-
Notifications
You must be signed in to change notification settings - Fork 0
/
trainInconSeg_with_RGB_Disparity.py
325 lines (278 loc) · 21.6 KB
/
trainInconSeg_with_RGB_Disparity.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
# By Yuxiang Sun, Dec. 4, 2019
# Email: [email protected]
import os, argparse, time, datetime, stat, shutil
import numpy as np
import torch
import torch.nn.functional as F
from torch.autograd import Variable
from torch.utils.data import DataLoader
import torchvision.utils as vutils
from util.RGB_Disparity_dataset import RGB_Disparity_dataset
from util.augmentation import RandomFlip, RandomCrop, RandomCropOut, RandomBrightness, RandomNoise
from util.util import compute_results
from sklearn.metrics import confusion_matrix
from torch.utils.tensorboard import SummaryWriter
from model import InconSeg
#############################################################################################
parser = argparse.ArgumentParser(description='Train with pytorch')
#############################################################################################
parser.add_argument('--model_name', '-m', type=str, default='InconSeg')
parser.add_argument('--batch_size', '-b', type=int, default=2)
parser.add_argument('--gpu', '-g', type=int, default=1)
parser.add_argument('--need_m', '-need_m', type=int, default=7)
parser.add_argument('--sleep', '-sleep', type=int, default=1)
parser.add_argument('--lr_start', '-ls', type=float, default=0.01)
#############################################################################################
parser.add_argument('--lr_decay', '-ld', type=float, default=0.95)
parser.add_argument('--epoch_max', '-em', type=int, default=200) # please stop training mannully
parser.add_argument('--epoch_from', '-ef', type=int, default=0)
parser.add_argument('--num_workers', '-j', type=int, default=8)
parser.add_argument('--n_class', '-nc', type=int, default=3)
parser.add_argument('--data_dir', '-dr', type=str, default='./dataset/')
args = parser.parse_args()
#############################################################################################
augmentation_methods = [
RandomFlip(prob=0.5),
RandomCrop(crop_rate=0.1, prob=1.0),
# RandomCropOut(crop_rate=0.2, prob=1.0),
# RandomBrightness(bright_range=0.15, prob=0.9),
# RandomNoise(noise_range=5, prob=0.9),
]
def fusion_loss(rgb_seg_1,depth_add,label):
feature_map_B, feature_map_C, feature_map_W, feature_map_H = rgb_seg_1.size()
label_B, label_W, label_H = label.size()
if feature_map_W != label_W:
label = torch.cuda.FloatTensor(label.unsqueeze(1).cpu().numpy())
label = F.interpolate(label,[feature_map_W,feature_map_H],mode="nearest")
label = torch.cuda.LongTensor(label.squeeze(1).cpu().numpy())
loss_pr_seg = F.cross_entropy(rgb_seg_1, label)
rgb_seg_1 = rgb_seg_1.detach()
rgb_seg_1=rgb_seg_1.argmax(1)
rgb_seg_1.eq_(label)
rgb_seg_1=rgb_seg_1.clone().detach_().requires_grad_(False)
add_map = (1-rgb_seg_1)*label
add_map=add_map.clone().detach_().requires_grad_(False)
loss_add = F.cross_entropy(depth_add,add_map)
return loss_pr_seg,loss_add,add_map
def train(epo, model, train_loader, optimizer):
model.train()
for it, (images, labels, names) in enumerate(train_loader):
images = Variable(images).cuda(args.gpu)
labels = Variable(labels).cuda(args.gpu)
start_t = time.time() # time.time() returns the current time
optimizer.zero_grad()
