file_sep = r'/'
log_save_Path = Result_save_Path + file_sep[0] + 'log'
model_save_Path = Result_save_Path + file_sep[0] + 'model'
os.system('mkdir ' + log_save_Path)
os.system('mkdir ' + model_save_Path)
# 开始训练:一下代码由main改变而来,但是不需要保存模型
if __name__ == "__main__":
# prepare container:准备网络输入输出容器 暂时测试新getbatch函数,所以注释掉了
# data_input_c = np.zeros([batch_size] + data_input_shape + [1], dtype=np.float32) # net input container
# label_input_c = np.zeros([batch_size] + label_shape)
# 构建网络并compile
# d_model_1 = vgg16_w_3d(use_bias_flag=True,classes=2)
d_model = resnet_or(use_bias_flag=False, classes=2)
# d_model_1 = resnext(classes=2, use_bias_flag=True)
# d_model_1 = SparseNetImageNet121(classes = 2,activation='softmax',dropout_rate = 0.5)
# d_model_1 = se_dense_net(nb_layers=[6, 12, 24, 16], growth_rate=32, nb_filter=64, bottleneck=True, reduction=0.0, dropout_rate=None, subsample_initial_block=True)
if multi_gpu_mode:
d_model = multi_gpu_model(d_model, gpus=gpu_munum)
# d_model.compile(optimizer=adam(lr=init_lr), loss='categorical_crossentropy', metrics=[y_t, y_pre, Acc])
d_model.compile(optimizer=SGD(lr=init_lr, momentum=0.9),
loss='categorical_crossentropy',
metrics=[y_t, y_pre, Acc])
# pause() # identify
# print(d_model.summary()) # view net
# pause() # identify
# extra param initialization:初始化一些用来记录和显示的参数
import keras.backend as K
from w_resnet import resnet_nobn, se_resnet, resnet_or
import keras.backend as K
# step2: import extra model finished
# step2: import extra model finished
os.environ["CUDA_VISIBLE_DEVICES"] = "1"
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
set_session(tf.Session(config=config))
# 構建網絡
model = resnet_or(use_bias_flag=True, classes=2)
# model.compile(optimizer=SGD(lr=1e-6, momentum=0.9), loss='binary_crossentropy')
# 構建輸入
data_input_c = np.zeros([1, 280, 280, 16, 1], dtype=np.float32)
H5_file = h5py.File(r'/data/@data_liaoxiao/4test/22_1.h5', 'r')
batch_x = H5_file['data'][:]
batch_y = H5_file['label3'][:]
H5_file.close()
batch_x = np.transpose(batch_x, (1, 2, 0))
data_input_c[0, :, :, :, 0] = batch_x[:, :, :]
# 保存或加載權重
# pre = model.predict_on_batch(data_input_c)
# model.save('G:\qweqweqweqwe\model.h5')
model.load_weights(
filepath=
os.system('mkdir '+log_save_Path)
os.system('mkdir '+model_save_Path)
os.system('mkdir '+test_log_save_path)
os.system('mkdir '+ver_log_save_path)
# 开始训练:一下代码由main改变而来,但是不需要保存模型
if __name__ == "__main__":
# prepare container:准备网络输入输出容器 暂时测试新getbatch函数,所以注释掉了
# data_input_c = np.zeros([batch_size] + data_input_shape + [1], dtype=np.float32) # net input container
# label_input_c = np.zeros([batch_size] + label_shape)
# 构建网络并compile
# d_model_1 = vgg16_w_3d(use_bias_flag=True,classes=2)
d_model = resnet_or(use_bias_flag=False,classes=2,inputshape=data_input_shape)
# d_model_1 = resnext(classes=2, use_bias_flag=True)
# d_model_1 = SparseNetImageNet121(classes = 2,activation='softmax',dropout_rate = 0.5)
# d_model_1 = se_dense_net(nb_layers=[6, 12, 24, 16], growth_rate=32, nb_filter=64, bottleneck=True, reduction=0.0, dropout_rate=None, subsample_initial_block=True)
if multi_gpu_mode:
d_model = multi_gpu_model(d_model, gpus=gpu_munum)
d_model.compile(optimizer=adam(lr=init_lr), loss=categorical_crossentropy, metrics=[y_t, y_pre, Acc])
# d_model.compile(optimizer=SGD(lr=init_lr, momentum=0.9), loss=categorical_crossentropy, metrics=[y_t, y_pre, Acc])
# d_model.compile(optimizer=SGD(lr=init_lr,momentum=0.9), loss='categorical_crossentropy', metrics=[y_t, y_pre, Acc])
# pause() # identify
# print(d_model.summary()) # view net
# pause() # identify
# extra param initialization:初始化一些用来记录和显示的参数