import pandas as pd
import numpy as np
import warnings
warnings.filterwarnings('ignore')
image_train=np.load('train.npy')
image_test=np.load('test.npy')
key_point=pd.read_csv('train.csv')
import matplotlib.pyplot as plt
XY=key_point.iloc[0,:].values.reshape(-1,2)
plt.scatter(XY[:,0],XY[:,1])
plt.imshow(image_train[:,:,0],cmap='gray')
fig,axes = plt.subplots(4,2,figsize=(8,14))
for i,ax in enumerate(axes.flatten()):
sns.boxplot(key_point.iloc[:,i],ax=ax)
fig,axes = plt.subplots(4,2,figsize=(8,14))
for i,ax in enumerate(axes.flatten()):
sns.boxplot(key_point.iloc[:,i],ax=ax)
#其实是可以删除的
key_point.isna().sum()
#众数填充
key_point=key_point.fillna(key_point.mode().iloc[0,:],axis=0)
from sklearn.multioutput import MultiOutputRegressor
from sklearn.neural_network import MLPRegressor
from sklearn.model_selection import train_test_split
def split_data(X,Y,random_state=2021):
y_test=Y.sample(frac=0.2,random_state=random_state)
test_index=y_test.index.tolist()
train_index=list(set(Y.index)-set(test_index))
y_train=Y.iloc[train_index,:]
x_test=X[:,:,test_index]
x_train=X[:,:,train_index]
return x_train,y_train,x_test,y_test
x_train,y_train,x_test,y_test=split_data(image_train,key_point)
x_train_=np.zeros((4000,96*96))
x_test_=np.zeros((1000,96*96))
for i in range(4000):
x_train_[i,:]=x_train[:,:,i].flatten()
for i in range(1000):
x_test_[i,:]=x_test[:,:,i].flatten()
#多输出回归
mor=MultiOutputRegressor(MLPRegressor(random_state=2021)).fit(x_train_,y_train)
y_pred=mor.predict(x_test_)
image_test_=np.zeros((2049,96*96))
for i in range(2049):
image_test_[i,:]=image_test[:,:,i].flatten()
y_test_pred_sk_mlp=mor.predict(image_test_)
from keras import models,Input,layers
from keras import backend as K
K.clear_session()
input_x = Input(shape=(96, 96,1))
x = layers.Flatten()(input_x)
for _ in range(3):
x = layers.Dense(128, activation='relu')(x)
x = layers.Dropout(0.2)(x)
out_put_left_eye_center_x = layers.Dense(1)(x)
out_put_left_eye_center_y = layers.Dense(1)(x)
out_put_right_eye_center_x = layers.Dense(1)(x)
out_put_right_eye_center_y = layers.Dense(1)(x)
out_put_nose_tip_x = layers.Dense(1)(x)
out_put_nose_tip_y = layers.Dense(1)(x)
out_put_mouth_center_bottom_lip_x = layers.Dense(1)(x)
out_put_mouth_center_bottom_lip_y = layers.Dense(1)(x)
model = models.Model(inputs=[input_x],
outputs=[
out_put_left_eye_center_x, out_put_left_eye_center_y,
out_put_right_eye_center_x,
out_put_right_eye_center_y, out_put_nose_tip_x,
out_put_nose_tip_y, out_put_mouth_center_bottom_lip_x,
out_put_mouth_center_bottom_lip_y
])
model.compile(optimizer='rmsprop', loss='mae', metrics=['mse'])
from tensorflow.keras.utils import plot_model
plot_model(model)
train_img = np.zeros((5000, 96, 96))
for i in range(5000):
train_img[i, :, :] = image_train[:, :, i]
train_img = train_img.reshape(5000, 96, 96, 1) / 255
test_img = np.zeros((2049, 96, 96))
for i in range(2049):
test_img[i, :, :] = image_test[:, :, i]
test_img = test_img.reshape(2049, 96, 96, 1) / 255
model.fit(train_img, [