depth_result, rgb_result,rgb_seg_f1,depth_add_f1,rgb_seg_f2,depth_add_f2,rgb_seg_f3,depth_add_f3 = model(images) # depth image result, rgb image result, rgb image pre-segmentaion, complement feature
loss_d = F.cross_entropy(depth_result, labels) # Note that the cross_entropy function has already include the softmax function
loss_r = F.cross_entropy(rgb_result, labels) # Note that the cross_entropy function has already include the softmax function
loss_pr_seg_f1,loss_add_f1,add_map_f1 = fusion_loss(rgb_seg_f1,depth_add_f1,label=labels)
loss_pr_seg_f2,loss_add_f2,add_map_f2 = fusion_loss(rgb_seg_f2,depth_add_f2,label=labels)
loss_pr_seg_f3,loss_add_f3,add_map_f3 = fusion_loss(rgb_seg_f3,depth_add_f3,label=labels)
#print(add_map.size())
loss = loss_d+loss_r+loss_pr_seg_f1+loss_add_f1+loss_pr_seg_f2+loss_add_f2+loss_pr_seg_f3+loss_add_f3
loss.backward()
optimizer.step()
lr_this_epo=0
for param_group in optimizer.param_groups:
lr_this_epo = param_group['lr']
print('Train: %s, epo %s/%s, iter %s/%s, lr %.8f, %.2f img/sec, loss %.4f, loss_d %.4f, loss_r %.4f, loss_pr_seg_f1 %.4f, loss_add_f1 %.4f, loss_pr_seg_f2 %.4f, loss_add_f2 %.4f,loss_pr_seg_f3 %.4f, loss_add_f3 %.4f, time %s' \
% (args.model_name, epo, args.epoch_max, it+1, len(train_loader), lr_this_epo, len(names)/(time.time()-start_t), float(loss), float(loss_d), float(loss_r), float(loss_pr_seg_f1), float(loss_add_f1), float(loss_pr_seg_f2), float(loss_add_f2), float(loss_pr_seg_f3), float(loss_add_f3),
datetime.datetime.now().replace(microsecond=0)-start_datetime))
if accIter['train'] % 1 == 0:
writer.add_scalar('Train/loss', loss, accIter['train'])
writer.add_scalar('Train/loss_d', loss_d, accIter['train'])
writer.add_scalar('Train/loss_r', loss_r, accIter['train'])
writer.add_scalar('Train/loss_seg_f1', loss_pr_seg_f1, accIter['train'])
writer.add_scalar('Train/loss_mse_f1', loss_add_f1, accIter['train'])
writer.add_scalar('Train/loss_seg_f2', loss_pr_seg_f2, accIter['train'])
writer.add_scalar('Train/loss_mse_f2', loss_add_f2, accIter['train'])
writer.add_scalar('Train/loss_seg_f3', loss_pr_seg_f3, accIter['train'])
writer.add_scalar('Train/loss_mse_f3', loss_add_f3, accIter['train'])
view_figure = True # note that I have not colorized the GT and predictions here
if accIter['train'] % 10 == 0:
if view_figure:
input_rgb_images = vutils.make_grid(images[:,:3], nrow=8, padding=10) # can only display 3-channel images, so images[:,:3]
writer.add_image('Train/input_rgb_images', input_rgb_images, accIter['train'])
scale = max(1, 255//args.n_class) # label (0,1,2..) is invisable, multiply a constant for visualization
groundtruth_tensor = labels.unsqueeze(1) * scale # mini_batch*480*640 -> mini_batch*1*480*640
groundtruth_tensor = torch.cat((groundtruth_tensor, groundtruth_tensor, groundtruth_tensor), 1) # change to 3-channel for visualization
groudtruth_images = vutils.make_grid(groundtruth_tensor, nrow=8, padding=10)
writer.add_image('Train/groudtruth_images', groudtruth_images, accIter['train'])
scale = max(1, 255//args.n_class) # label (0,1,2..) is invisable, multiply a constant for visualization
add_map_tensor_f1 = add_map_f1.unsqueeze(1) * scale # mini_batch*480*640 -> mini_batch*1*480*640
add_map_tensor_f1 = torch.cat((add_map_tensor_f1, add_map_tensor_f1, add_map_tensor_f1), 1) # change to 3-channel for visualization
add_map_images_f1 = vutils.make_grid(add_map_tensor_f1, nrow=8, padding=10)
