-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathtrain.py
179 lines (141 loc) · 8.06 KB
/
train.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
# Last edit 06.07.2020
import torch
import torch.nn as nn
import Pix2PixGAN.Generator as pix2pixG
import Pix2PixGAN.Discriminator as pix2pixD
import Pix2PixGAN.Initialization as pix2pixInit
import Pix2PixGAN.GANLoss as pix2pixLoss
import functools
import Data.RGBDFaceDataset as rgbdDataset
#import Data.MelSpecDataset as rgbdDataset
import Utils.Visualization as Vis
from tqdm import tqdm
from os import path
from torchsummary import summary
from torch.utils.tensorboard import SummaryWriter
import onnx
import os
import configFile as config
import statistics
import math
if __name__ == '__main__':
if not os.path.exists("Data/" + config.DatasetName + "/Result/"):
os.mkdir("Data/" + config.DatasetName + "/Result/")
os.mkdir("Data/" + config.DatasetName + "/Result/Snaps/")
os.mkdir("Data/" + config.DatasetName + "/Result/Examples/")
os.mkdir("Data/" + config.DatasetName + "/Result/Weights/")
### Define Networks ###
netG = pix2pixG.UnetGenerator(input_nc= config.INPUT_CHANNEL, output_nc=4, num_downs=int(math.log2(config.IMAGE_SIZE)), ngf=64, norm_layer=functools.partial(nn.BatchNorm2d, affine=True, track_running_stats=True), use_dropout=True)
netG = pix2pixInit.init_net(netG, gpu_ids=[0])
summary(netG, (config.INPUT_CHANNEL, config.IMAGE_SIZE, config.IMAGE_SIZE))
netD = pix2pixD.NLayerDiscriminator(input_nc=config.INPUT_CHANNEL + 4, ndf=64, n_layers=int(math.log2(config.IMAGE_SIZE) - 5), norm_layer=functools.partial(nn.BatchNorm2d, affine=True, track_running_stats=True))
netD = pix2pixInit.init_net(netD, gpu_ids=[0])
summary(netD, (config.INPUT_CHANNEL + 4, config.IMAGE_SIZE, config.IMAGE_SIZE))
### Load Exsting Model State ###
if path.exists("Data/" + config.DatasetName + "/Result/trainedGenerator.pth") and path.exists("Data/" + config.DatasetName + "/Result/trainedDiscriminator.pth"):
netG.load_state_dict(torch.load("Data/" + config.DatasetName + "/Result/trainedGenerator.pth"))
netD.load_state_dict(torch.load("Data/" + config.DatasetName + "/Result/trainedDiscriminator.pth"))
startEpoch = int(input('Enter startEpoch:'))
print('startEpoch:', startEpoch, type(startEpoch))
else:
startEpoch = 1
### Tensorboard
comment = f'_dataset={config.DatasetName}'
print("TensorBoard:" + comment)
tb = SummaryWriter(comment=comment)
### Training settings ###
learningRate = 0.0002
lambda_L1 = 100
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
criterionGAN = pix2pixLoss.GANLoss("vanilla").to(device)
criterionL1 = torch.nn.L1Loss()
optimizer_G = torch.optim.Adam(netG.parameters(), lr=learningRate, betas=(0.5, 0.999))
optimizer_D = torch.optim.Adam(netD.parameters(), lr=learningRate, betas=(0.5, 0.999))
lambda_rule = lambda epoch: 1.0 - max(0, epoch + startEpoch - 100) / float(100 + 1)
scheduler_G = torch.optim.lr_scheduler.LambdaLR(optimizer_G, lr_lambda=lambda_rule)
scheduler_D = torch.optim.lr_scheduler.LambdaLR(optimizer_D, lr_lambda=lambda_rule)
### Training ###
dataset = rgbdDataset.RGBDFaceDataset(imageSize=config.IMAGE_SIZE, path="Data/" + config.DatasetName + "/")
dataset = torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=True, num_workers=0)
for epoch in range(startEpoch, 200 + 1):
print("Epoche: ", epoch, "(LearningRates:", scheduler_G.get_last_lr(), scheduler_D.get_last_lr(), ")")
epoch_loss_D = []
epoch_loss_G = []
for i, data in enumerate(tqdm(dataset)):
heatmap = data['Heatmap'].to(device)
realRGBD = data['RGBD'].to(device)
#Vis.showDatapair(realRGBD[0], heatmap[0])
