try:
model.load_weights('LSTM+BN4--26--0.011.hdf5')
print("...Previous weight data...")
except:
print("...New weight data...")
pass
train_file_path = './DB/train/' ##该目录下是指定好的图片，没有其他的文件夹
tiger_train = TextImageGenerator(train_file_path, img_w, img_h, batch_size, downsample_factor)
tiger_train.build_data()
valid_file_path = './DB/test/'##该目录下是指定好的图片，没有其他的文件夹
tiger_val = TextImageGenerator(valid_file_path, img_w, img_h, val_batch_size, downsample_factor)
tiger_val.build_data()
ada = Adadelta()
early_stop = EarlyStopping(monitor='loss', min_delta=0.001, patience=4, mode='min', verbose=1)
checkpoint = ModelCheckpoint(filepath='LSTM+BN5--{epoch:02d}--{val_loss:.3f}.hdf5', monitor='loss', verbose=1, mode='min', period=1)
# the loss calc occurs elsewhere, so use a dummy lambda func for the loss
model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer=ada)
# captures output of softmax so we can decode the output during visualization
model.fit_generator(generator=tiger_train.next_batch(),
steps_per_epoch=int(tiger_train.n / batch_size),
epochs=30,
callbacks=[checkpoint],
validation_data=tiger_val.next_batch(),
validation_steps=int(tiger_val.n / val_batch_size))
patience=4,
mode='min',
verbose=1)
checkpoint = ModelCheckpoint(
filepath='LSTM+BN7--{epoch:02d}--{val_loss:.3f}.hdf5',
monitor='loss',
verbose=1,
mode='min',
period=1)
# the loss calc occurs elsewhere, so use a dummy lambda func for the loss
model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer=ada)
# captures output of softmax so we can decode the output during visualization
seqModel = model.fit_generator(
generator=tiger_train.next_batch(),
steps_per_epoch=int(num_train_images / batch_size),
epochs=15,
callbacks=[checkpoint],
validation_data=tiger_val.next_batch(),
validation_steps=int(num_validation_images / val_batch_size))
train_loss = seqModel.history['loss']
val_loss = seqModel.history['val_loss']
ep = range(1, 16)
plt.figure()
plt.plot(ep, train_loss, color='blue')
plt.xlabel('Epoch')
plt.ylabel('train_loss')
# valid_file_path = './DB/test/'
# tiger_val = TextImageGenerator(valid_file_path, img_w, img_h, val_batch_size, downsample_factor)
# tiger_val.build_data()
ada = tf.keras.optimizers.Adam()
early_stop = EarlyStopping(monitor='loss',
min_delta=0.001,
patience=4,
mode='min',
verbose=1)
checkpoint = ModelCheckpoint(
filepath='LSTM+BN5--{epoch:02d}--{val_loss:.3f}.hdf5',
monitor='loss',
verbose=1,
mode='min',
period=1)
# the loss calc occurs elsewhere, so use a dummy lambda func for the loss
model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer=ada)
model.summary()
# captures output of softmax so we can decode the output during visualization
model.fit_generator(
generator=tiger_train.next_batch(),
steps_per_epoch=100, #int(tiger_train.n / batch_size),
epochs=60,
callbacks=[checkpoint],
validation_data=tiger_train.next_batch_val(),
validation_steps=20,
) #int(tiger_train.n / val_batch_size))
print("...Previous weight data...")
except:
print("...New weight data...")
pass
train_file_path = './DB/synt_new/'
tiger_train = TextImageGenerator(train_file_path, img_w, img_h, batch_size, downsample_factor)
tiger_train.build_data()
#valid_file_path = './DB/test/'
#tiger_val = TextImageGenerator(valid_file_path, img_w, img_h, val_batch_size, downsample_factor)
#tiger_val.build_data()
ada = Adadelta()
early_stop = EarlyStopping(monitor='loss', min_delta=0.001, patience=4, mode='min', verbose=1)
#checkpoint = ModelCheckpoint(filepath='LSTM+BN5--{epoch:02d}--{val_loss:.3f}.hdf5', monitor='loss', verbose=1, mode='min', period=1)
checkpoint = ModelCheckpoint(filepath='new_model.hdf5', monitor='loss', verbose=1, mode='min', period=1)
# the loss calc occurs elsewhere, so use a dummy lambda func for the loss
model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer=ada)
# captures output of softmax so we can decode the output during visualization
model.fit_generator(generator=tiger_train.next_batch(),
steps_per_epoch=int(tiger_train.n / batch_size),
epochs=30,
callbacks=[checkpoint])
#validation_data=tiger_val.next_batch(),
#validation_steps=2)
#validation_steps=int(tiger_val.n / val_batch_size))