writer.add_image('Train/add_map_images_f1', add_map_images_f1, accIter['train'])
scale = max(1, 255//args.n_class) # label (0,1,2..) is invisable, multiply a constant for visualization
add_map_tensor_f2 = add_map_f2.unsqueeze(1) * scale # mini_batch*480*640 -> mini_batch*1*480*640
add_map_tensor_f2 = torch.cat((add_map_tensor_f2, add_map_tensor_f2, add_map_tensor_f2), 1) # change to 3-channel for visualization
add_map_images_f2 = vutils.make_grid(add_map_tensor_f2, nrow=8, padding=10)
writer.add_image('Train/add_map_images_f2', add_map_images_f2, accIter['train'])
scale = max(1, 255//args.n_class) # label (0,1,2..) is invisable, multiply a constant for visualization
add_map_tensor_f3 = add_map_f3.unsqueeze(1) * scale # mini_batch*480*640 -> mini_batch*1*480*640
add_map_tensor_f3 = torch.cat((add_map_tensor_f3, add_map_tensor_f3, add_map_tensor_f3), 1) # change to 3-channel for visualization
add_map_images_f3 = vutils.make_grid(add_map_tensor_f3, nrow=8, padding=10)
writer.add_image('Train/add_map_images_f3', add_map_images_f3, accIter['train'])
predicted_depth = depth_result.argmax(1).unsqueeze(1) * scale # mini_batch*args.n_class*480*640 -> mini_batch*480*640 -> mini_batch*1*480*640
predicted_depth = torch.cat((predicted_depth, predicted_depth, predicted_depth),1) # change to 3-channel for visualization, mini_batch*1*480*640
predicted_depth_images = vutils.make_grid(predicted_depth, nrow=8, padding=10)
writer.add_image('Train/predicted_depth_images', predicted_depth_images, accIter['train'])
predicted_rgb = rgb_result.argmax(1).unsqueeze(1) * scale # mini_batch*args.n_class*480*640 -> mini_batch*480*640 -> mini_batch*1*480*640
predicted_rgb = torch.cat((predicted_rgb, predicted_rgb, predicted_rgb),1) # change to 3-channel for visualization, mini_batch*1*480*640
predicted_rgb_images = vutils.make_grid(predicted_rgb, nrow=8, padding=10)
writer.add_image('Train/predicted_rgb_images', predicted_rgb_images, accIter['train'])
predicted_rgb_seg_f1= rgb_seg_f1.argmax(1).unsqueeze(1) * scale # mini_batch*args.n_class*480*640 -> mini_batch*480*640 -> mini_batch*1*480*640
predicted_rgb_seg_f1 = torch.cat((predicted_rgb_seg_f1, predicted_rgb_seg_f1, predicted_rgb_seg_f1),1) # change to 3-channel for visualization, mini_batch*1*480*640
predicted_rgb_seg_images_f1 = vutils.make_grid(predicted_rgb_seg_f1, nrow=8, padding=10)
writer.add_image('Train/predicted_rgb_seg_images_f1', predicted_rgb_seg_images_f1, accIter['train'])
predicted_depth_add_f1= depth_add_f1.argmax(1).unsqueeze(1) * scale # mini_batch*args.n_class*480*640 -> mini_batch*480*640 -> mini_batch*1*480*640
predicted_depth_add_f1 = torch.cat((predicted_depth_add_f1, predicted_depth_add_f1, predicted_depth_add_f1),1) # change to 3-channel for visualization, mini_batch*1*480*640
predicted_depth_add_images_f1 = vutils.make_grid(predicted_depth_add_f1, nrow=8, padding=10)
writer.add_image('Train/predicted_depth_add_images_f1', predicted_depth_add_images_f1, accIter['train'])
predicted_rgb_seg_f2= rgb_seg_f2.argmax(1).unsqueeze(1) * scale # mini_batch*args.n_class*480*640 -> mini_batch*480*640 -> mini_batch*1*480*640
predicted_rgb_seg_f2 = torch.cat((predicted_rgb_seg_f2, predicted_rgb_seg_f2, predicted_rgb_seg_f2),1) # change to 3-channel for visualization, mini_batch*1*480*640
predicted_rgb_seg_images_f2 = vutils.make_grid(predicted_rgb_seg_f2, nrow=8, padding=10)
writer.add_image('Train/predicted_rgb_seg_images_f2', predicted_rgb_seg_images_f2, accIter['train'])
