-
Notifications
You must be signed in to change notification settings - Fork 2
/
trainSGML.py
168 lines (146 loc) · 6.36 KB
/
trainSGML.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
# -*- coding: utf-8 -*-
import numpy as np
from GCNModel3 import GCNModel3
from BuildSPInst_A import *
import tensorflow as tf
import time
import os
os.environ['CUDA_VISIBLE_DEVICES']='1'
iter=0
for iter in range(1):
time_start=time.time()
def GCNevaluate(mask1, labels1):
t_test = time.time()
outs_val = sess.run([GCNmodel.loss, GCNmodel.accuracy], feed_dict={labels: labels1, mask: mask1})
return outs_val[0], outs_val[1], (time.time() - t_test)
data_name = 'IP'
num_classes = 16
learning_rate = 0.0005
epochs=500
img_gyh = data_name+'_gyh'
img_gt = data_name+'_gt'
Data = load_HSI_data(data_name)
model = GetInst_A(Data['useful_sp_lab'], Data[img_gyh], Data[img_gt], Data['trpos'])
def get_mask(gt, poses, sets):
pixel_mask_all=np.zeros(gt.shape, dtype='bool')
if sets=='all':
pixel_mask_all[gt>0]=True
gt_tr_te=gt[pixel_mask_all]
else:
pixel_mask_all[poses[:,0]-1,poses[:,1]-1]=True
pixel_mask=pixel_mask_all[gt>0]
return pixel_mask
# create gt_tr_te
pixel_mask_tr_te=np.zeros(Data[img_gt].shape, dtype='bool')
pixel_mask_tr_te[Data[img_gt]>0]=True
gt_tr_te=Data[img_gt][pixel_mask_tr_te]
# create pixel_mask_tr gt_nonzeros
pixel_mask_tr_te*=False
pixel_mask_tr_te[Data['trpos'][:,0]-1,Data['trpos'][:,1]-1]=True
pixel_mask_tr=pixel_mask_tr_te[Data[img_gt]>0]
pixel_mask_val=get_mask(Data[img_gt],Data['valpos'],'val')
pixel_mask_te=~(pixel_mask_tr^pixel_mask_val)
gt_nonzeros=(Data[img_gt])[Data[img_gt]>0]
#gt_nonzeros_tr=gt_nonzeros[pixel_mask_tr]
gt_1hot = np.zeros([pixel_mask_tr.shape[0], num_classes]) # one-hot coding
for row_idx in range(gt_1hot.shape[0]):
col_idx = int(gt_nonzeros[row_idx])-1
gt_1hot[row_idx, col_idx] = 1
gt_1hot_tr=np.array(gt_1hot)
gt_1hot_tr[pixel_mask_te^pixel_mask_val]*=False
gt_1hot_val=np.array(gt_1hot)
gt_1hot_val[pixel_mask_te^pixel_mask_tr]*=False
gt_1hot_te=np.array(gt_1hot)
gt_1hot_te[pixel_mask_tr^pixel_mask_val]*=False
pixel_mask_tr=np.expand_dims(pixel_mask_tr, axis=1)
pixel_mask_val=np.expand_dims(pixel_mask_val, axis=1)
pixel_mask_te=np.expand_dims(pixel_mask_te, axis=1)
useful_sp_lab = [np.array(x, dtype='int32') for x in model.useful_sp_lab]
sp_mean = [np.array(x, dtype='float32') for x in model.sp_mean]
sp_label = [ np.array(x, dtype='float32') for x in model.sp_label]
sp_support = []
#sp_support=list(scio.loadmat('data//'+data_name+'//result//support.mat')['sp_support'])
for A_x in model.sp_A:
sp_A = np.array(A_x, dtype='float32')
sp_support.append(np.array(model.CalSupport(sp_A), dtype='float32'))
############################################
mask = tf.placeholder("int32", [None, 1])
labels = tf.placeholder("float", [None, num_classes])
# Normalize sp_mean
#sp_mean /= sp_mean.sum(1).reshape(-1, 1)
