这是RSNA2017的一个挑战赛，在kaggle上下载了数据集
RSNA Bone Age | Kagglehttps://www.kaggle.com/kmader/rsna-bone-agegithub上也有一些代码，但是因为配置问题，都没有运行成功，也参考了一篇文章(15条消息) 【蓝蜗牛】骨龄检测（一）_新手村打蘑菇-CSDN博客_kaggle骨龄https://blog.csdn.net/weixin_43346901/article/details/99678300
下面是我参考了一些资料后写的一些代码
首先是用Xception网络
from tensorflow.keras.models import Model
from keras_preprocessing.image import ImageDataGenerator
from tensorflow.keras.layers import Dense, Dropout, BatchNormalization
from tensorflow.keras.layers import Input, Conv2D, multiply, LocallyConnected2D, Lambda, Flatten, concatenate
from tensorflow.keras.layers import GlobalAveragePooling2D, AveragePooling2D, MaxPooling2D
from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping, ReduceLRonPlateau
from tensorflow.keras import optimizers
from tensorflow.keras.metrics import mean_absolute_error
from tensorflow.keras.applications import Xception
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
import os
import matplotlib.pyplot as plt
# import seaborn as sns
import tensorflow as tf
os.environ["CUDA_VISIBLE_DEVICES"] = '1' # use GPU with ID=0
config = tf.compat.v1.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.5 # maximun alloc gpu50% of MEM
config.gpu_options.allow_growth = True # allocate dynamically
sess = tf.compat.v1.Session(config=config)
# %%
EPOCHS = 30
LEARNING_RATE = 0.001
BATCH_SIZE_TRAIN = 16 # 出现OOM，调小这个数值 8
BATCH_SIZE_VAL = 16 # 16
# 图像参数
PIXELS = 299 # Xception输入大小
CHANNELS = 3
IMG_SIZE = (PIXELS, PIXELS)
IMG_DIMS = (PIXELS, PIXELS, CHANNELS)
VALIDATION_FRACTION = 0.25
SEED = 1234
# %%
# 读取数据
path = '/home/user/X-ray/archive/'
train_path = path + 'boneage-training-dataset/' # 图像文件夹
# test_path = path + 'boneage-test-dataset/'
df = pd.read_csv(path + 'boneage-training-dataset.csv') # csv标注文件
files = [train_path + str(i) + '.png' for i in df['id']]
df['file'] = files
df['exists'] = df['file'].map(os.path.exists)
boneage_mean = df['boneage'].mean()
boneage_div = 2 * df['boneage'].std()
df['boneage_zscore'] = df['boneage'].map(lambda x:
(x - boneage_mean) / boneage_div)
df.dropna(inplace=True)
df['gender'] = df['male'].map(lambda x: 1 if x else 0)
df['boneage_category'] = pd.cut(df['boneage'], 10)
# Examine the distribution of age and gender
print("{} images found out of total {} images".format(df['exists'].sum(), df.shape[0]))
column_headers = list(df.columns.values)
print("column_handers = ", column_headers) # 列标签
print(df.sample(5)) # csv中随机抽取5行
# 导出数据 生成csv
test = pd.Dataframe(columns=column_headers, data=df)
test.to_csv('/home/user/tanminhui/X-ray/archive/test_df.csv') # 如果生成excel，可以用to_excel
# raw_train_df, test_df = train_test_split(df,
# test_size=0.2,
# random_state=2018,
# stratify=df['boneage_category'])
# raw_train_df, valid_df = train_test_split(raw_train_df,
# test_size=0.1,
# random_state=2018,
# stratify=raw_train_df['boneage_category'])
raw_train_df, raw_valid_df = train_test_split(df, test_size=0.25, random_state=1234,
stratify=df['boneage_category'])
train_df = raw_train_df.groupby(['boneage_category', 'male']).apply(lambda x: x.sample(500, replace=True)).reset_index(
drop=True)
valid_df, test_df = train_test_split(raw_valid_df, test_size=0.25, random_state=1234)
raw_train_df_size = raw_train_df.shape[0]
valid_size = valid_df.shape[0]
test_size = test_df.shape[0]
print("# Training images: {}".format(raw_train_df))
print("# Validation images: {}".format(valid_size))
print("# Testing images: {}".format(test_size))
optim = optimizers.Nadam(lr=LEARNING_RATE, beta_1=0.9, beta_2=0.999, epsilon=1e-08, schedule_decay=0.0003)
weight_path = "{}_weights.best.hdf5".format('bone_age')
checkpoint = ModelCheckpoint(weight_path, monitor='val_loss', verbose=1, save_best_only=True, mode='min',
save_weights_only=True)
reduceLRonPlat = ReduceLRonPlateau(monitor='val_loss', factor=0.8, patience=3, verbose=1, mode='auto', min_delta=0.0001,
