def main(test_desc_file, train_desc_file, load_dir):
import os
import tensorflow as tf
from keras.backend.tensorflow_backend import set_session
os.environ["CUDA_VISIBLE_DEVICES"] = "3"
# 进行配置，使用80%的GPU
config = tf.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.9
#config.gpu_options.allow_growth=True #不全部占满显存, 按需分配
set_session(tf.Session(config=config))
# Prepare the data generator
datagen = DataGenerator()
# Load the JSON file that contains the dataset
datagen.load_test_data(test_desc_file)
datagen.load_train_data(train_desc_file)
# Use a few samples from the dataset, to calculate the means and variance
# of the features, so that we can center our inputs to the network
# datagen.fit_train_test(load_dir)
datagen.fit_train(100)
# Compile a Recurrent Network with 1 1D convolution layer, GRU units
# and 1 fully connected layer
model = load_model(load_dir)
# Compile the CTC decode function
decode_fn = decode_ctc_fn(model)
# Compile the testing function
test_fn = compile_test_fn(model)
# Test the model
test_loss = test(model, test_fn, decode_fn, datagen)
print ("Test loss: {}".format(test_loss))
def main(train_desc_file, val_desc_file, epochs, save_dir, sortagrad):
if not os.path.exists(save_dir):
os.makedirs(save_dir)
# Configure logging
configure_logging(file_log_path=os.path.join(save_dir, 'train_log.txt'))
# Prepare the data generator
datagen = DataGenerator()
# Load the JSON file that contains the dataset
datagen.load_train_data(train_desc_file)
datagen.load_validation_data(val_desc_file)
# Use a few samples from the dataset, to calculate the means and variance
# of the features, so that we can center our inputs to the network
datagen.fit_train(100)
# Compile a Recurrent Network with 1 1D convolution layer, GRU units
# and 1 fully connected layer
model = compile_gru_model(recur_layers=3, nodes=1000, batch_norm=True)
# Compile the CTC training function
train_fn = compile_train_fn(model)
# Compile the validation function
val_fn = compile_test_fn(model)
# Train the model
train(model, train_fn, val_fn, datagen, save_dir, epochs=epochs,
do_sortagrad=sortagrad)
def main(test_desc_file, train_desc_file, load_dir, result_file='test_result'):
datagen = DataGenerator()
datagen.load_test_data(test_desc_file)
datagen.load_train_data(train_desc_file)
datagen.fit_train(100)
# load model
model = load_model(load_dir)
test_fn = compile_test_fn(model)
# test begin
start = time.time()
test(model, test_fn, datagen, result_file)
print('elapsed: %s' % (time.time() - start))
def main(train_desc_file, val_desc_file, epochs, save_dir, sortagrad):
if not os.path.exists(save_dir):
os.makedirs(save_dir)
# Configure logging
configure_logging(file_log_path=os.path.join(save_dir, 'train_log.txt'))
# Prepare the data generator
datagen = DataGenerator()
# Load the JSON file that contains the dataset
datagen.load_train_data(train_desc_file)
datagen.load_validation_data(val_desc_file)
# Use a few samples from the dataset, to calculate the means and variance
# of the features, so that we can center our inputs to the network
feats_mean, feats_std = datagen.fit_train(100)
np.savetxt(save_dir + '/feats_mean.txt', feats_mean, fmt='%0.8f')
np.savetxt(save_dir + '/feats_std.txt', feats_std, fmt='%0.8f')
# Compile a Recurrent Network with 1 1D convolution layer, GRU units
# and 1 fully connected layer
model = compile_gru_model(recur_layers=3, batch_norm=True)
# 输出模型结构
logger.info('*' * 20)
model_summary = model.summary()
logger.info(model_summary)
logger.info('*' * 20)
# 加载已训练好的模型
model.load_weights("model_0621_zhongjin_0624/model_46000_weights.h5")
# Compile the CTC training function
train_fn = compile_train_fn(model)
# Compile the validation function
val_fn = compile_test_fn(model)
# Compile the CTC decode function
decode_fn = decode_ctc_fn(model)
# Train the model
train(model,
train_fn,
val_fn,
decode_fn,
datagen,
save_dir,
epochs=epochs,
do_sortagrad=sortagrad)
def main(test_desc_file, train_desc_file, load_dir):
# Prepare the data generator
datagen = DataGenerator()
# Load the JSON file that contains the dataset
datagen.load_test_data(test_desc_file)
datagen.load_train_data(train_desc_file)
# Use a few samples from the dataset, to calculate the means and variance
# of the features, so that we can center our inputs to the network
datagen.fit_train(100)
# Compile a Recurrent Network with 1 1D convolution layer, GRU units
# and 1 fully connected layer
model = load_model(load_dir)
# Compile the testing function
test_fn = compile_test_fn(model)
# Test the model
test_loss = test(model, test_fn, datagen)
print("Test loss: {}".format(test_loss))