Related
import numpy as np
import keras
from keras.layers import Conv2D, MaxPooling2D, Flatten, Dense, Dropout, Input, Add, BatchNormalization, Lambda, Concatenate, Reshape
from keras.models import Model
import tensorflow as tf
import os
import pandas as pd
import pydicom as dicom
import matplotlib.pyplot as plt
import math
import pydicom
import cv2
import os
def resize_dcm(dcmdosyası, boyut):
dcmgörsel = pydicom.read_file(dcmdosyası)
img = dcmgörsel.pixel_array
img = cv2.resize(img, boyut, interpolation = cv2.INTER_AREA)
dcmgörsel.PixelData = img.tobytes()
dcmgörsel.Rows, dcmgörsel.Columns = img.shape
return dcmgörsel
folder = 'C:/Users/USER/Desktop/data/'
ana1 = [f.path for f in os.scandir(folder) if f.is_dir()]
for ana2 in ana1:
görseller = [f.path for f in os.scandir(ana2) if f.is_file()]
for görsellercücük in görseller:
resize2 = resize_dcm(görsellercücük, (800, 800))
yeni = os.path.join(ana2, os.path.basename(görsellercücük))
pydicom.write_file(yeni, resize2)
inputs = Input(shape=(800,800,1))
input2 = Input(shape=(1))
input3 = Input(shape=(1))
x = Conv2D(64, (3, 3), activation='relu', padding='same')(inputs)
x = BatchNormalization()(x)
x = Conv2D(64, (3, 3), activation='relu', padding='same')(x)
x = BatchNormalization()(x)
x = Conv2D(64, (3, 3), activation='relu', padding='same')(x)
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = Dropout(0.5)(x)
x = Conv2D(128, (3, 3), activation='relu', strides=2, padding='same')(x)
x = BatchNormalization()(x)
x = Conv2D(128, (3, 3), activation='relu', strides=2, padding='same')(x)
x = BatchNormalization()(x)
res1 = Conv2D(128, (3, 3), activation='relu', padding='same')(x)
res2 = Conv2D(128, (3, 3), activation='relu', padding='same')(res1)
x = Add()([x, res2])
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = Dropout(0.5)(x)
x = Conv2D(256, (3, 3), activation='relu', strides=2, padding='same')(x)
x = BatchNormalization()(x)
x = Conv2D(256, (3, 3), activation='relu', strides=2, padding='same')(x)
x = BatchNormalization()(x)
res1 = Conv2D(256, (3, 3), activation='relu', padding='same')(x)
res2 = Conv2D(256, (3, 3), activation='relu', padding='same')(res1)
x = Add()([x, res2])
res1 = Conv2D(256, (3, 3), activation='relu', padding='same')(x)
res2 = Conv2D(256, (3, 3), activation='relu', padding='same')(res1)
x = Add()([x, res2])
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = Conv2D(512, (3, 3), activation='relu', strides=2, padding='same')(x)
x = BatchNormalization()(x)
x = Conv2D(512, (3, 3), activation='relu', strides=2, padding='same')(x)
x = BatchNormalization()(x)
res1 = Conv2D(512, (3, 3), activation='relu', padding='same')(x)
res2 = Conv2D(512, (3, 3), activation='relu', padding='same')(res1)
x = Add()([x, res2])
res1 = Conv2D(512, (3, 3), activation='relu', padding='same')(x)
res2 = Conv2D(512, (3, 3), activation='relu', padding='same')(res1)
x = Add()([x, res2])
x = Flatten()(x)
x = Concatenate()([x,input2,input3])
x = Dense(4096, activation='relu')(x)
x = Dense(4096, activation='relu')(x)
output = Dense(4, activation='softmax')(x) #birads score