cooldown=5, min_lr=0.00006)
early = EarlyStopping(monitor="val_loss", mode="min", patience=6)
callbacks_list = [checkpoint, early, reduceLROnPlat]
def gen_2inputs(imgDatGen, df, batch_size, seed, img_size):
gen_img = imgDatGen.flow_from_dataframe(dataframe=df,
x_col='file', y_col='boneage_zscore',
batch_size=batch_size, seed=seed, shuffle=True, class_mode='other',
target_size=img_size, color_mode='rgb',
drop_duplicates=False)
gen_gender = imgDatGen.flow_from_dataframe(dataframe=df,
x_col='file', y_col='gender',
batch_size=batch_size, seed=seed, shuffle=True, class_mode='other',
target_size=img_size, color_mode='rgb',
drop_duplicates=False)
while True:
X1i = gen_img.next()
X2i = gen_gender.next()
yield [X1i[0], X2i[1]], X1i[1]
def test_gen_2inputs(imgDatGen, df, batch_size, img_size):
gen_img = imgDatGen.flow_from_dataframe(dataframe=df,
x_col='file', y_col='boneage_zscore',
batch_size=batch_size, shuffle=False, class_mode='other',
target_size=img_size, color_mode='rgb',
drop_duplicates=False)
gen_gender = imgDatGen.flow_from_dataframe(dataframe=df,
x_col='file', y_col='gender',
batch_size=batch_size, shuffle=False, class_mode='other',
target_size=img_size, color_mode='rgb',
drop_duplicates=False)
while True:
X1i = gen_img.next()
X2i = gen_gender.next()
yield [X1i[0], X2i[1]], X1i[1]
train_idg = ImageDataGenerator(zoom_range=0.2,
fill_mode='nearest',
rotation_range=25,
width_shift_range=0.25,
height_shift_range=0.25,
vertical_flip=False,
horizontal_flip=True,
shear_range=0.2,
samplewise_center=False,
samplewise_std_normalization=False)
val_idg = ImageDataGenerator(width_shift_range=0.25,
height_shift_range=0.25,
horizontal_flip=True)
test_idg = ImageDataGenerator()
train_flow = gen_2inputs(train_idg, train_df, BATCH_SIZE_TRAIN, SEED, IMG_SIZE)
valid_flow = gen_2inputs(val_idg, valid_df, BATCH_SIZE_VAL, SEED, IMG_SIZE)
test_flow = test_gen_2inputs(test_idg, test_df, 500, IMG_SIZE)
def mae_months(in_gt, in_pred):
return mean_absolute_error(boneage_div * in_gt, boneage_div * in_pred)
# 构建卷积神经网络
in_layer_img = Input(shape=IMG_DIMS, name='input_img')
in_layer_gender = Input(shape=(1,), name='input_gender')
base = Xception(input_shape=IMG_DIMS, weights='imagenet', include_top=False)
base_out = base(in_layer_img)
# base = GlobalAveragePooling2D()(base)
base = Dropout(0.5)(base_out)
bn_base = BatchNormalization()(base)
con_layer = Conv2D(512, kernel_size=(1, 1), padding='same', activation='relu')(bn_base)
con_layer = Dropout(0.5)(con_layer) # 要是跑不通，就注释这些
con_layer = Conv2D(512, kernel_size=(1, 1), padding='same', activation='relu')(con_layer) # 64
con_layer = Dropout(0.5)(con_layer) # 要是跑不通，就注释这些
feature_img = GlobalAveragePooling2D()(con_layer)
feature_gender = Dense(32, activation='relu')(in_layer_gender)
feature = concatenate([feature_img, feature_gender], axis=1)
out = Dense(512, activation='relu')(feature)
out=Dropout(0.5)(out)
out = Dense(512, activation='relu')(out)
out=Dropout(0.5)(out)
out = Dense(1, activation='linear')(out)
model = Model(inputs=[in_layer_img, in_layer_gender], outputs=out)
model.compile(loss='mean_absolute_error', optimizer=optim, metrics=[mae_months])
model.summary()
# from keras.utils import plot_model
# plot_model(model, show_shapes=True, to_file='model.png')
BATCH_SIZE_TEST = len(test_df) // 3
STEP_SIZE_TEST = 3
STEP_SIZE_TRAIN = len(train_df) // BATCH_SIZE_TRAIN
STEP_SIZE_VALID = len(valid_df) // BATCH_SIZE_VAL
model_history = model.fit_generator(generator=train_flow,
steps_per_epoch=STEP_SIZE_TRAIN,
validation_data=valid_flow,
validation_steps=STEP_SIZE_VALID,
epochs=EPOCHS,
callbacks=callbacks_list)
loss_history = model_history.history['loss']
history_df = pd.Dataframe.from_dict(model_history.history)
history_df.to_csv('/home/user/X-ray/xception_loss_history.csv')
