-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
177 lines (157 loc) · 7.08 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
import tensorflow as tf
import numpy as np
from tqdm import tqdm
import capslayer as cl
import os
import argparse
from dataloader import DataLoader
from model import DCCapsNet, CapsNet, DCCN2, DCCN3
from utils import LENGTH, calOA, selectData
from postProcess import TrainProcess, ProbMap
parser = argparse.ArgumentParser()
parser.add_argument("-e", "--epochs", default=50, type=int)
parser.add_argument("-b", "--batch_size", default=100, type=int)
parser.add_argument("-l", "--lr", default=0.001, type=float)
parser.add_argument("-g", "--gpu", default="0")
parser.add_argument("-r", "--ratio", default=0.1, type=float)
parser.add_argument("-a", "--aug", default=1, type=float)
parser.add_argument("-p", "--patch_size", default=9, type=int)
parser.add_argument("-m", "--model", default=1, type=int)
parser.add_argument("-d", "--directory", default="./save/default")
parser.add_argument("--model_path", default="./save/default/model")
parser.add_argument("--predict_only", action="store_true")
parser.add_argument("--restore", action="store_true")
parser.add_argument("--use_best_model", action="store_true")
parser.add_argument("--drop", default=1, type=float)
parser.add_argument("--data", default=0, type=int)
args = parser.parse_args()
print(args)
EPOCHS = args.epochs
LEARNING_RATE = args.lr
BATCH_SIZE = args.batch_size
RATIO = args.ratio
AUGMENT_RATIO = args.aug
PATCH_SIZE = args.patch_size
DROP_OUT = args.drop
DATA = args.data
DIRECTORY = args.directory
RESTORE = args.restore
PREDICT_ONLY = args.predict_only
MODEL_DIRECTORY = args.model_path
USE_BEST_MODEL = args.use_best_model
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
if not os.path.exists(DIRECTORY):
os.makedirs(os.path.join(DIRECTORY, "img"))
os.makedirs(os.path.join(DIRECTORY, "data"))
if not os.path.exists(MODEL_DIRECTORY):
os.makedirs(MODEL_DIRECTORY)
modelSavePath = os.path.join(MODEL_DIRECTORY, "model.ckpt")
imgSavePath = os.path.join(DIRECTORY, "img")
dataSavePath = os.path.join(DIRECTORY, "data")
pathName, matName = selectData(DATA)
dataloader = DataLoader(pathName, matName, PATCH_SIZE, RATIO, AUGMENT_RATIO)
trainPatch, trainSpectrum, trainLabel = dataloader.loadTrainData()
testPatch, testSpectrum, testLabel = dataloader.loadTestData()
allLabeledPatch, allLabeledSpectrum, allLabeledLabel, allLabeledIndex = dataloader.loadAllLabeledData()
w = tf.placeholder(shape=[None, dataloader.bands, 1], dtype=tf.float32)
x = tf.placeholder(shape=[None, dataloader.patchSize, dataloader.patchSize, dataloader.bands], dtype=tf.float32)
y = tf.placeholder(shape=[None, dataloader.numClasses], dtype=tf.float32)
k = tf.placeholder(dtype=tf.float32)
if args.model == 1:
pred = DCCapsNet(x, w, k, dataloader.numClasses)
print("USING DCCAPS***************************************")
elif args.model == 2:
pred = CapsNet(x, dataloader.numClasses)
print("USING CAPS*****************************************")
elif args.model == 3:
pred = DCCN2(x, w, k, dataloader.numClasses)
print("USING DCCN2****************************************")
else:
pred = DCCN3(x, w, k, dataloader.numClasses)
print("USING DCCN3****************************************")
pred = tf.divide(pred, tf.reduce_sum(pred, 1, keep_dims=True))
loss = tf.reduce_mean(cl.losses.margin_loss(y, pred))
optimizer = tf.train.AdamOptimizer(learning_rate=LEARNING_RATE).minimize(loss)
correctPredictions = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))