predicted_depth_add_f2= depth_add_f2.argmax(1).unsqueeze(1) * scale # mini_batch*args.n_class*480*640 -> mini_batch*480*640 -> mini_batch*1*480*640
predicted_depth_add_f2 = torch.cat((predicted_depth_add_f2, predicted_depth_add_f2, predicted_depth_add_f2),1) # change to 3-channel for visualization, mini_batch*1*480*640
predicted_depth_add_images_f2 = vutils.make_grid(predicted_depth_add_f2, nrow=8, padding=10)
writer.add_image('Train/predicted_depth_add_images_f2', predicted_depth_add_images_f2, accIter['train'])
predicted_rgb_seg_f3= rgb_seg_f3.argmax(1).unsqueeze(1) * scale # mini_batch*args.n_class*480*640 -> mini_batch*480*640 -> mini_batch*1*480*640
predicted_rgb_seg_f3 = torch.cat((predicted_rgb_seg_f3, predicted_rgb_seg_f3, predicted_rgb_seg_f3),1) # change to 3-channel for visualization, mini_batch*1*480*640
predicted_rgb_seg_images_f3 = vutils.make_grid(predicted_rgb_seg_f3, nrow=8, padding=10)
writer.add_image('Train/predicted_rgb_seg_images_f3', predicted_rgb_seg_images_f3, accIter['train'])
predicted_depth_add_f3= depth_add_f3.argmax(1).unsqueeze(1) * scale # mini_batch*args.n_class*480*640 -> mini_batch*480*640 -> mini_batch*1*480*640
predicted_depth_add_f3 = torch.cat((predicted_depth_add_f3, predicted_depth_add_f3, predicted_depth_add_f3),1) # change to 3-channel for visualization, mini_batch*1*480*640
predicted_depth_add_images_f3 = vutils.make_grid(predicted_depth_add_f3, nrow=8, padding=10)
writer.add_image('Train/predicted_depth_add_images_f3', predicted_depth_add_images_f3, accIter['train'])
depth_tensor = images[:,3:4] # mini_batch*args.n_class*480*640 -> mini_batch*480*640 -> mini_batch*1*480*640
depth_tensor = torch.cat((depth_tensor, depth_tensor, depth_tensor),1) # change to 3-channel for visualization, mini_batch*1*480*640
depth_tensor = vutils.make_grid(depth_tensor, nrow=8, padding=10)
writer.add_image('Train/depth_tensor', depth_tensor, accIter['train'])
accIter['train'] = accIter['train'] + 1
def validation(epo, model, val_loader):
model.eval()
with torch.no_grad():
for it, (images, labels, names) in enumerate(val_loader):
images = Variable(images).cuda(args.gpu)
labels = Variable(labels).cuda(args.gpu)
start_t = time.time() # time.time() returns the current time
depth_result, rgb_result,rgb_seg_f1,depth_add_f1,rgb_seg_f2,depth_add_f2,rgb_seg_f3,depth_add_f3 = model(images) # depth image result, rgb image result, rgb image pre-segmentaion, complement feature
loss_d = F.cross_entropy(depth_result, labels) # Note that the cross_entropy function has already include the softmax function
loss_r = F.cross_entropy(rgb_result, labels) # Note that the cross_entropy function has already include the softmax function
loss_pr_seg_f1,loss_add_f1,add_map_f1 = fusion_loss(rgb_seg_f1,depth_add_f1,label=labels)
loss_pr_seg_f2,loss_add_f2,add_map_f2 = fusion_loss(rgb_seg_f2,depth_add_f2,label=labels)
loss_pr_seg_f3,loss_add_f3,add_map_f3 = fusion_loss(rgb_seg_f3,depth_add_f3,label=labels)
#print(add_map.size())
loss = loss_d+loss_r+loss_pr_seg_f1+loss_add_f1+loss_pr_seg_f2+loss_add_f2+loss_pr_seg_f3+loss_add_f3
print('Val: %s, epo %s/%s, iter %s/%s, %.2f img/sec, loss %.4f, loss_d %.4f, loss_r %.4f, loss_pr_seg_f1 %.4f, loss_add_f1 %.4f, loss_pr_seg_f2 %.4f, loss_add_f2 %.4f, loss_pr_seg_f3 %.4f, loss_add_f3 %.4f, time %s' \