# %%
# 测试
model.load_weights("bone_age_weights.best.hdf5")
test_X, test_Y = next(test_flow)
# ------------------------------------------------------------------------------------------------
# column_headers_X = list(test_X.columns.values)
# print("column_handers = ", column_headers_X) # 列标签
# print(test_X.sample(5)) # csv中随机抽取5行
# # 导出数据 生成csv
# test_X_CSV = pd.Dataframe(columns=column_headers_X, data=test_X)
# test_X_CSV.to_csv('D:/test_X_CSV.csv') # 如果生成excel，可以用to_excel
#
# column_headers_Y = list(test_Y.columns.values)
# print("column_handers = ", column_headers_Y) # 列标签
# print(test_Y.sample(5)) # csv中随机抽取5行
# # 导出数据 生成csv
# test_Y_CSV = pd.Dataframe(columns=column_headers_Y, data=test_Y)
# test_Y_CSV.to_csv('D:/test_Y_CSV.csv') # 如果生成excel，可以用to_excel
# --------------------------------------------------------------------------------------------------------------------
# plt.style.use("dark_background")
plt.rcParams['font.family'] = 'sans-serif'
plt.rcParams['font.sans-serif'] = 'DejaVu Sans'
pred_Y = boneage_div * model.predict(test_X, batch_size=16, verbose=True) + boneage_mean
test_Y_months = boneage_div * test_Y + boneage_mean
from sklearn.metrics import mean_absolute_error as sk_mae
print("Mean absolute error on test data: " + str(sk_mae(test_Y_months, pred_Y)))
# # 导出数据 生成csv
# column_headers = list(test_Y.columns.values)
# column_headers.append('test_Y_months', 'pred_Y')
# data_data = [test_Y, test_Y_months, pred_Y]
# test_pred = pd.Dataframe(columns=column_headers, data=data_data)
# test_pred.to_csv('D:/test_pred.csv') # 如果生成excel，可以用to_excel
fig, ax1 = plt.subplots(1, 1, figsize=(6, 6))
ax1.plot(test_Y_months, pred_Y, 'b+', label='predictions')
ax1.plot(test_Y_months, test_Y_months, 'r-', label='actual')
ax1.legend()
ax1.set_xlabel('Actual Age (Months)')
ax1.set_ylabel('Predicted Age (Months)')
fig.savefig('test_result.png', dpi=300)
# test_X 四维数组
print('test_X', test_X)
print('test_Y', test_Y)
print('pred_Y',pred_Y)
print('test_Y_months',test_Y_months)
ord_idx = np.argsort(test_Y)
ord_idx = ord_idx[np.linspace(0, len(ord_idx) - 1, num=8).astype(int)] # take 8 evenly spaced ones
fig, m_axs = plt.subplots(2, 4, figsize=(16, 32))
for (idx, c_ax) in zip(ord_idx, m_axs.flatten()):
c_ax.imshow(test_X[0][idx, :, :, 0], cmap='bone')
title = 'Age: %2.1fnPredicted Age: %2.1fnGender: ' % (test_Y_months[idx], pred_Y[idx])
if test_X[1][idx] == 0:
title += "Femalen"
else:
title += "Malen"
c_ax.set_title(title)
c_ax.axis('off')
plt.show()
model.save('model_xception.h5') # HDF5文件
以下用Inception_v3网络
import numpy as np
import pandas as pd
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "0" # which gpu to use
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
from sklearn.model_selection import train_test_split
from tensorflow.keras.preprocessing.image import ImageDataGenerator, array_to_img, img_to_array, load_img
from tensorflow.keras.applications.inception_v3 import InceptionV3, preprocess_input
from tensorflow.keras.layers import Input, GlobalAveragePooling2D, Dense, Dropout, Flatten, Concatenate
from tensorflow.keras.models import Sequential, Model
from tensorflow.keras.metrics import mean_absolute_error
from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping, ReduceLRonPlateau
from sklearn.metrics import mean_absolute_error as sk_mae
import matplotlib.pyplot as plt
import tensorflow as tf
# from keras.backend.
config = tf.compat.v1.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.compat.v1.Session()
# tensorflow_backend import set_session
# from tensorflow.compat.v1.keras.backend import set_session
# config = tf.ConfigProto()
# config.gpu_options.allow_growth = True # dont allocate entire vram initially
# set_session(tf.Session(config=config))
# Reading data
print("Reading data...")
