-
Notifications
You must be signed in to change notification settings - Fork 0
/
TIANCHI-Training.py
127 lines (113 loc) · 6.34 KB
/
TIANCHI-Training.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
#!/usr/bin/env python
# coding: utf-8
from model.UVOS import UVOS
from utils.DataLoading import DataDiscovering, DataReading
from utils.tensorboard_evaluation import Evaluation
import torch.optim as optim
from torch.utils.data import DataLoader
import torchvision.transforms as transforms
import torch
import torch.nn as nn
from itertools import combinations
from tqdm import tqdm
import os
import argparse
import numpy as np
device = torch.device("cuda")
def train(root, num_epoch, Batch_Size, Random_Subset = None, learning_rate = 2.5e-4, model_dir="./saved_models", tensorboard_dir="./tensorboard"):
# create result and model folders
if not os.path.exists(model_dir):
os.mkdir(model_dir)
if not os.path.exists(tensorboard_dir):
os.mkdir(tensorboard_dir)
#tensorboard_Loss = Evaluation(tensorboard_dir,name = "Loss",stats=["val_Loss", "train_Loss"])
#tensorboard_Accuracy = Evaluation(tensorboard_dir,name = "Accuracy",stats=["val_accuracy"])
Dir = DataDiscovering(root)
transform = transforms.Compose([transforms.ToTensor()])
train_data, val_data = Dir.reading_training_data(train_val_split=0.7,random_subset=Random_Subset)
trainset = DataReading(train_data,transform = transform)
valset = DataReading(val_data,transform = transform)
#DL_network = DeepLab().cuda()
network = UVOS().cuda()
optimizer = optim.SGD( network.parameters(), lr=2.5e-4,momentum=0.9, weight_decay=1e-4 )
loss_fn = nn.L1Loss()
#for epoch in tqdm(range(num_epoch),ascii=True,desc="epoch -> "):
for epoch in range(num_epoch):
print("-----------------------------epoch {}-----------------------------".format(epoch))
train_length = len(list(train_data.keys()))
train_loss = 0
for item in range(train_length):
print("----- Reading training data -----")
training_data = list(combinations(trainset[item],2))
length = len(training_data)
if length%Batch_Size != 0:
Length = int(np.floor(length/Batch_Size)*Batch_Size)
train_loader = DataLoader(training_data[:Length],batch_size=Batch_Size,shuffle=False,num_workers=4)
else:
train_loader = DataLoader(training_data,batch_size=Batch_Size,shuffle=False,num_workers=4)
print("----- the total batches = {} -----".format(len(train_loader)))
print("----- Fininsh Reading training data -----")
folder_loss = 0
count = 0
#tensorboard_training_folder_Loss = Evaluation(tensorboard_dir,name = "training_the_folder_{}".format(list(train_data.keys())[item]),stats=["folder_train_Loss"])
for batch in train_loader:
Fa = batch[0][0]
Fb = batch[1][0]
Fa_label = batch[0][1].squeeze()
Fb_label = batch[1][1].squeeze()
optimizer.zero_grad()
pred_a, pred_b = network(Fa.to(device),Fb.to(device))
Loss = loss_fn(pred_a,Fa_label.to(device)) + loss_fn(pred_b, Fb_label.to(device))
del Fa, Fb, Fa_label, Fb_label
Loss.backward()
optimizer.step()
folder_loss += Loss
#tensorboard_training_folder_Loss.write_episode_data(count,{"folder_train_Loss":Loss.cpu().detach()})
count += 1
train_loss += folder_loss/count
del training_data, train_loader
print("-----training stage folder loss = {}-----".format(folder_loss/count))
train_loss = train_loss / train_length
torch.save(network.state_dict(), os.path.join(model_dir+'/'+ "train_saved_model_at_epoch_{}_loss={}.pth".format(epoch,train_loss)))
print("-----training stage folder loss = {}-----".format(epoch,train_loss))
#tensorboard_Loss.write_episode_data(epoch,{"train_Loss":train_loss.cpu().detach()})
with torch.no_grad():
val_length = len(list(val_data.keys()))
val_loss = 0
val_accuracy = 0
for item in range(val_length):
print("----- Reading validation data -----")
valid_data = list(combinations(valset[item],2))
length = len(valid_data)
if length%Batch_Size != 0:
Length = int(np.floor(length/Batch_Size)*Batch_Size)
val_loader = DataLoader(valid_data[:Length],batch_size=Batch_Size,shuffle=False,num_workers=4)
else:
val_loader = DataLoader(valid_data,batch_size=Batch_Size,shuffle=False,num_workers=4)
print("----- the total batches = {} -----".format(len(val_loader)))
print("----- Fininsh Reading validation data -----")
folder_loss = 0
count = 0
#tensorboard_val_folder_Loss = Evaluation(tensorboard_dir,name = "val_the_folder_{}".format(list(val_data.keys())[item]),stats=["folder_val_Loss"])
for batch in val_loader:
Fa = batch[0][0]
Fb = batch[1][0]
Fa_label = batch[0][1].squeeze()
Fb_label = batch[1][1].squeeze()
pred_a, pred_b = network(Fa.to(device),Fb.to(device))
Loss = loss_fn(pred_a,Fa_label.to(device)) + loss_fn(pred_b, Fb_label.to(device))
del Fa, Fb, Fa_label, Fb_label
#tensorboard_val_folder_Loss.write_episode_data(count,{"folder_val_Loss":Loss.cpu().detach()})
folder_loss += Loss
count += 1
#tensorboard_val_folder_Loss.save("val_the_folder_{}".format(list(val_data.keys())[item]))
val_loss += folder_loss/count
del valid_data, val_loader
print("-----val stagit pullge folder loss = {}-----".format(folder_loss/count))
val_loss = val_loss / val_length
print("-----epoch {}-- val stage folder loss = {}-----".format(epoch,val_loss))
#if val_loss <= 0.2 :
torch.save(network.state_dict(), os.path.join(model_dir +'/'+"val_saved_model_at_epoch_{}_loss_{}.pth".format(epoch,val_loss)))
print("-----Model saved in file-----")
#tensorboard_Loss.write_episode_data(epoch,{"val_Loss":val_loss.cpu().detach()})
train(root = 'E:/TIANCHI', num_epoch = 10, Batch_Size=2,Random_Subset = 6)