-
Notifications
You must be signed in to change notification settings - Fork 37
/
A2S2KResNet.py
800 lines (680 loc) · 25 KB
/
A2S2KResNet.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
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
#!/usr/bin/env python
# coding: utf-8
# # Imports
import argparse
import collections
import math
import time
import numpy as np
import scipy.io as sio
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from sklearn import metrics, preprocessing
from sklearn.decomposition import PCA
from sklearn.metrics import confusion_matrix
import geniter
import record
import torch_optimizer as optim2
import Utils
from torchsummary import summary
# # Setting Params
parser = argparse.ArgumentParser(description='Training for HSI')
parser.add_argument(
'-d', '--dataset', dest='dataset', default='IN', help="Name of dataset.")
parser.add_argument(
'-o',
'--optimizer',
dest='optimizer',
default='adam',
help="Name of optimizer.")
parser.add_argument(
'-e', '--epoch', type=int, dest='epoch', default=200, help="No of epoch")
parser.add_argument(
'-i', '--iter', type=int, dest='iter', default=3, help="No of iter")
parser.add_argument(
'-p', '--patch', type=int, dest='patch', default=4, help="Length of patch")
parser.add_argument(
'-k',
'--kernel',
type=int,
dest='kernel',
default=24,
help="Length of kernel")
parser.add_argument(
'-vs',
'--valid_split',
type=float,
dest='valid_split',
default=0.9,
help="Percentage of validation split.")
args = parser.parse_args()
PARAM_DATASET = args.dataset # UP,IN,SV, KSC
PARAM_EPOCH = args.epoch
PARAM_ITER = args.iter
PATCH_SIZE = args.patch
PARAM_VAL = args.valid_split
PARAM_OPTIM = args.optimizer
PARAM_KERNEL_SIZE = args.kernel
# # Data Loading
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# for Monte Carlo runs
seeds = [1331, 1332, 1333, 1334, 1335, 1336, 1337, 1338, 1339, 1340, 1341]
ensemble = 1
global Dataset # UP,IN,SV, KSC
dataset = PARAM_DATASET # input('Please input the name of Dataset(IN, UP, SV, KSC):')
Dataset = dataset.upper()
def load_dataset(Dataset, split=0.9):
data_path = '../dataset/'
if Dataset == 'IN':
mat_data = sio.loadmat(data_path + 'Indian_pines_corrected.mat')
mat_gt = sio.loadmat(data_path + 'Indian_pines_gt.mat')
data_hsi = mat_data['indian_pines_corrected']
gt_hsi = mat_gt['indian_pines_gt']
K = 200
TOTAL_SIZE = 10249
VALIDATION_SPLIT = split
TRAIN_SIZE = math.ceil(TOTAL_SIZE * VALIDATION_SPLIT)
if Dataset == 'UP':
uPavia = sio.loadmat(data_path + 'PaviaU.mat')
gt_uPavia = sio.loadmat(data_path + 'PaviaU_gt.mat')
data_hsi = uPavia['paviaU']
gt_hsi = gt_uPavia['paviaU_gt']
K = 103
TOTAL_SIZE = 42776
VALIDATION_SPLIT = split
TRAIN_SIZE = math.ceil(TOTAL_SIZE * VALIDATION_SPLIT)
if Dataset == 'SV':
SV = sio.loadmat(data_path + 'Salinas_corrected.mat')
gt_SV = sio.loadmat(data_path + 'Salinas_gt.mat')
data_hsi = SV['salinas_corrected']
gt_hsi = gt_SV['salinas_gt']
K = 15
TOTAL_SIZE = 54129
VALIDATION_SPLIT = split
TRAIN_SIZE = math.ceil(TOTAL_SIZE * VALIDATION_SPLIT)
if Dataset == 'KSC':
SV = sio.loadmat(data_path + 'KSC.mat')
gt_SV = sio.loadmat(data_path + 'KSC_gt.mat')
data_hsi = SV['KSC']
gt_hsi = gt_SV['KSC_gt']