output2 = Dense(4, activation='softmax')(x) #kompozisyon
output3 = Dense(2, activation='softmax')(x) #bolge 1
output4 = Dense(2, activation='softmax')(x) #bolge 2
output5 = Dense(2, activation='softmax')(x) #bolge 3
model = Model([inputs, input2,input3], [output,output2, output3])
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
model.summary()
data = pd.read_excel('C:/Users/USER/Desktop/asdsadasdas.xlsx')
data.rename(columns={'HASTANO': 'ID', 'BIRADS KATEGORİSİ':'birads', 'MEME KOMPOZİSYONU': 'comp', 'KADRAN BİLGİSİ (SAĞ)': 'areaR', 'KADRAN BİLGİSİ (SOL)': 'areaL'}, inplace=True)
data.replace(['BI-RADS0', 'BI-RADS1-2', 'BI-RADS4-5'], [0,1,2], inplace=True)
data.drop(data.columns[5], inplace=True, axis=1)
data['comp'].value_counts()
data.fillna(0, inplace=True)
def transform_to_hu(medical_image, image):
intercept = medical_image.RescaleIntercept
slope = medical_image.RescaleSlope
hu_image = image * slope + intercept
return hu_image
def train():
mainDir = 'C:/Users/USER/Desktop/data/'
for d in data.iterrows():
patientDir = mainDir + str(d[1][0])
for f in os.listdir(patientDir):
yon = 1 if f[0] == 'R' else 0
view = 1 if f[1] == 'M' else 0
dcm = dicom.dcmread(patientDir + '/' + f)
image = transform_to_hu(dcm, dcm.pixel_array)
o1 = 0
o2 = 0
o3 = 0
areaStr = d[1]['area' + ('R' if yon == 1 else 'L')]
if areaStr!= 0:
areaStr = areaStr.replace('[', '').replace(']', '')
lst = areaStr.split(',')
newLst = list(map(lambda x : x.replace('"', ''), lst))
print(newLst)
for a in newLst:
if(a == 'MERKEZ'):
o2 = 1
continue
t = a.split(' ')
if(view == 1): #mlo
if t[0] == 'ÜST':
o1 = 1
else:
o3 = 1
else:
if t[1] == 'DIŞ':
o1 = 1
else:
o3 = 1
print(o1, o2, o3, sep=', ')
model.fit([image, yon, view], [d[1]['birads'], d[1]['comp'], o1, o2, o3])
break
train()
Error:
Traceback (most recent call last):
File "C:\Users\USER\Desktop\aaac.py", line 157, in <module>
train()
File "C:\Users\USER\Desktop\aaac.py", line 155, in train
model.fit([image, yon, view], [d[1]['birads'], d[1]['comp'], o1, o2, o3])
File "C:\Users\USER\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\keras\utils\traceback_utils.py", line 70, in error_handler
raise e.with_traceback(filtered_tb) from None
File "C:\Users\USER\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\keras\engine\data_adapter.py", line 1081, in select_data_adapter
raise ValueError(
ValueError: Failed to find data adapter that can handle input: (<class 'list'> containing values of types {"<class 'numpy.ndarray'>", "<class 'int'>"}), (<class 'list'> containing values of types {"<class 'str'>", "<class 'int'>"})
Process finished with exit code 1
hello i wrote a deep learning code. At first, I shaped the code in accordance with the resnet architecture, and then I tried to obtain an output by combining the data on excel and visual. I have encountered such an error, I do not know why.
I am launching the following neural network, and I suspect that there is a mistake with some variables (especially in the MeanShift function) with the initialization of weights and biases.