fakeRGBD = netG(heatmap)
#Vis.showDatapair(fakeRGBD[0], heatmap[0])
### Update Discriminator ### similar to https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/master/models/pix2pix_model.py
optimizer_D.zero_grad() # set D's gradients to zero
"""Calculate GAN loss for the discriminator"""
# Fake; stop backprop to the generator by detaching fake_B
fake_AB = torch.cat((heatmap, fakeRGBD),1)
pred_fake = netD(fake_AB.detach())
loss_D_fake = criterionGAN(pred_fake, False)
# Real
real_AB = torch.cat((heatmap, realRGBD), 1)
pred_real = netD(real_AB)
loss_D_real = criterionGAN(pred_real, True)
# combine loss and calculate gradients
loss_D = (loss_D_fake + loss_D_real) * 0.5
loss_D.backward()
optimizer_D.step() # update D's weights
### Update Generator ### similar to https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/master/models/pix2pix_model.py
optimizer_G.zero_grad() # set G's gradients to zero
"""Calculate GAN and L1 loss for the generator"""
# First, G(A) should fake the discriminator
fake_AB = torch.cat((heatmap, fakeRGBD), 1)
with torch.no_grad():
pred_fake = netD(fake_AB)
loss_G_GAN = criterionGAN(pred_fake, True)
# Second, G(A) = B
loss_G_L1 = criterionL1(fakeRGBD, realRGBD) * lambda_L1
# combine loss and calculate gradients
loss_G = loss_G_GAN + loss_G_L1
loss_G.backward()
optimizer_G.step() # udpate G's weights
epoch_loss_D.append(loss_D.item())
epoch_loss_G.append(loss_G.item())
tb.add_scalars("Loss", {'Discriminator': statistics.mean(epoch_loss_D),
'Generator': statistics.mean(epoch_loss_G)}, epoch)
### Learning Rate Schedular
scheduler_G.step()
scheduler_D.step()
### Save Modell ###
torch.save(netG.cpu().state_dict(), "Data/" + config.DatasetName + "/Result/trainedGenerator.pth")
torch.save(netD.cpu().state_dict(), "Data/" + config.DatasetName + "/Result/trainedDiscriminator.pth")
if epoch % 10 == 0:
torch.save(netG.cpu().state_dict(), "Data/" + config.DatasetName + "/Result/Weights/trainedGenerator_epoch_" + str(epoch) + ".pth")
torch.save(netD.cpu().state_dict(), "Data/" + config.DatasetName + "/Result/Weights/trainedDiscriminator_epoch_" + str(epoch) + ".pth")
### Trace Modell ###
noise = torch.randn(heatmap.shape)
traced = torch.jit.trace(netG.eval(), noise)
netG.train().to(device)
netD.train().to(device)
traced.save("Data/" + config.DatasetName + "/Result/tracedGenerator.zip")
#print("LoadModel")
#loaded = torch.jit.load('trainedGenerator.zip')
#print(loaded)
#print(loaded.code)
#temp = loaded.forward(heatmap)
#Vis.showDatapair(temp[0],heatmap[0])
### Export Sample Image ###
if fakeRGBD[0].shape[0] == 4 and heatmap[0].shape[0] == 1:
Vis.exportExample(fakeRGBD[0], heatmap[0], "Data/" + config.DatasetName + "/Result/Examples/example_" + str(epoch) +".png")
### ONNX ####
x = torch.randn(1, config.INPUT_CHANNEL, 256, 256, requires_grad=True)
torch.onnx.export(netG.to("cpu"), # model being run
x, # model input (or a tuple for multiple inputs)
"Data/" + config.DatasetName + "/Result/tracedGenerator.onnx", # where to save the model (can be a file or file-like object)
export_params=True, # store the trained parameter weights inside the model file
opset_version=10, # the ONNX version to export the model to
do_constant_folding=True, # whether to execute constant folding for optimization
input_names=['input'], # the model's input names
output_names=['output'], # the model's output names
dynamic_axes={'input': {0: 'batch_size'}, # variable lenght axes
'output': {0: 'batch_size'}})
onnx_model = onnx.load("Data/" + config.DatasetName + "/Result/tracedGenerator.onnx")
print(onnx.checker.check_model(onnx_model))
##############