img_dir = "/home/user/X-ray/archive/boneage-training-dataset/"
csv_path = "/home/user/X-ray/archive/boneage-training-dataset.csv"
age_df = pd.read_csv(csv_path)
age_df['path'] = age_df['id'].map(lambda x: img_dir + "{}.png".format(x))
age_df['exists'] = age_df['path'].map(os.path.exists)
age_df['gender'] = age_df['male'].map(lambda x: "male" if x else "female")
mu = age_df['boneage'].mean()
sigma = age_df['boneage'].std()
age_df['zscore'] = age_df['boneage'].map(lambda x: (x - mu) / sigma)
age_df.dropna(inplace=True)
# # 查看图片大小 1514*2044
# from PIL import Image
# file_path = 'D:/X射线骨龄预测/archive/boneage-training-dataset/1500.png'
# img = Image.open(file_path)
# imgSize = img.size # 大小/尺寸
# w = img.width # 图片的宽
# h = img.height # 图片的高
# f = img.format # 图像格式
# print(imgSize)
# print(w, h, f)
# Examine the distribution of age and gender
print("{} images found out of total {} images".format(age_df['exists'].sum(), age_df.shape[0]))
column_headers = list(age_df.columns.values)
print("column_handers = ", column_headers) # 列标签
print(age_df.sample(5)) # csv中随机抽取5行
# # 导出数据 生成csv
# test = pd.Dataframe(columns=column_headers, data=age_df)
# test.to_csv('D:/test.csv') # 如果生成excel，可以用to_excel
# # # 绘出均衡前后骨龄的数量直方图
# # age_df[['boneage', 'zscore']].hist()
# # plt.xlabel('bone age')
# # plt.ylabel('Number of samples')
# # plt.show()
# # # 男女直方图
# # age_df[['male']]= age_df[['male']].astype(int)
# # age_df[['male']].hist(figsize=(10, 5))
# # plt.show()
# # 获取age_df的某一列
# age_df_col_path_zscore = age_df[['path', 'zscore']]
# age_df_path_zscore = np.array(age_df_col_path_zscore)
# print('age_df_col_path_zscoren ', age_df_col_path_zscore)
# # age_df[['boneage','male','zscore']].hist()
# # plt.show()
print("Reading complete !!!n")
# Split into training testing and validation datasets
print("Preparing training, testing and validation datasets ...")
age_df['boneage_category'] = pd.cut(age_df['boneage'], 10) # 按骨龄的梯度分为10个量级
raw_train_df, test_df = train_test_split(age_df,
test_size=0.2,
random_state=2018,
stratify=age_df['boneage_category'])
raw_train_df, valid_df = train_test_split(raw_train_df,
test_size=0.1,
random_state=2018,
stratify=raw_train_df['boneage_category'])
# # raw_train_df[['boneage']].hist(figsize=(10, 5)) # 绘图（均衡前）
# # plt.xlabel('bone age')
# # plt.ylabel('Number of samples')
# # plt.show()
# # # 男女直方图
# # raw_train_df[['male']] = raw_train_df[['male']].astype(int)
# # raw_train_df[['male']].hist(figsize=(10, 5)) # 绘图（均衡前）
# # plt.show()
# #
# raw_train_df_size = raw_train_df.shape[0]
# valid_size = valid_df.shape[0]
# test_size = test_df.shape[0]
# print("# Training images: {}".format(raw_train_df))
# print("# Validation images: {}".format(valid_size))
# print("# Testing images: {}".format(test_size))
# Training images: 9076 | Validation images: 1009 | Test images: 2523
# Balance the distribution in the training set
# raw_train_df中，boneage_category有10类，male有两类，排列组合共20种 每类重复采样500次，共得到 20 * 500 = 10000 个样本
train_df = raw_train_df.groupby(['boneage_category', 'male']).apply(lambda x: x.sample(500, replace=True)).reset_index(
drop=True)