K = data_hsi.shape[2]
TOTAL_SIZE = 5211
VALIDATION_SPLIT = split
TRAIN_SIZE = math.ceil(TOTAL_SIZE * VALIDATION_SPLIT)
shapeor = data_hsi.shape
data_hsi = data_hsi.reshape(-1, data_hsi.shape[-1])
data_hsi = PCA(n_components=K).fit_transform(data_hsi)
shapeor = np.array(shapeor)
shapeor[-1] = K
data_hsi = data_hsi.reshape(shapeor)
return data_hsi, gt_hsi, TOTAL_SIZE, TRAIN_SIZE, VALIDATION_SPLIT
# # Pytorch Data Loader Creation
data_hsi, gt_hsi, TOTAL_SIZE, TRAIN_SIZE, VALIDATION_SPLIT = load_dataset(
Dataset, PARAM_VAL)
print(data_hsi.shape)
image_x, image_y, BAND = data_hsi.shape
data = data_hsi.reshape(
np.prod(data_hsi.shape[:2]), np.prod(data_hsi.shape[2:]))
gt = gt_hsi.reshape(np.prod(gt_hsi.shape[:2]), )
CLASSES_NUM = max(gt)
print('The class numbers of the HSI data is:', CLASSES_NUM)
print('-----Importing Setting Parameters-----')
ITER = PARAM_ITER
PATCH_LENGTH = PATCH_SIZE
lr, num_epochs, batch_size = 0.001, 200, 32
loss = torch.nn.CrossEntropyLoss()
img_rows = 2 * PATCH_LENGTH + 1
img_cols = 2 * PATCH_LENGTH + 1
img_channels = data_hsi.shape[2]
INPUT_DIMENSION = data_hsi.shape[2]
ALL_SIZE = data_hsi.shape[0] * data_hsi.shape[1]
VAL_SIZE = int(TRAIN_SIZE)
TEST_SIZE = TOTAL_SIZE - TRAIN_SIZE
KAPPA = []
OA = []
AA = []
TRAINING_TIME = []
TESTING_TIME = []
ELEMENT_ACC = np.zeros((ITER, CLASSES_NUM))
data = preprocessing.scale(data)
data_ = data.reshape(data_hsi.shape[0], data_hsi.shape[1], data_hsi.shape[2])
whole_data = data_
padded_data = np.lib.pad(
whole_data, ((PATCH_LENGTH, PATCH_LENGTH), (PATCH_LENGTH, PATCH_LENGTH),
(0, 0)),
'constant',
constant_values=0)
# # Model
class ChannelSELayer3D(nn.Module):
"""
3D extension of Squeeze-and-Excitation (SE) block described in:
*Hu et al., Squeeze-and-Excitation Networks, arXiv:1709.01507*
*Zhu et al., AnatomyNet, arXiv:arXiv:1808.05238*
"""
def __init__(self, num_channels, reduction_ratio=2):
"""
:param num_channels: No of input channels
:param reduction_ratio: By how much should the num_channels should be reduced
"""
super(ChannelSELayer3D, self).__init__()
self.avg_pool = nn.AdaptiveAvgPool3d(1)
num_channels_reduced = num_channels // reduction_ratio
self.reduction_ratio = reduction_ratio
self.fc1 = nn.Linear(num_channels, num_channels_reduced, bias=True)
self.fc2 = nn.Linear(num_channels_reduced, num_channels, bias=True)
self.relu = nn.ReLU()
self.sigmoid = nn.Sigmoid()
def forward(self, input_tensor):
"""
:param input_tensor: X, shape = (batch_size, num_channels, D, H, W)
:return: output tensor
"""
batch_size, num_channels, D, H, W = input_tensor.size()
# Average along each channel
squeeze_tensor = self.avg_pool(input_tensor)
# channel excitation
fc_out_1 = self.relu(
self.fc1(squeeze_tensor.view(batch_size, num_channels)))
fc_out_2 = self.sigmoid(self.fc2(fc_out_1))
output_tensor = torch.mul(
input_tensor, fc_out_2.view(batch_size, num_channels, 1, 1, 1))
return output_tensor
class SpatialSELayer3D(nn.Module):
"""
3D extension of SE block -- squeezing spatially and exciting channel-wise described in:
*Roy et al., Concurrent Spatial and Channel Squeeze & Excitation in Fully Convolutional Networks, MICCAI 2018*
"""
