monitor='val_loss',
save_weights_only=True,
save_best_only=False,
period=1)
reduce_lr = ReduceLROnPlateau(monitor='val_loss',
factor=0.5,
patience=2,
verbose=1)
early_stopping = EarlyStopping(monitor='val_loss',
min_delta=0,
patience=6,
verbose=1)
BATCH_SIZE = 2
gen = Generator(bbox_util, BATCH_SIZE, lines[:num_train],
lines[num_train:], (input_shape[0], input_shape[1]),
NUM_CLASSES)
for i in range(174):
model.layers[i].trainable = False
model.compile(loss={
'regression': smooth_l1(),
'classification': focal()
},
optimizer=keras.optimizers.Adam(lr=1e-4, clipnorm=0.001))
model.fit_generator(
gen.generate(True),
steps_per_epoch=num_train // BATCH_SIZE,
validation_data=gen.generate(False),
#------------------------------------------------------#
# 主干特征提取网络特征通用，冻结训练可以加快训练速度
# 也可以在训练初期防止权值被破坏。
# Init_Epoch为起始世代
# Freeze_Epoch为冻结训练的世代
# Epoch总训练世代
#------------------------------------------------------#
if True:
Init_epoch = 0
Freeze_epoch = 50
# batch_size大小，每次喂入多少数据
batch_size = 8
# 最大学习率
learning_rate_base = 1e-3
gen = Generator(training_dataset_path,img_dim,batch_size,bbox_util)
model.compile(loss={
'bbox_reg' : box_smooth_l1(),
'cls' : conf_loss(),
'ldm_reg' : ldm_smooth_l1()
},optimizer=keras.optimizers.Adam(lr=learning_rate_base)
)
model.fit(gen.generate(False),
steps_per_epoch=gen.get_len()//batch_size,
verbose=1,
epochs=Freeze_epoch,
initial_epoch=Init_epoch,
callbacks=[logging, checkpoint, reduce_lr, early_stopping])