-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
271 lines (212 loc) · 9.55 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
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
#%%
import numpy as np
import argparse
import yaml
from pathlib import Path
import os
from tqdm import tqdm
import tensorflow as tf
from tensorflow import keras as tk
import tensorflow.keras.backend as K
# tf.compat.v1.disable_eager_execution()
from models.hrnet import HRNet
from models.vggunet import Vggunet
from models.subject4 import Subject4
from models.bisenet import Bisenet
from models.callback import Custom_Callback
from dataparser.inria import Inria, Inria_v
from dataparser.ade20k import Ade20k, Ade20k_v
from dataparser.cityscape import Cityscape, Cityscape_v
#%%
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--config", type=str, required=True)
args = parser.parse_args()
config = yaml.load("".join(Path(args.config).open("r").readlines()), Loader=yaml.FullLoader)
# config = yaml.load("".join(Path("configs/ade20k_hrnet.yaml").open("r").readlines()), Loader=yaml.FullLoader)
print("=====================config=====================")
for v in config.keys() :
print("%s : %s" %(v, config[v]))
print("================================================")
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = ",".join([str(i) for i in config["gpu_indices"]])
if not (config["mode"] == 0 or config["mode"] == 1) :
print("Config mode is not for training!")
quit()
# for multi gpu
# tf.compat.v1.disable_eager_execution()
gpus = tf.config.experimental.list_physical_devices('GPU')
if gpus:
try:
for i in range(len(config["gpu_indices"])) :
tf.config.experimental.set_memory_growth(gpus[i], True)
except RuntimeError as e:
print(e)
if config["dataset_name"] == "inria" :
data_parser = Inria(config)
data_parserv = Inria_v(config)
if config["dataset_name"] == "ade20k" :
data_parser = Ade20k(config)
data_parserv = Ade20k_v(config)
if config["dataset_name"] == "cityscape" :
data_parser = Cityscape(config)
data_parserv = Cityscape_v(config)
mirrored_strategy = tf.distribute.MirroredStrategy()
with mirrored_strategy.scope():
if config["model_name"] == "hrnet" :
model = HRNet(configs=config)
elif config["model_name"] == "vggunet" :
model = Vggunet(configs=config)
elif config["model_name"] == "subject4" :
model = Subject4(configs=config)
elif config["model_name"] == "bisenet" :
model = Bisenet(configs=config)
with mirrored_strategy.scope():
dataset = tf.data.Dataset.from_generator(
data_parser.generator,
(tf.float32, tf.float32),
# (tf.TensorShape([config["image_size"][0], config["image_size"][1], 3]), tf.TensorShape([config["image_size"][0], config["image_size"][1], 3]))
(tf.TensorShape([None, None, 3]), tf.TensorShape([None, None]))
).batch(config["batch_size"], drop_remainder=True)
datasetv = tf.data.Dataset.from_generator(
data_parserv.generator,
(tf.float32, tf.float32),
# (tf.TensorShape([config["image_size"][0], config["image_size"][1], 3]), tf.TensorShape([config["image_size"][0], config["image_size"][1], 3]))
(tf.TensorShape([None, None, 3]), tf.TensorShape([None, None]))
).batch(config["batch_size"], drop_remainder=True)
dist_dataset = mirrored_strategy.experimental_distribute_dataset(dataset)
dist_datasetv = mirrored_strategy.experimental_distribute_dataset(datasetv)
# %%
@tf.function
def train_step(dist_inputs) :
def train_fn(inputs) :
x, y = inputs
with tf.GradientTape() as tape :
output = model.model(x, training=True)
loss = model.sce_loss(y, output)
grads = tape.gradient(loss, model.model.trainable_variables)
model.optim.apply_gradients(list(zip(grads, model.model.trainable_variables)))
return loss
per_example_losses = mirrored_strategy.experimental_run_v2(train_fn, args=(dist_inputs,))
mean_loss = mirrored_strategy.reduce(tf.distribute.ReduceOp.MEAN, per_example_losses, axis=None)
return mean_loss
@tf.function
def test_step(dist_inputs) :
def test_fn(inputs) :
x, y = inputs
output = model.model(x, training=False)
accu = model.pixel_accuracy(y, output)
miou = model.miou(y, output)
return accu, miou