def __init__(self, num_channels):
"""
:param num_channels: No of input channels
"""
super(SpatialSELayer3D, self).__init__()
self.conv = nn.Conv3d(num_channels, 1, 1)
self.sigmoid = nn.Sigmoid()
def forward(self, input_tensor, weights=None):
"""
:param weights: weights for few shot learning
:param input_tensor: X, shape = (batch_size, num_channels, D, H, W)
:return: output_tensor
"""
# channel squeeze
batch_size, channel, D, H, W = input_tensor.size()
if weights:
weights = weights.view(1, channel, 1, 1)
out = F.conv2d(input_tensor, weights)
else:
out = self.conv(input_tensor)
squeeze_tensor = self.sigmoid(out)
# spatial excitation
output_tensor = torch.mul(input_tensor,
squeeze_tensor.view(batch_size, 1, D, H, W))
return output_tensor
class ChannelSpatialSELayer3D(nn.Module):
"""
3D extension of concurrent spatial and channel squeeze & excitation:
*Roy et al., Concurrent Spatial and Channel Squeeze & Excitation in Fully Convolutional Networks, arXiv:1803.02579*
"""
def __init__(self, num_channels, reduction_ratio=2):
"""
:param num_channels: No of input channels
:param reduction_ratio: By how much should the num_channels should be reduced
"""
super(ChannelSpatialSELayer3D, self).__init__()
self.cSE = ChannelSELayer3D(num_channels, reduction_ratio)
self.sSE = SpatialSELayer3D(num_channels)
def forward(self, input_tensor):
"""
:param input_tensor: X, shape = (batch_size, num_channels, D, H, W)
:return: output_tensor
"""
output_tensor = torch.max(
self.cSE(input_tensor), self.sSE(input_tensor))
return output_tensor
class ProjectExciteLayer(nn.Module):
"""
Project & Excite Module, specifically designed for 3D inputs
*quote*
"""
def __init__(self, num_channels, reduction_ratio=2):
"""
:param num_channels: No of input channels
:param reduction_ratio: By how much should the num_channels should be reduced
"""
super(ProjectExciteLayer, self).__init__()
num_channels_reduced = num_channels // reduction_ratio
self.reduction_ratio = reduction_ratio
self.relu = nn.ReLU()
self.conv_c = nn.Conv3d(
in_channels=num_channels,
out_channels=num_channels_reduced,
kernel_size=1,
stride=1)
self.conv_cT = nn.Conv3d(
in_channels=num_channels_reduced,
out_channels=num_channels,
kernel_size=1,
stride=1)
self.sigmoid = nn.Sigmoid()
def forward(self, input_tensor):
"""
:param input_tensor: X, shape = (batch_size, num_channels, D, H, W)
:return: output tensor
"""
batch_size, num_channels, D, H, W = input_tensor.size()
# Average along channels and different axes
squeeze_tensor_w = F.adaptive_avg_pool3d(input_tensor, (1, 1, W))
squeeze_tensor_h = F.adaptive_avg_pool3d(input_tensor, (1, H, 1))
squeeze_tensor_d = F.adaptive_avg_pool3d(input_tensor, (D, 1, 1))
# tile tensors to original size and add:
final_squeeze_tensor = sum([
squeeze_tensor_w.view(batch_size, num_channels, 1, 1, W),
squeeze_tensor_h.view(batch_size, num_channels, 1, H, 1),
squeeze_tensor_d.view(batch_size, num_channels, D, 1, 1)
])
# Excitation:
final_squeeze_tensor = self.sigmoid(
self.conv_cT(self.relu(self.conv_c(final_squeeze_tensor))))
output_tensor = torch.mul(input_tensor, final_squeeze_tensor)
return output_tensor
class eca_layer(nn.Module):
"""Constructs a ECA module.