################
##### RCAN #####
################
def RCAN(n_RCAB=3, n_RG=5, rgb_mean = [0.4488], rgb_std = 1, **kwargs):
def _func(input_shape):
import sys
sys.setrecursionlimit(25000)
# Start of the Model
inputs = Input(input_shape)
scaled_inputs = tf.keras.layers.experimental.preprocessing.Rescaling(255)(inputs)
scaled_inputs = MeanShift(rgb_range=255, rgb_mean=rgb_mean, rgb_std=rgb_std)(inputs)
# Head module
x = Conv2D(64, 3, padding='same')(scaled_inputs)
_inp = x
# Body module
for i in range(n_RG):
x = ResidualGroup(x, n_RCAB)
x = Conv2D(64, 3, padding='same')(x)
add_global = tf.keras.layers.Add()([x, _inp])
# Tail module
x = UpSampling(add_global)
x = Conv2D(1, 3, padding='same')(x)
outputs = MeanShift(rgb_range=255, rgb_mean=rgb_mean, rgb_std=rgb_std, sign=1)(x)
outputs = tf.keras.layers.experimental.preprocessing.Rescaling(1.0 / 255)(outputs)
# Defining the model
model = Model(inputs=inputs, outputs=outputs)
return model
return _func
######################
##### ABOUT RCAN #####
######################
def RCAB(x):
_res = x
x = Conv2D(64, 3, padding='same')(x)
x = tf.keras.layers.LeakyReLU()(x)
x = Conv2D(64, 3, padding='same')(x)
x = ChannelAttention(x)
x = tf.keras.layers.Add()([x, _res])
return x
def ChannelAttention(x):
_res = x
avg_pool = tf.math.reduce_mean(x, axis=[1, 2, 3], keepdims=True)
feat_mix = Conv2D(4, 1, padding='same', activation='relu')(avg_pool)
feat_mix = Conv2D(64, 1, padding='same', activation='sigmoid')(feat_mix)
multi = tf.keras.layers.Multiply()([feat_mix, _res])
return multi
def ResidualGroup(x, n_RCAB):
skip_connection = x
for i in range(n_RCAB):
x = RCAB(x)
x = Conv2D(64, 3, padding='same')(x)
x = tf.keras.layers.Add()([x, skip_connection])
return x
def MeanShift(rgb_range, rgb_mean, rgb_std, sign=-1):
def _func(x):
# Initilize weights
weight_initer = tf.constant(value=1., dtype=tf.float32, shape=[1])
W = tf.Variable(weight_initer, name="Weight", dtype=tf.float32)
# Initilize bias
bias_initer = tf.constant(value=[m * rgb_range * sign / rgb_std for m in rgb_mean], dtype=tf.float32)
b = tf.Variable(bias_initer, name="Bias", dtype=tf.float32)
x = Conv2D(1, 1, padding="same", kernel_initializer=keras.initializers.Constant(W), bias_initializer=keras.initializers.Constant(b), activation="relu")(x)
return x
return _func
def UpSampling(x, act=False):
features = Conv2D(256, 3, padding='same')(x)
high_res = tf.nn.depth_to_space(features, 2)
if act:
high_res = tf.keras.layers.ReLU(max_value=1)(high_res)
return high_res
The error is when saving the weights (with checkpoints every 5 epochs), but the update of any version (2.4 or 2.9) of tensorflow isn't working.
ValueError: Unable to serialize VariableSpec(shape=(1,),
dtype=tf.float32, trainable=True, alias_id=None) to JSON, because the
TypeSpec class <class
'tensorflow.python.ops.resource_variable_ops.VariableSpec'> has not
been registered
Moreover, my inputs and target are percentile normalized. And I would like to know what layers or operations to add so that predictions matches best the targets (on the variables scaled_inputs and outputs of RCAN function for example).
I would appreciate any help. Thank you in advance.
EDIT:
Ok, found the problem: i think that kernel and bias were not trainable due to keras.initializers.Constant(W) (and same for b) in the Conv2D layer, or due to the tensorflow initializer (I was a bit lost because multiple ones exist). So I replaced it by: and it works
def MeanShift(rgb_range, rgb_mean, rgb_std, sign=-1):
def _func(x):
# Initilize weights and bias
W = tf.keras.initializers.Constant(1.)
b = tf.keras.initializers.Constant(rgb_mean[0] * rgb_range * sign / rgb_std)
x = Conv2D(1, 1, padding="same", kernel_initializer=W, bias_initializer=b)(x)
return x
return _func
I am dealing with audio data (.wav) files, I have created two CNN network to pass the data. so the two models at the end will use the function of Concatenate the output of these two models and give out the classification. but I am having problem for training it, it says the model expects to see 2 arrays but it got one array. any one who can help me with this problem I will appreciate, I have been struggling for weeks.
