model = SSD300(input_shape, NUM_CLASS)
model.load_weights("model_data/ssd_weights.h5", by_name = True, skip_mismatch=True)
log = TensorBoard(log_dir = log_path)
checkpoint = ModelCheckpoint(log_path + 'ep{epoch:03d}-loss{loss:.3f}-val_loss{val_loss:.3f}.h5',
monitor='val_loss', save_weights_only=True, save_best_only=True, 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 = 16
priors = pickle.load(open('model_data/prior_boxes_ssd300.pkl','rb'))
bbox_util = BBoxUtility(NUM_CLASS,priors)
gen = Generator(bbox_util, batch_size, lines[:num_train],lines[num_train:],(input_shape[0],input_shape[1]), NUM_CLASS, do_crop=True)
if True:
model.compile(optimizer=Adam(1e-5),loss=MultiboxLoss(NUM_CLASS,neg_pos_ratio=2.0).compute_loss)
model.fit_generator(gen.generate(True),
steps_per_epoch=num_train/batch_size,
validation_data=gen.generate(False),
validation_steps=num_val/batch_size,
epochs=30,
initial_epoch=0,
callbacks=[log, checkpoint, reduce_lr, early_stopping])
model = SSD300(input_shape, num_classes=NUM_CLASSES)
model.load_weights('logs/ep052-loss0.561-val_loss0.968.h5', by_name=True,skip_mismatch=True)
# 训练参数设置
logging = tf.keras.callbacks.TensorBoard()
checkpoint = tf.keras.callbacks.ModelCheckpoint(log_dir + 'ep{epoch:03d}-loss{loss:.3f}-val_loss{val_loss:.3f}.h5',
monitor='loss', save_weights_only=True, save_best_only=True, period=1)
reduce_lr = tf.keras.callbacks.ReduceLROnPlateau(monitor='loss', factor=0.5, patience=2, verbose=1)
early_stopping = tf.keras.callbacks.EarlyStopping(monitor='loss', min_delta=0, patience=16, verbose=1)
BATCH_SIZE = 12
gen = Generator(bbox_util, BATCH_SIZE, lines[:num_train], lines[num_train:],
(input_shape[0], input_shape[1]),NUM_CLASSES, do_crop=True)
if False:
model.compile(optimizer=tf.keras.optimizers.Adam(lr=1e-4),loss=MultiboxLoss(NUM_CLASSES, neg_pos_ratio=5.0).compute_loss)
model.fit_generator(gen.generate(True),
steps_per_epoch=num_train//BATCH_SIZE,
validation_data=gen.generate(False),
validation_steps=num_val//BATCH_SIZE,
epochs=20,
initial_epoch=0,
callbacks=[logging, checkpoint, reduce_lr, early_stopping])
if True:
model.compile(optimizer=tf.keras.optimizers.Adamax(lr=1e-5),loss=MultiboxLoss(NUM_CLASSES, neg_pos_ratio=3.0).compute_loss)
model.fit_generator(gen.generate(True),
steps_per_epoch=num_train//BATCH_SIZE,
validation_data=gen.generate(False),
validation_steps=num_val//BATCH_SIZE,
epochs=250,
Freeze_epoch = 50
BATCH_SIZE = 16
learning_rate_base = 5e-4
gen = Generator(bbox_util, BATCH_SIZE, lines[:num_train],
lines[num_train:], (input_shape[0], input_shape[1]),
NUM_CLASSES)
epoch_size = num_train // BATCH_SIZE
epoch_size_val = num_val // BATCH_SIZE
if epoch_size == 0 or epoch_size_val == 0:
raise ValueError("数据集过小，无法进行训练，请扩充数据集。")
model.compile(optimizer=Adam(lr=learning_rate_base),
loss=MultiboxLoss(NUM_CLASSES,
neg_pos_ratio=3.0).compute_loss)
model.fit_generator(gen.generate(True),
steps_per_epoch=epoch_size,
validation_data=gen.generate(False),
validation_steps=epoch_size_val,
epochs=Freeze_epoch,
initial_epoch=Init_epoch,
callbacks=[
logging, checkpoint, reduce_lr, early_stopping,
loss_history
])
for i in range(80):
model.layers[i].trainable = True
if True:
Freeze_epoch = 50
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 = 4
gen = Generator(bbox_util, BATCH_SIZE, lines[:num_train], lines[num_train:],
(input_shape[0], input_shape[1]),NUM_CLASSES)
#------------------------------------------------------#
# 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。、
# 提示OOM或者显存不足请调小Batch_size
#------------------------------------------------------#
for i in range(21):
model.layers[i].trainable = False
if True:
model.compile(optimizer=Adam(lr=5e-4),loss=MultiboxLoss(NUM_CLASSES, neg_pos_ratio=3.0).compute_loss)
model.fit_generator(gen.generate(True),
steps_per_epoch=num_train//BATCH_SIZE,
validation_data=gen.generate(False),
validation_steps=num_val//BATCH_SIZE,
epochs=15,
initial_epoch=0,
callbacks=[logging, checkpoint, reduce_lr, early_stopping])
if True:
model.compile(optimizer=Adam(lr=2e-4),loss=MultiboxLoss(NUM_CLASSES, neg_pos_ratio=3.0).compute_loss)
model.fit_generator(gen.generate(True),
steps_per_epoch=num_train//BATCH_SIZE,
validation_data=gen.generate(False),
validation_steps=num_val//BATCH_SIZE,
epochs=30,
initial_epoch=15,
# Freeze_Epoch为冻结训练的世代
# Epoch总训练世代
# 提示OOM或者显存不足请调小Batch_size
#------------------------------------------------------#
for i in range(21):
