I'm trying to implement an app that can predict bmi, age, gender using facial image. To be able to do this i'm trying to use the resnet-50 cnn model knowledge and change the input and output of the network.
For Training/Testing the new model i'm using FACE2BMI Dataset contain almost 60k images with bmi, age, gender data.
face2bmi csv file
I've used this code to create the model
IMG_SHAPE = (256,256, 3)
base_model = tf.keras.applications.EfficientNetB6(weights='imagenet', include_top=False, input_shape=IMG_SHAPE)
base_model.trainable = False
optimizer = tf.keras.optimizers.Adam()
model_inputs = tf.keras.Input(shape=(256, 256, 3))
x = base_model(model_inputs, training=False)
x = tf.keras.layers.MaxPooling2D()(x)
x = tf.keras.layers.Flatten()(x)
#let's add a fully-connected layer
'''x = tf.keras.layers.Dense(64,activation='relu')(x)
x = tf.keras.layers.Dropout(0.2)(x)
x = tf.keras.layers.Dense(64,activation='relu')(x)
x = tf.keras.layers.Dropout(0.2)(x)
'''
#start passing that fully connected block output to all the different model heads
y1 = tf.keras.layers.Dense(32,activation='relu')(x)
y1 = tf.keras.layers.Dropout(0.2)(y1)
y1 = tf.keras.layers.Dense(16,activation='relu')(y1)
y1 = tf.keras.layers.Dropout(0.2)(y1)
y2 = tf.keras.layers.Dense(32,activation='relu')(x)
y2 = tf.keras.layers.Dropout(0.2)(y2)
y2 = tf.keras.layers.Dense(16,activation='relu')(y2)
y2 = tf.keras.layers.Dropout(0.2)(y2)
y3 = tf.keras.layers.Dense(32,activation='sigmoid')(x)
y3 = tf.keras.layers.Dropout(0.2)(y3)
y3 = tf.keras.layers.Dense(16,activation='sigmoid')(y3)
y3 = tf.keras.layers.Dropout(0.2)(y3)
# Predictions for each task
y1 = tf.keras.layers.Dense(units=3,activation="linear",name='bmi')(y1)
y2 = tf.keras.layers.Dense(units=3,activation="linear",name='age')(y2)
y3 = tf.keras.layers.Dense(units=3,activation="sigmoid",name='sex')(y3)
custom_model = tf.keras.Model(inputs=model_inputs,outputs=[y1,y2,y3])
custom_model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=1e-5),
loss={'bmi':'mean_squared_error','age':'mean_squared_error','sex':'hinge'},metrics=['accuracy'])
The summary of the model is:
Model: "model_1"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_4 (InputLayer) [(None, 256, 256, 3 0 []
)]
efficientnetb6 (Functional) (None, 8, 8, 2304) 40960143 ['input_4[0][0]']
max_pooling2d_1 (MaxPooling2D) (None, 4, 4, 2304) 0 ['efficientnetb6[0][0]']
flatten_1 (Flatten) (None, 36864) 0 ['max_pooling2d_1[0][0]']
dense_6 (Dense) (None, 32) 1179680 ['flatten_1[0][0]']
dense_8 (Dense) (None, 32) 1179680 ['flatten_1[0][0]']
dense_10 (Dense) (None, 32) 1179680 ['flatten_1[0][0]']
dropout_6 (Dropout) (None, 32) 0 ['dense_6[0][0]']
dropout_8 (Dropout) (None, 32) 0 ['dense_8[0][0]']
dropout_10 (Dropout) (None, 32) 0 ['dense_10[0][0]']
dense_7 (Dense) (None, 16) 528 ['dropout_6[0][0]']
dense_9 (Dense) (None, 16) 528 ['dropout_8[0][0]']
dense_11 (Dense) (None, 16) 528 ['dropout_10[0][0]']
dropout_7 (Dropout) (None, 16) 0 ['dense_7[0][0]']
dropout_9 (Dropout) (None, 16) 0 ['dense_9[0][0]']
dropout_11 (Dropout) (None, 16) 0 ['dense_11[0][0]']
bmi (Dense) (None, 3) 51 ['dropout_7[0][0]']
age (Dense) (None, 3) 51 ['dropout_9[0][0]']
sex (Dense) (None, 3) 51 ['dropout_11[0][0]']
==================================================================================================
Total params: 44,500,920
Trainable params: 3,540,777
Non-trainable params: 40,960,143
_____________________________________
my question is how can i train and test this new custom model with the FACE2BMI Dataset ?