key_point[['left_eye_center_x']], key_point[['left_eye_center_y']],
key_point[['right_eye_center_x']], key_point[['right_eye_center_y']],
key_point[['nose_tip_x']], key_point[['nose_tip_y']],
key_point[['mouth_center_bottom_lip_x']], key_point[['mouth_center_bottom_lip_y']]
],
batch_size=64,
epochs=10)
K.clear_session()
t=[64,128,128]
input_x = Input(shape=(96, 96,1))
x = layers.Conv2D(32,(3,3),activation='relu')(input_x)
x = layers.MaxPooling2D((2,2))(x)
for i in range(3):
x = layers.Conv2D(t[i],(3,3),activation='relu')(x)
x = layers.MaxPooling2D((2,2))(x)
x=layers.Flatten()(x)
x = layers.Dropout(0.2)(x)
out_put_left_eye_center_x = layers.Dense(1)(x)
out_put_left_eye_center_y = layers.Dense(1)(x)
out_put_right_eye_center_x = layers.Dense(1)(x)
out_put_right_eye_center_y = layers.Dense(1)(x)
out_put_nose_tip_x = layers.Dense(1)(x)
out_put_nose_tip_y = layers.Dense(1)(x)
out_put_mouth_center_bottom_lip_x = layers.Dense(1)(x)
out_put_mouth_center_bottom_lip_y = layers.Dense(1)(x)
model = models.Model(inputs=[input_x],
outputs=[
out_put_left_eye_center_x, out_put_left_eye_center_y,
out_put_right_eye_center_x,
out_put_right_eye_center_y, out_put_nose_tip_x,
out_put_nose_tip_y, out_put_mouth_center_bottom_lip_x,
out_put_mouth_center_bottom_lip_y
])
model.compile(optimizer='rmsprop', loss='mae', metrics=['mse'])
model.fit(train_img, [
key_point[['left_eye_center_x']], key_point[['left_eye_center_y']],
key_point[['right_eye_center_x']], key_point[['right_eye_center_y']],
key_point[['nose_tip_x']], key_point[['nose_tip_y']],
key_point[['mouth_center_bottom_lip_x']], key_point[['mouth_center_bottom_lip_y']]
],
batch_size=64,
epochs=10)
from tensorflow.keras.applications import ResNet50
K.clear_session()
ResNet50_base = ResNet50(include_top=False,input_shape=(96,96,1),weights=None)
input_x = Input(shape=(96, 96, 1))
x = ResNet50_base(input_x)
x = layers.Flatten()(x)
x = layers.Dropout(0.2)(x)
out_put_left_eye_center_x = layers.Dense(1)(x)
out_put_left_eye_center_y = layers.Dense(1)(x)
out_put_right_eye_center_x = layers.Dense(1)(x)
out_put_right_eye_center_y = layers.Dense(1)(x)
out_put_nose_tip_x = layers.Dense(1)(x)
out_put_nose_tip_y = layers.Dense(1)(x)
out_put_mouth_center_bottom_lip_x = layers.Dense(1)(x)
out_put_mouth_center_bottom_lip_y = layers.Dense(1)(x)
model = models.Model(inputs=[input_x],
outputs=[
out_put_left_eye_center_x, out_put_left_eye_center_y,
out_put_right_eye_center_x,
out_put_right_eye_center_y, out_put_nose_tip_x,
out_put_nose_tip_y, out_put_mouth_center_bottom_lip_x,
out_put_mouth_center_bottom_lip_y
])
model.compile(optimizer='rmsprop', loss='mae', metrics=['mse'])
model.fit(train_img, [
key_point[['left_eye_center_x']], key_point[['left_eye_center_y']],
key_point[['right_eye_center_x']], key_point[['right_eye_center_y']],
key_point[['nose_tip_x']], key_point[['nose_tip_y']],
key_point[['mouth_center_bottom_lip_x']], key_point[['mouth_center_bottom_lip_y']]
],
batch_size=64,
epochs=10)