% (args.model_name, epo, args.epoch_max, it + 1, len(val_loader), len(names)/(time.time()-start_t), float(loss), float(loss_d), float(loss_r), float(loss_pr_seg_f1), float(loss_add_f1), float(loss_pr_seg_f2), float(loss_add_f2), float(loss_pr_seg_f3), float(loss_add_f3),
datetime.datetime.now().replace(microsecond=0)-start_datetime))
if accIter['val'] % 1 == 0:
writer.add_scalar('Validation/loss', loss, accIter['val'])
writer.add_scalar('Validation/loss_d', loss_d, accIter['val'])
writer.add_scalar('Validation/loss_r', loss_r, accIter['val'])
writer.add_scalar('Validation/loss_seg_f1', loss_pr_seg_f1, accIter['train'])
writer.add_scalar('Validation/loss_mse_f1', loss_add_f1, accIter['train'])
writer.add_scalar('Validation/loss_seg_f2', loss_pr_seg_f2, accIter['train'])
writer.add_scalar('Validation/loss_mse_f2', loss_add_f2, accIter['train'])
writer.add_scalar('Validation/loss_seg_f3', loss_pr_seg_f3, accIter['train'])
writer.add_scalar('Validation/loss_mse_f3', loss_add_f3, accIter['train'])
def testing(epo, model, test_loader):
model.eval()
conf_total = np.zeros((args.n_class, args.n_class))
label_list = ["unlabeled", "pothole", "car"]
testing_results_file = os.path.join(weight_dir, 'testing_results_file.txt')
with torch.no_grad():
for it, (images, labels, names) in enumerate(test_loader):
images = Variable(images).cuda(args.gpu)
labels = Variable(labels).cuda(args.gpu)
depth_result, rgb_result,rgb_seg_f1,depth_add_f1,rgb_seg_f2,depth_add_f2,rgb_seg_f3,depth_add_f3 = model(images) # depth image result, rgb image result, rgb image pre-segmentaion, complement feature
logits = rgb_result
label = labels.cpu().numpy().squeeze().flatten()
prediction = logits.argmax(1).cpu().numpy().squeeze().flatten() # prediction and label are both 1-d array, size: minibatch*640*480
conf = confusion_matrix(y_true=label, y_pred=prediction, labels=[0,1,2]) # conf is args.n_class*args.n_class matrix, vertical axis: groundtruth, horizontal axis: prediction
conf_total += conf
print('Test: %s, epo %s/%s, iter %s/%s, time %s' % (args.model_name, epo, args.epoch_max, it+1, len(test_loader),
datetime.datetime.now().replace(microsecond=0)-start_datetime))
precision, recall, IoU,F1 = compute_results(conf_total)
writer.add_scalar('Test/average_precision',precision.mean(), epo)
writer.add_scalar('Test/average_recall', recall.mean(), epo)
writer.add_scalar('Test/average_IoU', IoU.mean(), epo)
writer.add_scalar('Test/average_F1', F1.mean(), epo)
for i in range(len(precision)):
writer.add_scalar("Test(class)/precision_class_%s" % label_list[i], precision[i], epo)
writer.add_scalar("Test(class)/recall_class_%s"% label_list[i], recall[i],epo)
writer.add_scalar('Test(class)/Iou_%s'% label_list[i], IoU[i], epo)
writer.add_scalar('Test(class)/F1_%s'% label_list[i], F1[i], epo)
if epo==0:
with open(testing_results_file, 'w') as f:
f.write("# %s, initial lr: %s, batch size: %s, date: %s \n" %(args.model_name, args.lr_start, args.batch_size, datetime.date.today()))
f.write("# epoch: unlabeled, pothole, car, average(nan_to_num). (Acc %, IoU %)\n")
with open(testing_results_file, 'a') as f:
f.write(str(epo)+': ')
for i in range(len(precision)):
f.write('%0.4f,%0.4f,%0.4f,%0.4f, ' % (100*precision[i], 100*recall[i], 100*IoU[i], 100*F1[i]))
f.write('%0.4f,%0.4f,%0.4f,%0.4f\n' % (100*np.mean(np.nan_to_num(precision)), 100*np.mean(np.nan_to_num(recall)), 100*np.mean(np.nan_to_num(IoU)),100*np.mean(np.nan_to_num(F1))))
print('saving testing results.')
