-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
240 lines (210 loc) · 9.66 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
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
import argparse
import provider
import numpy as np
import datetime
import os
import config
import tensorflow as tf
from model_Att import DGCNN, DGCNNLoss
# from pointnet_model import PointNet, PointNetLoss
# Parameter Set
parser = argparse.ArgumentParser()
parser.add_argument('--test_area', type=str, default=config.TEST_AREA, help='The ID of test area.')
parser.add_argument('--learning_rate', type=float, default=config.LEARNING_RATE,
help='Initial learning rate [default: 0.001]')
parser.add_argument('--decay_steps', type=int, default=config.DECAY_STEPS,
help='Learning rate decay when it going around steps')
parser.add_argument('--decay_rate', type=float, default=config.DECAY_RATE, help='Learning_rate *= decay_rate')
parser.add_argument('--log_dir', type=str, default=config.LOG, help='Log dir [default: log]')
parser.add_argument('--epoch', type=int, default=config.EPOCH, help='The counts of training epoches')
parser.add_argument('--batch_size', type=int, default=config.BATCH_SIZE,
help='Batch Size during training for each GPU [default: 24]')
parser.add_argument('--test_frequency', type=int, default=config.TEST_FREQUENCY, help='Test frequency')
parser.add_argument('--save_frequency', type=int, default=config.SAVE_FREQUENCY, help='Save frequency')
parser.add_argument('--num_classes', type=int, default=config.NUM_CLASSES, help='number of classes')
FLAGS = parser.parse_args()
TEST_AREA = FLAGS.test_area
LEARNING_RATE = FLAGS.learning_rate
LOG_DIR = FLAGS.log_dir
EPOCH = FLAGS.epoch
BATCH_SIZE = FLAGS.batch_size
DECAY_STEPS = FLAGS.decay_steps
DECAY_RATE = FLAGS.decay_rate
TEST_FREQUENCY = FLAGS.test_frequency
SAVE_FREQUENCY = FLAGS.save_frequency
# Load Data
DATA_PATH = './data'
data_batch_list = []
label_batch_list = []
ALL_FILES = provider.getDataFiles(DATA_PATH + '/all_files.txt')
ROOM_LIST = provider.getDataFiles(DATA_PATH + '/rooms_name.txt')
for _ in ALL_FILES:
data_batch, label_batch = provider.loadH5Files(DATA_PATH + '/' + _.split('/')[-1])
data_batch_list.append(data_batch)
label_batch_list.append(label_batch)
data_batches = np.concatenate(data_batch_list, 0)
label_batches = np.concatenate(label_batch_list, 0)
test_area = TEST_AREA
print('Test Area: ' + test_area)
train_idx = []
test_idx = []
for room_id, room_name in enumerate(ROOM_LIST):
if test_area in room_name:
test_idx.append(room_id)
else:
train_idx.append(room_id)
train_data = data_batches[train_idx, ...].astype(np.float32)
train_label = label_batches[train_idx].astype(np.int32)
test_data = data_batches[test_idx, ...].astype(np.float32)
test_label = label_batches[test_idx].astype(np.int32)
TRAIN_BATCHES = train_data.shape[0]
print('Train Batches: {} Test Batches:{}'.format(TRAIN_BATCHES, test_data.shape[0]))
print('Train Points : {} Test Points:{}'.format(TRAIN_BATCHES * train_data.shape[1],
test_data.shape[0] * test_data.shape[1]))
train_weights = np.zeros((config.NUM_CLASSES,),dtype=np.float32)
for l in train_label:
for i in range(config.NUM_CLASSES):
mask_array = np.full_like(l, i, dtype=np.int32)
train_weights[i] += np.sum(np.equal(mask_array,l))
all_points_count = np.sum(train_weights)
for _ in range(len(train_weights)):
train_weights[_] /= all_points_count
train_weights[_] = 1./np.log(1.2+train_weights[_])
# train_weights /= np.max(train_weights)
def rotate_point_cloud_z(x):
batch_data = x.numpy()
for k in range(batch_data.shape[0]):
rotation_angle = np.random.uniform() * 2 * np.pi
cosval = np.cos(rotation_angle)
sinval = np.sin(rotation_angle)
rotation_matrix = np.array([[cosval, sinval, 0.],
[-sinval, cosval, 0.],
[0., 0., 1.]], dtype=np.float32)
batch_data[k, :, :3] = np.matmul(batch_data[k, :, :3], rotation_matrix)
batch_data[k, :, 3:6] = np.matmul(batch_data[k, :, 3:6], rotation_matrix)
return batch_data
# Dataset
train_dataset = tf.data.Dataset.from_tensor_slices((train_data, train_label))
train_dataset = train_dataset.shuffle(1000)
train_dataset = train_dataset.batch(BATCH_SIZE)
train_dataset = train_dataset.prefetch(20 * BATCH_SIZE)
test_dataset = tf.data.Dataset.from_tensor_slices((test_data, test_label))
del train_data, train_label, test_data, test_label
# Model
dgcnn_net = DGCNN(k=20, num_classes=12)