Args:
channel: Number of channels of the input feature map
k_size: Adaptive selection of kernel size
"""
def __init__(self, channel, k_size=3):
super(eca_layer, self).__init__()
self.avg_pool = nn.AdaptiveAvgPool3d(1)
self.conv = nn.Conv2d(
1, 1, kernel_size=k_size, padding=(k_size - 1) // 2, bias=False)
self.sigmoid = nn.Sigmoid()
def forward(self, x):
# x: input features with shape [b, c, h, w]
b, c, h, w, t = x.size()
# feature descriptor on the global spatial information
# 24, 1, 1, 1
y = self.avg_pool(x)
# Two different branches of ECA module
y = self.conv(y.squeeze(-1).transpose(-1, -3)).transpose(
-1, -3).unsqueeze(-1)
# Multi-scale information fusion
y = self.sigmoid(y)
return x * y.expand_as(x)
class Residual(nn.Module): # pytorch
def __init__(
self,
in_channels,
out_channels,
kernel_size,
padding,
use_1x1conv=False,
stride=1,
start_block=False,
end_block=False,
):
super(Residual, self).__init__()
self.conv1 = nn.Sequential(
nn.Conv3d(
in_channels,
out_channels,
kernel_size=kernel_size,
padding=padding,
stride=stride), nn.ReLU())
self.conv2 = nn.Conv3d(
out_channels,
out_channels,
kernel_size=kernel_size,
padding=padding,
stride=stride)
if use_1x1conv:
self.conv3 = nn.Conv3d(
in_channels, out_channels, kernel_size=1, stride=stride)
else:
self.conv3 = None
if not start_block:
self.bn0 = nn.BatchNorm3d(in_channels)
self.bn1 = nn.BatchNorm3d(out_channels)
self.bn2 = nn.BatchNorm3d(out_channels)
if start_block:
self.bn2 = nn.BatchNorm3d(out_channels)
if end_block:
self.bn2 = nn.BatchNorm3d(out_channels)
# ECA Attention Layer
self.ecalayer = eca_layer(out_channels)
# start and end block initialization
self.start_block = start_block
self.end_block = end_block
def forward(self, X):
identity = X
if self.start_block:
out = self.conv1(X)
else:
out = self.bn0(X)
out = F.relu(out)
out = self.conv1(out)
out = self.bn1(out)
out = F.relu(out)
out = self.conv2(out)
if self.start_block:
out = self.bn2(out)
out = self.ecalayer(out)
out += identity
if self.end_block:
out = self.bn2(out)
out = F.relu(out)
return out
class S3KAIResNet(nn.Module):
def __init__(self, band, classes, reduction):
super(S3KAIResNet, self).__init__()
self.name = 'SSRN'
self.conv1x1 = nn.Conv3d(
in_channels=1,
out_channels=PARAM_KERNEL_SIZE,
kernel_size=(1, 1, 7),
stride=(1, 1, 2),
padding=0)
self.conv3x3 = nn.Conv3d(
in_channels=1,
out_channels=PARAM_KERNEL_SIZE,
kernel_size=(3, 3, 7),
stride=(1, 1, 2),
padding=(1, 1, 0))
self.batch_norm1x1 = nn.Sequential(
nn.BatchNorm3d(
PARAM_KERNEL_SIZE, eps=0.001, momentum=0.1,
affine=True), # 0.1
nn.ReLU(inplace=True))
self.batch_norm3x3 = nn.Sequential(
nn.BatchNorm3d(
PARAM_KERNEL_SIZE, eps=0.001, momentum=0.1,
affine=True), # 0.1
nn.ReLU(inplace=True))
self.pool = nn.AdaptiveAvgPool3d(1)
self.conv_se = nn.Sequential(
nn.Conv3d(
PARAM_KERNEL_SIZE, band // reduction, 1, padding=0, bias=True),