this is my config file
class Config:
def __init__(self, mode='conv', nfilt=26, nfeat=40, nfft=4096, rate=16000):
self.mode = mode
self.nfilt = nfilt
self.nfeat = nfeat
self.nfft = nfft
self.rate = rate
self.step = int(rate / 10)
self.model_path = os.path.join('models', mode + '.model')
self.p_path = os.path.join('pickles', mode + '.p')
I build one function for my mfcc
build a function for build
def build_rand_feat():
X = []
y = []
_min, _max = float('inf'), -float('inf')
for _ in tqdm(range(n_sample)):
rand_class = np.random.choice(class_dist.index, p=prob_dist)
file = np.random.choice(df[df.label == rand_class].index)
data, rate = sf.read('audios/' + file)
label = df.at[file, 'label']
rand_index = np.random.randint(0, data.shape[0] - config.step)
sample = data[rand_index:rand_index + config.step]
X_sample = mfcc(sample, rate, winlen=0.05, winstep=0.02, numcep=config.nfeat, nfilt=config.nfilt, nfft=config.nfft)
_min = min(np.amin(X_sample), _min)
_max = max(np.amax(X_sample), _max)
X.append(X_sample)
y.append(classes.index(label))
config.min = _min
X = np.array(X)
y = np.array(y)
X = (X - _min) / (_max - _min)
if config.mode == 'conv':
X = X.reshape(X.shape[0], X.shape[1], X.shape[2], 1)
print(X.shape)
elif config.mode == 'time':
X = X.reshape(X.shape[0], X.shape[1], X.shape[2])
y = to_categorical(y, num_classes=10)
config.data = (X, y)
return X, y
and this is my multi scale network
def get_conv_model():
main_model = Sequential()
main_model.add(Conv2D(128, (3, 3), activation='relu', strides=(1, 1), padding='same', input_shape=input_shape))
main_model.add(MaxPool2D((2, 2)))
main_model.add(Conv2D(128, (3, 3), activation='relu', strides=(1, 1), padding='same'))
main_model.add(MaxPool2D((2, 2), padding='same'))
main_model.add(Flatten())
second_model = Sequential()
second_model.add(Conv2D(256, (3, 3), activation='relu', strides=(1, 1), padding='same', input_shape=input_shape))
second_model.add(MaxPool2D((2, 2), padding='same'))
second_model.add(Conv2D(256, (3, 3), activation='relu', strides=(1, 1), padding='same'))
second_model.add(MaxPool2D((2, 2)))
second_model.add(Flatten())
# first model upper
main_model = Sequential()
main_model.add(Conv2D(64, (3, 3), activation='relu', strides=(1, 1), padding='same', input_shape=input_shape))
main_model.add(BatchNormalization())
main_model.add(Dropout(0.3))
main_model.add(Conv2D(128, (3, 3), activation='relu', strides=(1, 1), padding='same'))
main_model.add(BatchNormalization())
main_model.add(Dropout(0.3))
main_model.add(Conv2D(256, (3, 3), activation='relu', strides=(1, 1), padding='same'))
main_model.add(BatchNormalization())
main_model.add(Dropout(0.3))
main_model.add(Flatten())
# second model lower
lower_model1 = Sequential()
lower_model1.add(MaxPool2D(strides=(1, 1), padding='same', input_shape=input_shape))
lower_model1.add(Conv2D(128, (3, 3), activation='relu', strides=(1, 1), padding='same', input_shape=input_shape))
lower_model1.add(BatchNormalization())
lower_model1.add(Dropout(0.3))
lower_model1.add(Conv2D(256, (3, 3), activation='relu', strides=(1, 1), padding='same'))
lower_model1.add(BatchNormalization())
lower_model1.add(Dropout(0.3))
lower_model1.add(Conv2D(512, (3, 3), activation='relu', strides=(1, 1), padding='same'))
lower_model1.add(Flatten())
# merged models
merged_model = Concatenate()([main_model.output, lower_model1.output])
x = Dense(256, activation='relu')(merged_model)
x = Dropout(0.3)(x)
x = Dense(512, activation='relu')(x)
x = Dropout(0.3)(x)
x = Dense(128, activation='relu')(x)
x = Dropout(0.3)(x)
output = Dense(10, activation='softmax')(x)
final_model = Model(inputs=[main_model.input, lower_model1.input], outputs=[output])
final_model.summary()
final_model.compile(loss="categorical_crossentropy", optimizer=Adam(0.001), metrics=['acc'])
print(K.eval(final_model.optimizer.lr))
#class_weight = compute_class_weight('balanced', np.unique(y_flat), y_flat)
final_model.fit(X,y, epochs=10, batch_size=64, shuffle=True, validation_split=0.3)
return main_model, lower_model1, merged_model
I'm trying to do semantic segmentation on skin lesion. I used SegNet as my model. But after running this model, training loss was decreasing but validation loss was not decreasing. i.e. overfitting problem is occured. I tuned learning rate many times and reduced number of number dense layer but no solution came. I also used dropout but still overfitting is happening.
