),self.opt.save_path+'/'+'task_'+str(self.opt.task_id)+'/discrimator_%d.pkl' % (epoch))
###########################################################################
def test(self,ind_epoch):
self.generator.load_state_dict(torch.load(self.opt.save_path+'/'+'task_'+str(self.opt.task_id)+'/'+ 'generator_'+str(ind_epoch)+'.pkl'),strict=False)
# Load data
te_data = MultiModalityData_load(self.opt,train=False,test=True)
te_loader = DataLoader(te_data,batch_size=self.opt.batch_size,shuffle=False)
pred_eva_set = []
for ii, inputs in enumerate(te_loader):
#print(ii)
# define diferent synthesis tasks
[x_in1, x_in2, x_out] = model_task(inputs,self.opt.task_id)
x_fusion = torch.cat([x_in1,x_in2],dim=1)
if self.opt.use_gpu:
x_fusion = x_fusion.cuda()
# pred_out -- [batch_size*4,1,128,128]
# x3 -- [batch_size*4,1,128,128]
pred_out,pred_out1,pred_out2 = self.generator(x_fusion)
errors = prediction_syn_results(pred_out,x_out)
print(errors)
pred_eva_set.append([errors['MSE'],errors['SSIM'],errors['PSNR']])
mean_values = [ind_epoch,np.array(pred_eva_set)[:,0].mean(),np.array(pred_eva_set)[:,1].mean(),np.array(pred_eva_set)[:,2].mean(),np.array(pred_eva_set)[:,3].mean(),np.array(pred_eva_set)[:,4].mean(),np.array(pred_eva_set)[:,5].mean()]
return mean_values