with open(testing_results_file, "r") as file:
writer.add_text('testing_results', file.read().replace('\n', ' \n'), epo)
if __name__ == '__main__':
torch.cuda.set_device(args.gpu)
print("\nthe pytorch version:", torch.__version__)
print("the gpu count:", torch.cuda.device_count())
print("the current used gpu:", torch.cuda.current_device(), '\n')
model = eval(args.model_name)(n_class=args.n_class)
if args.gpu >= 0: model.cuda(args.gpu)
optimizer = torch.optim.SGD(model.parameters(), lr=args.lr_start, momentum=0.9, weight_decay=0.0005)
scheduler = torch.optim.lr_scheduler.ExponentialLR(optimizer, gamma=args.lr_decay, last_epoch=-1)
# preparing folders
if os.path.exists("./runs"):
shutil.rmtree("./runs")
weight_dir = os.path.join("./runs", args.model_name)
os.makedirs(weight_dir)
os.chmod(weight_dir, stat.S_IRWXO) # allow the folder created by docker read, written, and execuated by local machine
writer = SummaryWriter("./runs/tensorboard_log")
os.chmod("./runs/tensorboard_log", stat.S_IRWXO) # allow the folder created by docker read, written, and execuated by local machine
os.chmod("./runs", stat.S_IRWXO)
print('training %s on GPU #%d with pytorch' % (args.model_name, args.gpu))
print('from epoch %d / %s' % (args.epoch_from, args.epoch_max))
print('weight will be saved in: %s' % weight_dir)
train_dataset = RGB_Disparity_dataset(data_dir=args.data_dir, split='train', transform=augmentation_methods)
val_dataset = RGB_Disparity_dataset(data_dir=args.data_dir, split='validation')
test_dataset = RGB_Disparity_dataset(data_dir=args.data_dir, split='test')
train_loader = DataLoader(
dataset = train_dataset,
batch_size = args.batch_size,
shuffle = True,
num_workers = args.num_workers,
pin_memory = True,
drop_last = True
)
val_loader = DataLoader(
dataset = val_dataset,
batch_size = args.batch_size,
shuffle = False,
num_workers = args.num_workers,
pin_memory = True,
drop_last = True
)
test_loader = DataLoader(
dataset = test_dataset,
batch_size = args.batch_size,
shuffle = False,
num_workers = args.num_workers,
pin_memory = True,
drop_last = True
)
start_datetime = datetime.datetime.now().replace(microsecond=0)
accIter = {'train': 0, 'val': 0}
for epo in range(args.epoch_from, args.epoch_max):
print('\ntrain %s, epo #%s begin...' % (args.model_name, epo))
#scheduler.step() # if using pytorch 0.4.1, please put this statement here
train(epo, model, train_loader, optimizer)
validation(epo, model, val_loader)
checkpoint_model_file = os.path.join(weight_dir, str(epo) + '.pth')
print('saving check point %s: ' % checkpoint_model_file)
torch.save(model.state_dict(), checkpoint_model_file)
testing(epo, model, test_loader) # testing is just for your reference, you can comment this line during training
scheduler.step() # if using pytorch 1.1 or above, please put this statement here