model.layers[i].trainable = False
if True:
Init_epoch = 0
Freeze_epoch = 50
BATCH_SIZE = 16
learning_rate_base = 5e-4
gen = Generator(bbox_util, BATCH_SIZE, lines[:num_train], lines[num_train:],
(input_shape[0], input_shape[1]),NUM_CLASSES)
model.compile(optimizer=Adam(lr=learning_rate_base),loss=MultiboxLoss(NUM_CLASSES, neg_pos_ratio=3.0).compute_loss)
model.fit_generator(gen.generate(True),
steps_per_epoch=num_train//BATCH_SIZE,
validation_data=gen.generate(False),
validation_steps=num_val//BATCH_SIZE,
epochs=Freeze_epoch,
initial_epoch=Init_epoch,
callbacks=[logging, checkpoint, reduce_lr, early_stopping])
for i in range(21):
model.layers[i].trainable = True
if True:
Freeze_epoch = 50
Epoch = 100
BATCH_SIZE = 8
learning_rate_base = 1e-4
else:
no_load_key.append(k)
model_dict.update(temp_dict)
model.load_state_dict(model_dict)
#------------------------------------------------------#
# 显示没有匹配上的Key
#------------------------------------------------------#
if local_rank == 0:
print("\nSuccessful Load Key:", str(load_key)[:500], "……\nSuccessful Load Key Num:", len(load_key))
print("\nFail To Load Key:", str(no_load_key)[:500], "……\nFail To Load Key num:", len(no_load_key))
print("\n\033[1;33;44m温馨提示，head部分没有载入是正常现象，Backbone部分没有载入是错误的。\033[0m")
#----------------------#
# 获得损失函数
#----------------------#
criterion = MultiboxLoss(num_classes, neg_pos_ratio=3.0)
#----------------------#
# 记录Loss
#----------------------#
if local_rank == 0:
time_str = datetime.datetime.strftime(datetime.datetime.now(),'%Y_%m_%d_%H_%M_%S')
log_dir = os.path.join(save_dir, "loss_" + str(time_str))
loss_history = LossHistory(log_dir, model, input_shape=input_shape)
else:
loss_history = None
#------------------------------------------------------------------#
# torch 1.2不支持amp，建议使用torch 1.7.1及以上正确使用fp16
# 因此torch1.2这里显示"could not be resolve"
#------------------------------------------------------------------#
if fp16:
BATCH_SIZE = 16
gen = Generator(bbox_util, BATCH_SIZE, lines[:num_train], lines[num_train:],
(input_shape[0], input_shape[1]),NUM_CLASSES)
#------------------------------------------------------#
# 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。
# 提示OOM或者显存不足请调小Batch_size
#------------------------------------------------------#
for i in range(81):
model.layers[i].trainable = False
if True:
model.compile(optimizer=Adam(lr=5e-4, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0),
loss=MultiboxLoss(NUM_CLASSES, neg_pos_ratio=3.0).compute_loss)
model.fit_generator(gen.generate(True),
steps_per_epoch=num_train//BATCH_SIZE,
validation_data=gen.generate(False),
validation_steps=num_val//BATCH_SIZE,
epochs=50,
initial_epoch=0,
callbacks=[logging, checkpoint, reduce_lr, early_stopping])
for i in range(81):
model.layers[i].trainable = True
if True:
model.compile(optimizer=Adam(lr=1e-4, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0),loss=MultiboxLoss(NUM_CLASSES, neg_pos_ratio=3.0).compute_loss)
# model.compile(optimizer=SGD(lr=1e-4,momentum=0.9,decay=5e-4),loss=MultiboxLoss(NUM_CLASSES, neg_pos_ratio=3.0).compute_loss)
model.fit_generator(gen.generate(True),
steps_per_epoch=num_train//BATCH_SIZE,
print('Load weights {}.'.format(model_path))
model.load_weights(model_path,
by_name=True,
skip_mismatch=True)
else:
model = SSD300((input_shape[0], input_shape[1], 3),
num_classes,
weight_decay=weight_decay)
if model_path != '':
#------------------------------------------------------#
# 载入预训练权重
#------------------------------------------------------#
print('Load weights {}.'.format(model_path))
model.load_weights(model_path, by_name=True, skip_mismatch=True)
multiloss = MultiboxLoss(num_classes, neg_pos_ratio=3.0).compute_loss
#---------------------------#
# 读取数据集对应的txt
#---------------------------#
with open(train_annotation_path, encoding='utf-8') as f:
train_lines = f.readlines()
with open(val_annotation_path, encoding='utf-8') as f:
val_lines = f.readlines()
num_train = len(train_lines)
num_val = len(val_lines)
show_config(
classes_path = classes_path, model_path = model_path, input_shape = input_shape, \
Init_Epoch = Init_Epoch, Freeze_Epoch = Freeze_Epoch, UnFreeze_Epoch = UnFreeze_Epoch, Freeze_batch_size = Freeze_batch_size, Unfreeze_batch_size = Unfreeze_batch_size, Freeze_Train = Freeze_Train, \
Init_lr = Init_lr, Min_lr = Min_lr, optimizer_type = optimizer_type, momentum = momentum, lr_decay_type = lr_decay_type, \