accuracy = tf.reduce_mean(tf.cast(correctPredictions, "float"))
init = tf.global_variables_initializer()
saver = tf.train.Saver()
with tf.Session() as sess:
leastLoss = 100.0
if RESTORE or PREDICT_ONLY:
saver.restore(sess, modelSavePath)
else:
sess.run(init)
if not PREDICT_ONLY:
trainProcess = TrainProcess(DIRECTORY)
for epoch in range(EPOCHS):
if epoch % 5 == 0:
permutation = np.random.permutation(trainPatch.shape[0])
trainPatch = trainPatch[permutation, :, :, :]
trainSpectrum = trainSpectrum[permutation, :]
trainLabel = trainLabel[permutation, :]
iter = dataloader.trainNum // BATCH_SIZE
with tqdm(total=iter, desc="epoch %3d" % (epoch + 1), ncols=LENGTH) as pbar:
for i in range(iter):
batch_w = trainSpectrum[i * BATCH_SIZE:(i + 1) * BATCH_SIZE, :, :]
batch_x = trainPatch[i * BATCH_SIZE:(i + 1) * BATCH_SIZE, :, :, :]
batch_y = trainLabel[i * BATCH_SIZE:(i + 1) * BATCH_SIZE, :]
_, batchLoss, trainAcc = sess.run([optimizer, loss, accuracy],
feed_dict={w: batch_w, x: batch_x, y: batch_y, k: DROP_OUT})
pbar.set_postfix_str(
"loss: %.6f, accuracy:%.2f, testLoss:-.---, testAcc:-.--" % (batchLoss, trainAcc))
pbar.update(1)
if i == 0 and epoch == 0:
idx = np.random.choice(dataloader.testNum, size=BATCH_SIZE, replace=False)
test_batch_w = testSpectrum[idx, :, :]
test_batch_x = testPatch[idx, :, :, :]
test_batch_y = testLabel[idx, :]
ac, ls = sess.run([accuracy, loss],
feed_dict={w: test_batch_w, x: test_batch_x, y: test_batch_y, k: 1})
trainProcess.addData(batchLoss, trainAcc, ls, ac)
if batchLoss < leastLoss:
saver.save(sess, save_path=modelSavePath)
leastLoss = batchLoss
if iter * BATCH_SIZE < dataloader.trainNum:
batch_w = trainSpectrum[iter * BATCH_SIZE:, :, :]
batch_x = trainPatch[iter * BATCH_SIZE:, :, :, :]
batch_y = trainLabel[iter * BATCH_SIZE:, :]
_, bl, ta = sess.run([optimizer, loss, accuracy],
feed_dict={w: batch_w, x: batch_x, y: batch_y, k: DROP_OUT})
idx = np.random.choice(dataloader.testNum, size=BATCH_SIZE, replace=False)
test_batch_w = testSpectrum[idx, :, :]
test_batch_x = testPatch[idx, :, :, :]
test_batch_y = testLabel[idx, :]
ac, ls = sess.run([accuracy, loss], feed_dict={w: test_batch_w, x: test_batch_x, y: test_batch_y, k: 1})
pbar.set_postfix_str(
"loss: %.6f, accuracy:%.2f, testLoss:%.3f, testAcc:%.2f" % (batchLoss, trainAcc, ls, ac))
trainProcess.addData(batchLoss, trainAcc, ls, ac)
if USE_BEST_MODEL:
saver.restore(sess, modelSavePath)
iter = dataloader.allLabeledNum // BATCH_SIZE
probMap = ProbMap(dataloader.numClasses, dataSavePath, allLabeledLabel, allLabeledIndex, dataloader.height,
dataloader.width, dataloader.trainIndex)
with tqdm(total=iter, desc="predicting...") as pbar:
for i in range(iter):
batch_w = allLabeledSpectrum[i * BATCH_SIZE:(i + 1) * BATCH_SIZE, :, :]
batch_x = allLabeledPatch[i * BATCH_SIZE:(i + 1) * BATCH_SIZE, :, :, :]
batch_y = allLabeledLabel[i * BATCH_SIZE:(i + 1) * BATCH_SIZE, :]
tmp = sess.run(pred, feed_dict={w: batch_w, x: batch_x, y: batch_y, k: 1})
probMap.addData(tmp)
pbar.update()
if iter * BATCH_SIZE < dataloader.allLabeledNum:
batch_w = allLabeledSpectrum[iter * BATCH_SIZE:, :, :]
batch_x = allLabeledPatch[iter * BATCH_SIZE:, :, :, :]
batch_y = allLabeledLabel[iter * BATCH_SIZE:, :]
tmp = sess.run(pred, feed_dict={w: batch_w, x: batch_x, y: batch_y, k: 1})
probMap.addData(tmp)
probMap.finish()
print(np.shape(probMap.map))
if not PREDICT_ONLY:
trainProcess.save()
probMap.save()
OA = calOA(probMap.map, allLabeledLabel)
print(OA)
with open(os.path.join(DIRECTORY, "summary.txt"), "w+") as f:
print("OA:", OA, file=f)
print(args, file=f)