seed=123
np.random.seed(seed)
tf.set_random_seed(seed)
# 构建proposed model
GCNmodel = GCNModel3(features = sp_mean, labels = labels,l=Data[img_gt],idx=useful_sp_lab, learning_rate = learning_rate,
num_classes = num_classes, mask = mask, support = sp_support, scale_num = 3, h = 32)
sess=tf.Session()
sess.run(tf.global_variables_initializer())
train_t = time.time()
for epoch in range(epochs):
# Training step
outs = sess.run([GCNmodel.opt_op, GCNmodel.loss, GCNmodel.accuracy], feed_dict={ labels:gt_1hot_tr,
mask:pixel_mask_tr })
outsval = sess.run([GCNmodel.loss, GCNmodel.accuracy], feed_dict={ labels:gt_1hot_val,
mask:pixel_mask_val })
print("Epoch:", '%04d' % (epoch + 1),
"train_loss=", "{:.2f}".format(outs[1]),
"train_acc=", "{:.2f}".format(outs[2]),
"val_loss=", "{:.2f}".format(outsval[0]),
"val_acc=", "{:.2f}".format(outsval[1]))
# print("Epoch:", '%04d' % (epoch + 1), "val_loss=", "{:.5f}".format(outsval[0]),
# "val_acc=", "{:.5f}".format(outsval[1]))
print("Optimization Finished!")
training_time = time.time() - train_t
print("Training time =", str(time.time() - train_t))
# Testing
test_cost, test_acc, test_duration = GCNevaluate(pixel_mask_te, gt_1hot_te)
print("Test set results:", "cost=", "{:.5f}".format(test_cost),
"accuracy=", "{:.5f}".format(test_acc), "time=", "{:.5f}".format(test_duration))
#######
pred_map=np.zeros_like(Data[img_gt])
#superPixel-wise accuracy
outputs = sess.run(GCNmodel.outputs)
predictions=np.argmax(outputs, axis=1)+1
pred_map[Data[img_gt]>0]=predictions
#######
resultpath='data//'+data_name+'//result//'
matrix = np.zeros((num_classes, num_classes))
outputs_decode = np.argmax(outputs[pixel_mask_te[:,0]], 1)
test_labels = np.argmax(gt_1hot_te[pixel_mask_te[:,0]],1)
n = pixel_mask_te.sum()
with open(resultpath+'prediction.txt', 'w') as f:
for i in range(n):
pre_label = int(outputs_decode[i])
f.write(str(pre_label)+'\n')
matrix[pre_label][test_labels[i]] += 1
np.savetxt(resultpath+'result_matrix.txt', matrix, fmt='%d', delimiter=',')
print(''+str(np.int_(matrix)))
print(np.sum(np.trace(matrix)))
# print('OA = '+str(OA)+'\n')
ua = np.diag(matrix)/np.sum(matrix, axis=0)
AA=np.sum(ua)/matrix.shape[0]
precision = np.diag(matrix)/np.sum(matrix, axis=1)
matrix = np.mat(matrix)
OA = np.sum(np.trace(matrix)) / float(n)
Po = OA
xsum = np.sum(matrix, axis=1)
ysum = np.sum(matrix, axis=0)
Pe = float(ysum*xsum)/(np.sum(matrix)**2)
Kappa = float((Po-Pe)/(1-Pe))
AP=np.sum(precision)/matrix.shape[0]
# print('ua =')
for i in range(num_classes):
print(ua[i])
print(AA)
print(OA)
print(Kappa)
print()
for i in range(num_classes):
print(precision[i])
print(AP)
f.close()
#################################################
scio.savemat('data/'+data_name+'/result/pred.mat',{'pred':pred_map})
scio.savemat('data/'+data_name+'/result/support.mat',{'sp_support':sp_support})
iter=iter+1