def train(self,
img_folder,
ann_folder,
nb_epoch,
saved_weights_name,
batch_size=8,
jitter=True,
learning_rate=1e-4,
train_times=1,
valid_times=1,
valid_img_folder="",
valid_ann_folder="",
first_trainable_layer=None,
is_only_detect=False):
# 1. get annotations
train_annotations, valid_annotations = get_train_annotations(
self._labels, img_folder, ann_folder, valid_img_folder,
valid_ann_folder, is_only_detect)
print("No error getting annotations")
# 1. get batch generator
train_batch_generator = self._get_batch_generator(train_annotations,
batch_size,
train_times,
jitter=jitter)
valid_batch_generator = self._get_batch_generator(valid_annotations,
batch_size,
valid_times,
jitter=False)
print("No error getting batch generator")
# 2. To train model get keras model instance & loss fucntion
model = self._yolo_network.get_model(first_trainable_layer)
loss = self._get_loss_func(batch_size)
print("No error getting keras model instance & loss fucntion")
# 3. Run training loop
train(model,
loss,
train_batch_generator,
valid_batch_generator,
learning_rate=learning_rate,
nb_epoch=nb_epoch,
saved_weights_name=saved_weights_name)
print("No error running training loop")
def train(self,
img_folder,
ann_folder,
nb_epoch,
saved_weights_name,
batch_size=8,
jitter=True,
learning_rate=1e-4,
train_times=1,
valid_times=1,
warmup_epochs=None,
valid_img_folder="",
valid_ann_folder="",
first_trainable_layer=None,
is_only_detect=False):
# 1. get annotations
train_annotations, valid_annotations = get_train_annotations(self._labels,
img_folder,
ann_folder,
valid_img_folder,
valid_ann_folder,
is_only_detect)
# 1. get batch generator
train_batch_generator = self._get_batch_generator(train_annotations, batch_size, train_times, jitter=jitter)
valid_batch_generator = self._get_batch_generator(valid_annotations, batch_size, valid_times, jitter=False)
# 2. To train model get keras model instance & loss fucntion
model = self._yolo_network.get_model(first_trainable_layer)
loss = self._get_loss_func(batch_size,
warmup_epochs,
train_times,
valid_times)
# 3. Run training loop
train(model,
loss,
train_batch_generator,
valid_batch_generator,
learning_rate = learning_rate,
nb_epoch = nb_epoch,
saved_weights_name = saved_weights_name)
def train(
self,
img_folder,
ann_folder,
img_in_mem, # datasets in mem, format: list
ann_in_mem, # datasets's annotation in mem, format: list
nb_epoch,
save_best_weights_path,
save_final_weights_path,
batch_size=8,
jitter=True,
learning_rate=1e-4,
train_times=1,
valid_times=1,
valid_img_folder="",
valid_ann_folder="",
valid_img_in_mem=None,
valid_ann_in_mem=None,
first_trainable_layer=None,
is_only_detect=False,
progress_callbacks=[]):
# 1. get annotations
train_annotations, valid_annotations = get_train_annotations(
self._labels, img_folder, ann_folder, valid_img_folder,
valid_ann_folder, img_in_mem, ann_in_mem, valid_img_in_mem,
valid_ann_in_mem, is_only_detect)
# 1. get batch generator
train_batch_generator = self._get_batch_generator(train_annotations,
batch_size,
train_times,
jitter=jitter)
valid_batch_generator = self._get_batch_generator(valid_annotations,
batch_size,
valid_times,
jitter=False)
# 2. To train model get keras model instance & loss function
model = self._yolo_network.get_model(first_trainable_layer)
loss = self._get_loss_func(batch_size)
# 3. Run training loop
history = train(model,
loss,
train_batch_generator,
valid_batch_generator,
learning_rate=learning_rate,
nb_epoch=nb_epoch,
save_best_weights_path=save_best_weights_path,
save_final_weights_path=save_final_weights_path,
progress_callbacks=progress_callbacks)
return history