nn.ReLU(inplace=True))
self.conv_ex = nn.Conv3d(
band // reduction, PARAM_KERNEL_SIZE, 1, padding=0, bias=True)
self.softmax = nn.Softmax(dim=1)
self.res_net1 = Residual(
PARAM_KERNEL_SIZE,
PARAM_KERNEL_SIZE, (1, 1, 7), (0, 0, 3),
start_block=True)
self.res_net2 = Residual(PARAM_KERNEL_SIZE, PARAM_KERNEL_SIZE,
(1, 1, 7), (0, 0, 3))
self.res_net3 = Residual(PARAM_KERNEL_SIZE, PARAM_KERNEL_SIZE,
(3, 3, 1), (1, 1, 0))
self.res_net4 = Residual(
PARAM_KERNEL_SIZE,
PARAM_KERNEL_SIZE, (3, 3, 1), (1, 1, 0),
end_block=True)
kernel_3d = math.ceil((band - 6) / 2)
# print(kernel_3d)
self.conv2 = nn.Conv3d(
in_channels=PARAM_KERNEL_SIZE,
out_channels=128,
padding=(0, 0, 0),
kernel_size=(1, 1, kernel_3d),
stride=(1, 1, 1))
self.batch_norm2 = nn.Sequential(
nn.BatchNorm3d(128, eps=0.001, momentum=0.1, affine=True), # 0.1
nn.ReLU(inplace=True))
self.conv3 = nn.Conv3d(
in_channels=1,
out_channels=PARAM_KERNEL_SIZE,
padding=(0, 0, 0),
kernel_size=(3, 3, 128),
stride=(1, 1, 1))
self.batch_norm3 = nn.Sequential(
nn.BatchNorm3d(
PARAM_KERNEL_SIZE, eps=0.001, momentum=0.1,
affine=True), # 0.1
nn.ReLU(inplace=True))
self.avg_pooling = nn.AvgPool3d(kernel_size=(5, 5, 1))
self.full_connection = nn.Sequential(
nn.Linear(PARAM_KERNEL_SIZE, classes)
# nn.Softmax()
)
def forward(self, X):
x_1x1 = self.conv1x1(X)
x_1x1 = self.batch_norm1x1(x_1x1).unsqueeze(dim=1)
x_3x3 = self.conv3x3(X)
x_3x3 = self.batch_norm3x3(x_3x3).unsqueeze(dim=1)
x1 = torch.cat([x_3x3, x_1x1], dim=1)
U = torch.sum(x1, dim=1)
S = self.pool(U)
Z = self.conv_se(S)
attention_vector = torch.cat(
[
self.conv_ex(Z).unsqueeze(dim=1),
self.conv_ex(Z).unsqueeze(dim=1)
],
dim=1)
attention_vector = self.softmax(attention_vector)
V = (x1 * attention_vector).sum(dim=1)
x2 = self.res_net1(V)
x2 = self.res_net2(x2)
x2 = self.batch_norm2(self.conv2(x2))
x2 = x2.permute(0, 4, 2, 3, 1)
x2 = self.batch_norm3(self.conv3(x2))
x3 = self.res_net3(x2)
x3 = self.res_net4(x3)
x4 = self.avg_pooling(x3)
x4 = x4.view(x4.size(0), -1)
return self.full_connection(x4)
model = S3KAIResNet(BAND, CLASSES_NUM, 2).cuda()
summary(model, input_data=(1, img_rows, img_cols, BAND), verbose=1)
def train(net,
train_iter,
valida_iter,
loss,
optimizer,
device,
epochs,
early_stopping=True,
early_num=20):
loss_list = [100]
early_epoch = 0
net = net.to(device)
print("training on ", device)
start = time.time()
train_loss_list = []
valida_loss_list = []
train_acc_list = []
valida_acc_list = []
for epoch in range(epochs):
train_acc_sum, n = 0.0, 0
time_epoch = time.time()
lr_adjust = torch.optim.lr_scheduler.CosineAnnealingLR(
optimizer, 15, eta_min=0.0, last_epoch=-1)
for X, y in train_iter:
batch_count, train_l_sum = 0, 0
X = X.to(device)
y = y.to(device)
y_hat = net(X)
l = loss(y_hat, y.long())
optimizer.zero_grad()
l.backward()
optimizer.step()
train_l_sum += l.cpu().item()