# print(train_df.sample(5))
# train_df[['boneage']].hist(figsize=(10, 5)) # 绘图 (均衡后)
# plt.title('Equalized image')
# plt.xlabel('bone age')
# plt.ylabel('Number of samples')
# plt.show()
# # 男女直方图
# train_df[['male']] = train_df[['male']].astype(int)
# train_df[['male']].hist(figsize=(10, 5)) # 绘图（均衡后）
# plt.title('Equalized male')
# plt.show()
# # Training images: 10000 | Validation images: 1009 | Test images: 2523
train_size = train_df.shape[0]
valid_size = valid_df.shape[0]
test_size = test_df.shape[0]
print("# Training images: {}".format(train_size))
print("# Validation images: {}".format(valid_size))
print("# Testing images: {}".format(test_size))
# Make training, validation and testing dataset
IMG_SIZE = (299, 299) # default size for inception_v3
img_data_gen = ImageDataGenerator(samplewise_center=False,
samplewise_std_normalization=False,
horizontal_flip=True,
vertical_flip=False,
height_shift_range=0.25,
width_shift_range=0.25,
rotation_range=25,
shear_range=0.2,
fill_mode='reflect',
zoom_range=0.2,
preprocessing_function=preprocess_input)
def gen_2inputs(imgDatGen, df, batch_size, seed, img_size):
gen_img = imgDatGen.flow_from_dataframe(dataframe=df,
x_col='path', y_col='zscore',
batch_size=batch_size, seed=seed, shuffle=True, class_mode='raw',
target_size=img_size, color_mode='rgb')
gen_gender = imgDatGen.flow_from_dataframe(dataframe=df,
x_col='path', y_col='male',
batch_size=batch_size, seed=seed, shuffle=True, class_mode='raw',
target_size=img_size, color_mode='rgb')
while True:
X1i = gen_img.next()
X2i = gen_gender.next()
yield [X1i[0], X2i[1]], X1i[1]
def test_gen_2inputs(imgDatGen, df, batch_size, img_size):
gen_img = imgDatGen.flow_from_dataframe(dataframe=df,
x_col='path', y_col='zscore',
batch_size=batch_size, shuffle=False, class_mode='raw',
target_size=img_size, color_mode='rgb')
gen_gender = imgDatGen.flow_from_dataframe(dataframe=df,
x_col='path', y_col='male',
batch_size=batch_size, shuffle=False, class_mode='raw',
target_size=img_size, color_mode='rgb')
while True:
X1i = gen_img.next()
X2i = gen_gender.next()
yield [X1i[0], X2i[1]], X1i[1]
BATCH_SIZE_TRAIN = 16
SEED = 8309
BATCH_SIZE_VAL = 16
train_flow = gen_2inputs(img_data_gen, train_df, BATCH_SIZE_TRAIN, SEED, IMG_SIZE)
valid_flow = gen_2inputs(img_data_gen, valid_df, BATCH_SIZE_VAL, SEED, IMG_SIZE)
test_flow = test_gen_2inputs(img_data_gen, test_df, test_size, IMG_SIZE)
# Model definition
print("Compiling deep model ...")
IMG_SHAPE = (299, 299, 3) # 224
# 1、两个输入分别是原始图像和性别输入
img = Input(shape=IMG_SHAPE)
gender = Input(shape=(1,))
# 2、预训练主干网络（图像）
cnn_vec = InceptionV3(input_shape=IMG_SHAPE, include_top=False, weights='imagenet')(img)
# 3、主干网络输出
cnn_vec = GlobalAveragePooling2D()(cnn_vec)
cnn_vec = Dropout(0.2)(cnn_vec)
# 4.性别输入网络
gender_vec = Dense(32, activation='relu')(gender)
# 5.两个网络拼接
features = Concatenate(axis=-1)([cnn_vec, gender_vec])
dense_layer = Dense(512, activation='relu')(features)
dense_layer = Dropout(0.2)(dense_layer)
dense_layer = Dense(512, activation='relu')(dense_layer)
dense_layer = Dropout(0.2)(dense_layer)
output_layer = Dense(1, activation='linear')(dense_layer) # linear is what 16bit did
bone_age_model = Model(inputs=[img, gender], outputs=output_layer)
