def train():
net = model.model(128).model()
net.compile(optimizer=SGD(lr=0.001),
loss='categorical_crossentropy',
metrics=['acc'])
net.summary()
x_train, y_train = loader.loader("./data/FF-PP/train")
net.fit(x_train, y_train, epochs=100, batch_size=50, validation_split=0.2)
x_test, y_test = loader.loader("./data/FF-PP/test")
score = net.evaluate(x_test, y_test)
print(score)
## Save the Weights
json_string = net.to_json()
open('./models/FF-PP/AlexNet.json', 'w').write(json_string)
net.save_weights("./models/FF-PP/weights.h5")
from keras import backend as K
from keras.models import load_model, model_from_json
import numpy as np
from md import loader
import cv2
from tqdm import tqdm
import copy
import os
# Read model
json_string = open('./models/FF-FM/AlexNet.json', 'r').read()
model = model_from_json(json_string)
model.load_weights('./models/FF-FM/weights.h5')
model.summary()
# Read images and names
ld = loader.loader("./data/FF-FM/train", option=True)
x = ld[0][3]
cv2.imwrite("./sample.png", x * 255)
x = np.array([x])
for x, path in zip(ld[0], ld[2]):
x = np.array([x])
name = os.path.basename(path)
# define functions
func_Yc = model.output[:, 0]
func_Aij6 = model.get_layer("conv2d_6").output
func_Aij7 = model.get_layer("conv2d_7").output
func_Aij8 = model.get_layer("conv2d_8").output
func_Aij9 = model.get_layer("conv2d_9").output
func_Aij10 = model.get_layer("conv2d_10").output
func_grads6 = K.gradients(func_Yc, func_Aij6)[0]