train_acc_sum += (y_hat.argmax(dim=1) == y).sum().cpu().item()
n += y.shape[0]
batch_count += 1
lr_adjust.step()
valida_acc, valida_loss = record.evaluate_accuracy(
valida_iter, net, loss, device)
loss_list.append(valida_loss)
train_loss_list.append(train_l_sum) # / batch_count)
train_acc_list.append(train_acc_sum / n)
valida_loss_list.append(valida_loss)
valida_acc_list.append(valida_acc)
print(
'epoch %d, train loss %.6f, train acc %.3f, valida loss %.6f, valida acc %.3f, time %.1f sec'
% (epoch + 1, train_l_sum / batch_count, train_acc_sum / n,
valida_loss, valida_acc, time.time() - time_epoch))
PATH = "./net_DBA.pt"
if early_stopping and loss_list[-2] < loss_list[-1]:
if early_epoch == 0:
torch.save(net.state_dict(), PATH)
early_epoch += 1
loss_list[-1] = loss_list[-2]
if early_epoch == early_num:
net.load_state_dict(torch.load(PATH))
break
else:
early_epoch = 0
print('epoch %d, loss %.4f, train acc %.3f, time %.1f sec'
% (epoch + 1, train_l_sum / batch_count, train_acc_sum / n,
time.time() - start))
def sampling(proportion, ground_truth):
train = {}
test = {}
labels_loc = {}
m = max(ground_truth)
for i in range(m):
indexes = [
j for j, x in enumerate(ground_truth.ravel().tolist())
if x == i + 1
]
np.random.shuffle(indexes)
labels_loc[i] = indexes
if proportion != 1:
nb_val = max(int((1 - proportion) * len(indexes)), 3)
else:
nb_val = 0
train[i] = indexes[:nb_val]
test[i] = indexes[nb_val:]
train_indexes = []
test_indexes = []
for i in range(m):
train_indexes += train[i]
test_indexes += test[i]
np.random.shuffle(train_indexes)
np.random.shuffle(test_indexes)
return train_indexes, test_indexes
def select(groundTruth): #divide dataset into train and test datasets
labels_loc = {}
train = {}
test = {}
m = max(groundTruth)
#amount = [3, 41, 29, 7, 14, 20, 2, 15, 3, 36, 64, 22, 4, 28, 10, 2]
#amount = [43, 1387, 801, 230, 469, 710, 26, 463, 17, 936, 2391, 571, 201, 1237, 376, 91]
if Dataset == 'IN':
amount = [
35, 1011, 581, 167, 344, 515, 19, 327, 12, 683, 1700, 418, 138,
876, 274, 69
] #IP 20%
#amount = [6, 144, 84, 24, 50, 75, 3, 49, 2, 97, 247, 62, 22, 130, 38, 10] #IP 20%
if Dataset == 'UP':
amount = [5297, 14974, 1648, 2424, 1076, 4026, 1046, 2950, 755] #UP
if Dataset == 'KSC':
amount = [
530, 165, 176, 170, 110, 161, 80, 299, 377, 283, 296, 341, 654
] #KSC
for i in range(m):
indices = [
j for j, x in enumerate(groundTruth.ravel().tolist()) if x == i + 1
]
np.random.shuffle(indices)
labels_loc[i] = indices
nb_val = int(amount[i])
train[i] = indices[:-nb_val]
test[i] = indices[-nb_val:]
# whole_indices = []
train_indices = []
test_indices = []
for i in range(m):
# whole_indices += labels_loc[i]
train_indices += train[i]
test_indices += test[i]
np.random.shuffle(train_indices)
np.random.shuffle(test_indices)
return train_indices, test_indices