# # VGG Model definition
# print("Compiling deep model ...")
# img = Input(shape=IMG_SHAPE)
# gender = Input(shape=(1,))
# cnn_vec = VGG16(input_shape=IMG_SHAPE, include_top=False, weights='imagenet')(img)
# cnn_vec = GlobalAveragePooling2D()(cnn_vec)
# cnn_vec = Dropout(0.2)(cnn_vec)
# gender_vec = Dense(32, activation='relu')(gender)
# features = Concatenate(axis=-1)([cnn_vec, gender_vec])
# dense_layer = Dense(1024, activation='relu')(features)
# dense_layer = Dropout(0.2)(dense_layer)
# dense_layer = Dense(1024, activation='relu')(dense_layer)
# dense_layer = Dropout(0.2)(dense_layer)
# output_layer = Dense(1, activation='linear')(dense_layer) # linear is what 16bit did
# bone_age_model = Model(inputs=[img, gender], outputs=output_layer)
# Compile model编译模型
def mae_months(in_gt, in_pred):
return mean_absolute_error(mu + sigma * in_gt, mu + sigma * in_pred)
bone_age_model.compile(optimizer='adam', loss='mse', metrics=[mae_months])
bone_age_model.summary()
print("Model compiled !!!n")
# Training deep model
print("Training deep model ...")
# 测试步长 学习率： 步长更大= 学习率更高
EPOCHS = 30
BATCH_SIZE_TEST = len(test_df) // 3
STEP_SIZE_TEST = 3
STEP_SIZE_TRAIN = len(train_df) // BATCH_SIZE_TRAIN
STEP_SIZE_VALID = len(valid_df) // BATCH_SIZE_VAL
from tensorflow.keras import optimizers
# Model Callbacks
weight_path = "{}_weights.best.hdf5".format('bone_age')
checkpoint = ModelCheckpoint(weight_path, monitor='val_loss', verbose=1, save_best_only=True, mode='min',
save_weights_only=True)
optim = optimizers.Nadam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-08, schedule_decay=0.0003)
reduceLRonPlat = ReduceLRonPlateau(monitor='val_loss', factor=0.8, patience=3, verbose=1, mode='auto', epsilon=0.0001,
cooldown=5, min_lr=0.0006)
early = EarlyStopping(monitor="val_loss", mode="min",
patience=10) # probably needs to be more patient, but kaggle time is limited
callbacks_list = [checkpoint, early, reduceLROnPlat]
bone_age_model_history = bone_age_model.fit_generator(generator=train_flow,
steps_per_epoch=STEP_SIZE_TRAIN,
validation_data=valid_flow,
validation_steps=STEP_SIZE_VALID,
epochs=EPOCHS,
callbacks=callbacks_list)
# 保存训练的loss
loss_history = bone_age_model_history.history['loss']
history_df = pd.Dataframe.from_dict(bone_age_model_history.history)
history_df.to_csv('/home/user/X-ray/incption_v3_loss_history1r.csv')
bone_age_model.load_weights("bone_age_weights.best.hdf5")
print("Training complete !!!n")
# evaluate model on test dataset
print("evaluating model on test data ...n")
print("Preparing testing dataset...")
test_X, test_Y = next(test_flow) # one big batch
print("Data prepared !!!")
pred_Y = mu + sigma * bone_age_model.predict(x=test_X, batch_size=16, verbose=1) # 25
test_Y_months = mu + sigma * test_Y
print("Mean absolute error on test data: " + str(sk_mae(test_Y_months, pred_Y)))
fig, ax1 = plt.subplots(1, 1, figsize=(6, 6))
ax1.plot(test_Y_months, pred_Y, 'r.', label='predictions')
ax1.plot(test_Y_months, test_Y_months, 'b-', label='actual')
ax1.legend()
ax1.set_xlabel('Actual Age (Months)')
ax1.set_ylabel('Predicted Age (Months)')
ord_idx = np.argsort(test_Y)
ord_idx = ord_idx[np.linspace(0, len(ord_idx) - 1, num=8).astype(int)] # take 8 evenly spaced ones
fig, m_axs = plt.subplots(2, 4, figsize=(16, 32))
for (idx, c_ax) in zip(ord_idx, m_axs.flatten()):
c_ax.imshow(test_X[0][idx, :, :, 0], cmap='bone')
title = 'Age: %2.1fnPredicted Age: %2.1fnGender: ' % (test_Y_months[idx], pred_Y[idx])
if test_X[1][idx] == 0:
title += "Femalen"
else:
title += "Malen"
c_ax.set_title(title)
c_ax.axis('off')
plt.show()
bone_age_model.save('bone_age_model_inception.h5') # HDF5文件
欢迎分享，转载请注明来源：内存溢出
微信扫一扫
支付宝扫一扫
评论列表（0条）