-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathunet_train.py
229 lines (156 loc) · 6.46 KB
/
unet_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
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
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
import os
import monai
import torch
import torch.nn as nn
import torch.optim as optim
from monai.networks.nets import UNet
from monai.data import DataLoader, CacheDataset, ImageDataset, Dataset, image_reader
from monai.data.utils import pad_list_data_collate
from monai.transforms import Compose, ScaleIntensity, ToTensor,LoadImaged, ScaleIntensityd, ToTensord,EnsureChannelFirstd
import SimpleITK as sitk
from monai.data.image_reader import ITKReader
from monai.losses import DiceLoss, DiceCELoss, DiceFocalLoss, FocalLoss, GeneralizedDiceLoss
from monai.metrics import MSEMetric
import dataset
from dataset import PelvisDataset
import numpy as np
import sklearn
from monai.metrics import compute_hausdorff_distance
from UNETR import UNETR
import ipywidgets as widgets
from IPython.display import display
import time
import torch.nn.functional as F
# train
root_dir ="/home/paulagmtz/TFM_PAULA_24/DATA_TFM/train_files"
input_dim = (194,148,105)
n_labels= 6
train_dataset = PelvisDataset(root_dir,input_dim, n_labels, transform=None)
train_dataloader = DataLoader(train_dataset, collate_fn=pad_list_data_collate,batch_size =2)
#val
# root_dir ="/home/paulagmtz/TFM_PAULA_24/DATA_TFM/val_files"
# input_dim = (194,148,105)
# n_labels= 2
# val_dataset = PelvisDataset(root_dir,input_dim, n_labels, transform=None)
# val_dataloader = DataLoader(val_dataset, batch_size =2)
# #test
# root_dir ="/home/paulagmtz/TFM_PAULA_24/DATA_TFM/test_files"
# input_dim = (194,148,105)
# n_labels= 2
# test_dataset = PelvisDataset(root_dir,input_dim, n_labels, transform=None)
# test_dataloader = DataLoader(test_dataset, batch_size =2)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = UNet(
spatial_dims=3 ,
in_channels=1,
out_channels=7,
channels=(64,128,256,512),
strides=(2,2,2),
# num_res_units=2,
# norm=Norm.BATCH,
).to(device)
# Unetr model
# model = UNETR(in_channels=1, out_channels=1, img_size=(112, 160, 208), feature_size=32, norm_name='batch')
# model.to(device)
# loss and optimizer
dice_loss_function = DiceLoss(sigmoid=True, squared_pred=True)
diceCE_loss_function = DiceCELoss(sigmoid=True)
dicefocall_loss_function = DiceFocalLoss(sigmoid=True)
generalizedDice_loss_function = GeneralizedDiceLoss(sigmoid=True)
optimizer = optim.Adam(model.parameters(), lr=1e-3)
# Función para mostrar imágenes de inputs, targets y outputs
def plot_results(inputs, targets, outputs):
depth_slice = inputs[:, :, :, :, :]
depth_slice_target = targets[:, :, :, :, :]
depth_slice_outputs = outputs[:, :, :, :, :]
def plot_image(i):
fig, axes = plt.subplots(1, 3, figsize=(10, 5))
image_array = depth_slice[0, :, i, :, :].cpu().detach().numpy()
image_array_target = depth_slice_target[0, :, i, :, :].cpu().detach().numpy()
image_array_outputs = depth_slice_outputs[0, :, i, :, :].cpu().detach().numpy()
axes[0].imshow(image_array[0, :, :], cmap="gray")
axes[0].set_title("inputs")
axes[0].axis('off')
axes[1].imshow(image_array_target[0, :, :], cmap="gray")
axes[1].set_title("target")
axes[1].axis('off')
axes[2].imshow(image_array_outputs[0, :, :], cmap="gray")
axes[2].set_title("output")
axes[2].axis('off')
plt.show()
widgets.interact(plot_image, i=widgets.IntSlider(min=0, max=depth_slice.shape[2]-1, step=1, value=22))
# Train model
num_epochs = 300
# num_epochs =50
loss_evolution = []
# Guarda el tiempo de inicio
inicio = time.time()
loss_evolution_train = []
count = 0
for epoch in range(num_epochs):
model.train()
for batch in train_dataloader:
inputs, targets = batch[0].to(device), batch[1].to(device)
optimizer.zero_grad()
outputs = model(inputs)
out = nn.functional.sigmoid(outputs)
# out = nn.functional.softmax(outputs, dim = 0)
# one_hot_targets = F.one_hot(targets, num_classes=6)
loss = dicefocall_loss_function(outputs, targets)
loss.backward()
optimizer.step()
# plot_results(inputs,targets,outputs)
loss_evolution_train.append(loss)
# if count == 50:
# plot_results(inputs,targets,out)
print(f"Epoch [{epoch + 1}/{num_epochs}] - Loss: {loss.item()}")
# count = 0
# count = count + 1
print("Training complete.")
# Guarda el tiempo de finalización
fin = time.time()
# Calcula la diferencia
tiempo_total = fin - inicio
print(f"Tiempo de ejecución: {tiempo_total} segundos")
torch.save(model.state_dict(), "/home/paulagmtz/TFM_PAULA_24/MONAI-UNET/UNET_TRUE_MULTILABEL_1_7_model_200_dicefocall_loss.pth")
# def euclidean_distance(predictions, targets):
# print(predictions)
# print(targets)
# print(predictions.shape)
# print(targets.shape)
# # Aplanar los tensores
# predictions_flat = predictions.view(2, -1)
# targets_flat = targets.view(2, -1)
# dist = torch.norm(targets_flat-predictions_flat,dim=1)
# return dist
# def validate_model(model, val_dataloader, device):
# model.eval() # Modo de evaluación
# # Definir la función de pérdida
# loss_function = DiceLoss(sigmoid=True, squared_pred=True)
# # Variables para almacenar la evolución de la métrica
# metric_history = []
# loss_evolution = []
# with torch.no_grad():
# for batch in val_dataloader:
# inputs, targets = batch[0].to(device), batch[1].to(device)
# # Realizar predicciones
# outputs = model(inputs)
# one_hot_targets = F.one_hot(targets, num_clases=6)
# # Calcular la pérdida
# loss = loss_function(outputs, targets)
# # caluclar la distancia euclidea
# dist = euclidean_distance(outputs, targets)
# # Guardar la métrica
# loss_evolution.append(loss.item())
# metric_history.append(dist)
# # Mostrar imágenes de inputs, targets y outputs
# plot_results(inputs, targets, outputs)
# # Calcular la métrica promedio
# average_metric = sum(metric_history) / len(metric_history)
# print(f'Average Metric: {average_metric}')
# # Guardar la evolución de la métrica en un archivo (por ejemplo, un archivo de texto)
# with open("metric_history.txt", "w") as f:
# for metric in metric_history:
# f.write(f"{metric}\n")
# # Realizar la validación
# validate_model(model, val_dataloader, device)