# # Training
for index_iter in range(ITER):
print('iter:', index_iter)
#define the model
net = S3KAIResNet(BAND, CLASSES_NUM, 2)
if PARAM_OPTIM == 'diffgrad':
optimizer = optim2.DiffGrad(
net.parameters(),
lr=lr,
betas=(0.9, 0.999),
eps=1e-8,
weight_decay=0) # weight_decay=0.0001)
if PARAM_OPTIM == 'adam':
optimizer = optim.Adam(
net.parameters(),
lr=1e-3,
betas=(0.9, 0.999),
eps=1e-8,
weight_decay=0)
time_1 = int(time.time())
np.random.seed(seeds[index_iter])
# train_indices, test_indices = select(gt)
train_indices, test_indices = sampling(VALIDATION_SPLIT, gt)
_, total_indices = sampling(1, gt)
TRAIN_SIZE = len(train_indices)
print('Train size: ', TRAIN_SIZE)
TEST_SIZE = TOTAL_SIZE - TRAIN_SIZE
print('Test size: ', TEST_SIZE)
VAL_SIZE = int(TRAIN_SIZE)
print('Validation size: ', VAL_SIZE)
print('-----Selecting Small Pieces from the Original Cube Data-----')
train_iter, valida_iter, test_iter, all_iter = geniter.generate_iter(
TRAIN_SIZE, train_indices, TEST_SIZE, test_indices, TOTAL_SIZE,
total_indices, VAL_SIZE, whole_data, PATCH_LENGTH, padded_data,
INPUT_DIMENSION, 16, gt) #batchsize in 1
tic1 = time.time()
train(
net,
train_iter,
valida_iter,
loss,
optimizer,
device,
epochs=PARAM_EPOCH)
toc1 = time.time()
pred_test = []
tic2 = time.time()
with torch.no_grad():
for X, y in test_iter:
# print('Shape of X', X.shape, 'Shape of y', y.shape)
# X = X.permute(0, 3, 1, 2)
X = X.to(device)
net.eval()
y_hat = net(X)
pred_test.extend(np.array(net(X).cpu().argmax(axis=1)))
toc2 = time.time()
collections.Counter(pred_test)
gt_test = gt[test_indices] - 1
overall_acc = metrics.accuracy_score(pred_test, gt_test[:-VAL_SIZE])
confusion_matrix = metrics.confusion_matrix(pred_test, gt_test[:-VAL_SIZE])
each_acc, average_acc = record.aa_and_each_accuracy(confusion_matrix)
kappa = metrics.cohen_kappa_score(pred_test, gt_test[:-VAL_SIZE])
torch.save(
net.state_dict(), "./models/S3KAIResNetpatch_" + str(img_rows) + '_' +
Dataset + '_split_' + str(VALIDATION_SPLIT) + '_lr_' + str(lr) +
PARAM_OPTIM + '_kernel_' + str(PARAM_KERNEL_SIZE) + str(
round(overall_acc, 3)) + '.pt')
KAPPA.append(kappa)
OA.append(overall_acc)
AA.append(average_acc)
TRAINING_TIME.append(toc1 - tic1)
TESTING_TIME.append(toc2 - tic2)
ELEMENT_ACC[index_iter, :] = each_acc
# # Map, Records
print("--------" + " Training Finished-----------")
record.record_output(
OA, AA, KAPPA, ELEMENT_ACC, TRAINING_TIME, TESTING_TIME,
'./report/' + 'S3KAIResNetpatch:' + str(img_rows) + '_' + Dataset + 'split'
+ str(VALIDATION_SPLIT) + 'lr' + str(lr) + PARAM_OPTIM + '_kernel_' +
str(PARAM_KERNEL_SIZE) + '.txt')
Utils.generate_png(
all_iter, net, gt_hsi, Dataset, device, total_indices,
'./classification_maps/' + 'S3KAIResNetpatch:' + str(img_rows) + '_' +
Dataset + 'split' + str(VALIDATION_SPLIT) + 'lr' + str(lr) + PARAM_OPTIM +
'_kernel_' + str(PARAM_KERNEL_SIZE))