def train_lstm(self):
"""
Train LSTM pipeline using pre-generated CNN codes.
:param:
:return:
"""
codesTrainData = json.load(open(self.params.files['lstm_training_struct']))
codesStats = json.load(open(self.params.files['cnn_codes_stats']))
metadataStats = json.load(open(self.params.files['dataset_stats']))
model = get_lstm_model(self.params, codesStats)
#model = make_parallel(model, 4)
model.compile(optimizer=Adam(lr=self.params.lstm_adam_learning_rate), loss='categorical_crossentropy', metrics=['accuracy'])
train_datagen = codes_metadata_generator(self.params, codesTrainData, metadataStats, codesStats)
print("training")
filePath = os.path.join(self.params.directories['lstm_checkpoint_weights'], 'weights.{epoch:02d}.hdf5')
checkpoint = ModelCheckpoint(filepath=filePath, monitor='loss', verbose=0, save_best_only=False, save_weights_only=False, mode='auto', period=1)
callbacks_list = [checkpoint]
#model.fit_generator(train_datagen,
# steps_per_epoch=(len(codesTrainData) / self.params.batch_size_lstm + 1),
# epochs=self.params.lstm_epochs, callbacks=callbacks_list,
# max_queue_size=20)
model.fit_generator(train_datagen,
steps_per_epoch=(len(codesTrainData) / self.params.batch_size_lstm + 1),
epochs=self.params.lstm_epochs, max_queue_size=20)
model.save(self.params.files['lstm_model'])
def train_multi(self):
"""
Train LSTM pipeline using pre-generated CNN codes.
:param:
:return:
"""
allCodesTrainingData = json.load(
open(self.params.files['multi_training_struct']))
# add 50% of the /val/ data (which has false_detection instances) to train dataset, leave rest for validation
codesTrainData = {}
codesValidData = {}
for k, v in allCodesTrainingData.iteritems():
if k.startswith('val/'):
if np.random.randint(2):
codesTrainData[k] = v
else:
codesValidData[k] = v
else:
codesTrainData[k] = v
assert len(
allCodesTrainingData) == len(codesTrainData) + len(codesValidData)
codesStats = json.load(open(self.params.files['cnn_codes_stats']))
if self.params.max_temporal != 0:
codesStats['max_temporal'] = self.params.max_temporal
metadataStats = json.load(open(self.params.files['dataset_stats']))
loaded_filename = None
if self.params.args.load_model:
model = load_model(self.params.args.load_model)
loaded_filename = os.path.basename(self.params.args.load_model)
else:
model = get_multi_model(self.params, codesStats)
if self.params.args.load_weights:
model.load_weights(self.params.args.load_weights, by_name=True)
loaded_filename = os.path.basename(self.params.args.load_weights)
initial_epoch = self.get_initial_epoch(loaded_filename)
if self.params.print_model_summary:
model.summary()
model = multi_gpu_model(model, gpus=self.params.gpus)
model.compile(optimizer=RMSprop(lr=self.params.learning_rate),
loss='categorical_crossentropy',
metrics=['accuracy'])
preffix = self.get_preffix()
print("training multi-image model: " + preffix)
filePath = os.path.join(
self.params.directories['multi_checkpoint_weights'],
preffix + '-epoch_' + '{epoch:02d}' + '-acc_' + '{acc:.4f}' +
'-val_acc_' + '{val_acc:.4f}.hdf5')
checkpoint = ModelCheckpoint(filepath=filePath,
monitor='loss',
verbose=1,
save_best_only=False,
save_weights_only=False,
mode='auto',
period=1)
reduce_lr = ReduceLROnPlateau(monitor='val_acc',
factor=0.1,
patience=1,
min_lr=1e-7,
epsilon=0.01,
verbose=1)
model.fit_generator(generator=codes_metadata_generator(self.params, \
codesTrainData, metadataStats, codesStats, \
class_aware_sampling = not self.params.leave_unbalanced, \
temporal_dropout = self.params.temporal_dropout),
steps_per_epoch=int(math.ceil((len(codesTrainData) / self.params.batch_size))),
epochs=self.params.epochs,
callbacks=[checkpoint, FMOW_Callback(), reduce_lr],
initial_epoch = initial_epoch,
validation_data = codes_metadata_generator(self.params, \
codesValidData, metadataStats, codesStats, \
class_aware_sampling = False,\
temporal_dropout = 0.),
validation_steps = int(math.ceil((len(codesValidData) / self.params.batch_size))),
)
model.save(self.params.files['multi_model'])