# dgcnn_net = PointNet(num_point=8192,num_attribute=9,num_classes=12)
# Loss
dgcnn_loss = DGCNNLoss()
# dgcnn_loss = PointNetLoss()
# Optimizer
lr_schedule = tf.keras.optimizers.schedules.ExponentialDecay(
initial_learning_rate=LEARNING_RATE,
decay_steps=DECAY_STEPS,
decay_rate=DECAY_RATE,
staircase=True
)
optimizer = tf.optimizers.Adam(learning_rate=lr_schedule)
# @tf.function
def trainOneBatch(batch_data, batch_label, sample_weight=None):
batch_data = rotate_point_cloud_z(batch_data)
with tf.GradientTape() as tape:
outputs = dgcnn_net(batch_data, training=True)
loss_value = tf.reduce_sum(dgcnn_loss(y_true=batch_label, y_pred=outputs,sample_weight=sample_weight))
gradients = tape.gradient(loss_value, dgcnn_net.trainable_variables)
optimizer.apply_gradients(grads_and_vars=zip(gradients, dgcnn_net.trainable_variables))
correct = tf.reduce_sum(
tf.cast(tf.equal(tf.argmax(outputs, 2, output_type=tf.int32), batch_label), tf.float32)
)
return loss_value, correct
# @tf.function
def testOneBatch(batch_data, batch_label):
if len(batch_data.shape) < 3:
batch_data = tf.expand_dims(batch_data, 0)
if len(batch_label.shape) < 2:
batch_label = tf.expand_dims(batch_label, 0)
outputs = dgcnn_net(batch_data, training=False)
loss_value = tf.reduce_sum(dgcnn_loss(y_true=batch_label, y_pred=outputs))
correct = tf.reduce_sum(
tf.cast(tf.equal(tf.argmax(outputs, 2, output_type=tf.int32), batch_label), tf.float32)
)
return loss_value, correct
# LOGS Tracing
current_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
summary_writer = tf.summary.create_file_writer(os.path.join(LOG_DIR, current_time))
# Training
GLOBAL_STEP = 1
BEST_TEST_ACC_LIST = [-1 * np.inf]
for epoch in range(1, EPOCH + 1):
train_epoch_loss = 0.
train_correct = 0
train_count = 0
step = 1
for data, label in train_dataset:
sw = train_weights[label.numpy()]
train_batch_loss, train_batch_correct = trainOneBatch(data, label, sw)
train_batch_count = tf.cast(label.shape[0] * label.shape[1], tf.float32)
if train_batch_loss < 0:
print("Skip this step due to negative loss.")
continue
train_epoch_loss += train_batch_loss
train_correct += train_batch_correct
train_count += train_batch_count
with summary_writer.as_default():
tf.summary.scalar('train_batch_loss', train_batch_loss, step=GLOBAL_STEP)
print(
"Epoch: {}/{}, Step: {}/{},"
"batch_loss: {:.5f}, global_steps:{}, lr:{:.9f}.".format(
epoch,
EPOCH,
step,
TRAIN_BATCHES // BATCH_SIZE + 1,
train_batch_loss,
GLOBAL_STEP,
lr_schedule.__call__(GLOBAL_STEP)
))
GLOBAL_STEP += 1
step += 1
train_epoch_loss /= step
train_epoch_acc = train_correct / train_count
with summary_writer.as_default():
tf.summary.scalar('train_epoch_loss', train_epoch_loss, step=epoch)
tf.summary.scalar('train_epoch_accuracy', train_epoch_acc, step=epoch)
print("Epoch: {}/{},Train Average Loss:{:.5f} Accuracy:{:.5f}.".format(
epoch, EPOCH, train_epoch_loss, train_epoch_acc)
)
# Testing
if epoch % TEST_FREQUENCY == 0 and test_area != 'None':
test_loss = 0.
test_correct = 0
test_count = 0
step = 1
for data, label in test_dataset:
test_loss_value, test_batch_correct = testOneBatch(data, label)
test_batch_count = tf.cast(label.shape[-1], tf.float32)
if test_loss_value < 0:
print("Skip this step due to negative loss.")
continue
test_loss += test_loss_value
test_correct += test_batch_correct
test_count += test_batch_count
step += 1
test_loss /= step
test_acc = test_correct / test_count
with summary_writer.as_default():
tf.summary.scalar('test_loss', test_loss, step=epoch)
tf.summary.scalar('test_accuracy', test_acc, step=epoch)
print("Test_loss: {:.5f} Accuracy: {:.5f}".format(test_loss, test_acc))
if test_acc > BEST_TEST_ACC_LIST[-1]:
dgcnn_net.save_weights(os.path.join(LOG_DIR, 'dgcnn'), save_format='tf')
print(
"The best accuracy on test dataset has declined from {} to {}, and saving model weight to {}."
.format(BEST_TEST_ACC_LIST[-1], test_acc, os.path.join(LOG_DIR, 'dgcnn_arg'))
)
BEST_TEST_ACC_LIST.append(test_acc)
if epoch % SAVE_FREQUENCY == 0:
dgcnn_net.save_weights(filepath=os.path.join(LOG_DIR, 'dgcnn' + "epoch-{}".format(epoch)), save_format='tf')
print("Save Model in {}.".format(os.path.join(LOG_DIR, 'dgcnn' + "epoch-{}".format(epoch))) + "epoch-{}".format(
epoch))
with open('record.txt', 'a') as fo:
fo.write('\nTest Area: ' + test_area + '\n')
fo.write('Overall Accuracy: ' + str(max(BEST_TEST_ACC_LIST)))