pe_accu, pe_miou = mirrored_strategy.experimental_run_v2(test_fn, args=(dist_inputs,))
mean_accu = mirrored_strategy.reduce(tf.distribute.ReduceOp.MEAN, pe_accu, axis=None)
mean_miou = mirrored_strategy.reduce(tf.distribute.ReduceOp.MEAN, pe_miou, axis=None)
return mean_accu, mean_miou
def set_lr (e) :
updated_lr = config["lr"] * ((1/2)**(e // 15))
# if e > 20 :
# updated_lr = config["lr"]/2
# elif e > 40 :
# updated_lr = config["lr"]/4
# elif e > 60 :
# updated_lr = config["lr"]/8
# elif e > 80 :
# updated_lr = config["lr"]/16
# elif e > 100 :
# updated_lr = config["lr"]/32
# elif e > 120 :
# updated_lr = config["lr"]/64
# else :
# updated_lr = config["lr"]
return updated_lr
def train () :
logfile = Path(config["logger_file"])
if not logfile.exists() :
tmpf = logfile.open("w+")
top_text = "epoch,loss,pixel_acc,miou\n"
tmpf.write(top_text)
tmpf.close()
# model.model.summary()
for e in range(config["present_epoch"], config["epoch"]) :
# K.set_value(model.optim.learning_rate, set_lr(e))
losses = []
# tqdm_s = tqdm(range(data_parser.steps))
tqdm_s = tqdm(dist_dataset)
for s in tqdm_s :
# mean_loss = train_step(data_parser.get_batch())
mean_loss = train_step(s)
losses.append(mean_loss.numpy())
tqdm_s.set_description_str(f"Loss : {str(np.mean(losses))}")
print(f"Epoch : {str(e)}, Loss : {str(np.mean(losses))}")
accus, mious = [], []
# tqdm_sv = tqdm(range(data_parserv.steps))
tqdm_sv = tqdm(dist_datasetv)
for s in tqdm_sv :
# accu, miou = test_step(data_parserv.get_batch())
accu, miou = test_step(s)
accus.append(accu), mious.append(miou)
tqdm_sv.set_description_str(f"accu : {str(np.mean(accus))}, miou : {str(np.mean(mious))}")
print(f"Epoch : {str(e)}, Accu : {str(np.mean(accus))}, miou : {str(np.mean(mious))}")
model.miou_op.reset_states()
tmpf = logfile.open("a+")
tmpf.write(",".join([str(e), str(np.mean(losses)), str(np.mean(accus)), str(np.mean(mious))]) + "\n")
tmpf.close()
data_parser.on_epoch_end()
data_parserv.on_epoch_end()
if not Path(config['save_path']).exists() :
Path(config['save_path']).mkdir(parents=True)
# break
if e % config["saving_interval"] == config["saving_interval"]-1 :
model.model.save(f"{config['save_path']}/model_{str(e+1)}.h5")
with mirrored_strategy.scope():
train()
# repeat = config["epoch"]*data_parser.steps
# repeatv = config["epoch"]*data_parserv.steps
# with mirrored_strategy.scope():
# dataset = tf.data.Dataset.from_generator(
# data_parser.generator,
# (tf.float32, tf.float32),
# (tf.TensorShape([config["image_size"][0], config["image_size"][1], 3]), tf.TensorShape([config["image_size"][0], config["image_size"][1], 3]))
# ).batch(config["batch_size"], drop_remainder=True)
# datasetv = tf.data.Dataset.from_generator(
# data_parserv.generator,
# (tf.float32, tf.float32),
# (tf.TensorShape([config["image_size"][0], config["image_size"][1], 3]), tf.TensorShape([config["image_size"][0], config["image_size"][1], 3]))
# ).batch(config["batch_size"], drop_remainder=True)
# logger = tk.callbacks.CSVLogger(config["logger_file"], append=True)
# model_cb = Custom_Callback(config)
# # lr_scheduler = tk.callbacks.ReduceLROnPlateau(monitor="val_loss")
# def lr_sched(epoch) :
# if epoch < 50 :
# return config["lr"]
# elif epoch < 100 :
# return 0.5 * config["lr"]
# else :
# return 0.25 * config["lr"]
# lr_scheduler = tf.keras.callbacks.LearningRateScheduler(lr_sched)
# print(model.model)
# print("model loaded")
# # print(data_parser.steps)
# model.model.fit(
# dataset,
# epochs=config["epoch"],
# callbacks=[model_cb, logger, lr_scheduler],
# # callbacks=[model_cb, logger],
# validation_data=datasetv,
# validation_freq=1,
# # steps_per_epoch=data_parser.steps,
# # class_weight=config["class_weight"],
# initial_epoch=config["present_epoch"]
# )
# model.model.save(str(Path(config["save_path"])/f"model_{config['epoch']}.h5"))
# %%
if False :
tmpx, tmpy = data_parser.get_batch()
with tf.GradientTape() as tape :
output = model.model(tmpx[:4, :, :, :], training=True)
loss = model.sce_loss(tmpy[:4, :, :, :], output)
grads = tape.gradient(loss, model.model.trainable_variables)
grads
# %%
for v in grads :
print(v.shape)