Here is the graph
Here is the code of my model:
def segnet(input_size=(512, 512, 1)):
# Encoding layer
img_input = Input(input_size)
x = Conv2D(64, (3, 3), padding='same', name='conv1',strides= (1,1))(img_input)
x = BatchNormalization(name='bn1')(x)
x = Activation('relu')(x)
x = Conv2D(64, (3, 3), padding='same', name='conv2')(x)
x = BatchNormalization(name='bn2')(x)
x = Activation('relu')(x)
x = MaxPooling2D()(x)
x = Dropout(0.7)(x)
x = Conv2D(128, (3, 3), padding='same', name='conv3')(x)
x = BatchNormalization(name='bn3')(x)
x = Activation('relu')(x)
x = Conv2D(128, (3, 3), padding='same', name='conv4')(x)
x = BatchNormalization(name='bn4')(x)
x = Activation('relu')(x)
x = MaxPooling2D()(x)
x = Dropout(0.7)(x)
x = Conv2D(256, (3, 3), padding='same', name='conv5')(x)
x = BatchNormalization(name='bn5')(x)
x = Activation('relu')(x)
x = Conv2D(256, (3, 3), padding='same', name='conv6')(x)
x = BatchNormalization(name='bn6')(x)
x = Activation('relu')(x)
x = Conv2D(256, (3, 3), padding='same', name='conv7')(x)
x = BatchNormalization(name='bn7')(x)
x = Activation('relu')(x)
x = MaxPooling2D()(x)
x = Dropout(0.7)(x)
x = Conv2D(512, (3, 3), padding='same', name='conv8')(x)
x = BatchNormalization(name='bn8')(x)
x = Activation('relu')(x)
x = Conv2D(512, (3, 3), padding='same', name='conv9')(x)
x = BatchNormalization(name='bn9')(x)
x = Activation('relu')(x)
x = Conv2D(512, (3, 3), padding='same', name='conv10')(x)
x = BatchNormalization(name='bn10')(x)
x = Activation('relu')(x)
x = MaxPooling2D()(x)
x = Dropout(0.7)(x)
x = Conv2D(512, (3, 3), padding='same', name='conv11')(x)
x = BatchNormalization(name='bn11')(x)
x = Activation('relu')(x)
x = Conv2D(512, (3, 3), padding='same', name='conv12')(x)
x = BatchNormalization(name='bn12')(x)
x = Activation('relu')(x)
x = Conv2D(512, (3, 3), padding='same', name='conv13')(x)
x = BatchNormalization(name='bn13')(x)
x = Activation('relu')(x)
x = MaxPooling2D()(x)
x = Dropout(0.7)(x)
x = Dense(256, activation = 'relu', name='fc1')(x)
x = Dense(256, activation = 'relu', name='fc2')(x)
# Decoding Layer
x = UpSampling2D()(x)
x = Conv2DTranspose(512, (3, 3), padding='same', name='deconv1')(x)
x = BatchNormalization(name='bn14')(x)
x = Activation('relu')(x)
x = Conv2DTranspose(512, (3, 3), padding='same', name='deconv2')(x)
x = BatchNormalization(name='bn15')(x)
x = Activation('relu')(x)
x = Conv2DTranspose(512, (3, 3), padding='same', name='deconv3')(x)
x = BatchNormalization(name='bn16')(x)
x = Activation('relu')(x)
x = Dropout(0.7)(x)
x = UpSampling2D()(x)
x = Conv2DTranspose(512, (3, 3), padding='same', name='deconv4')(x)
x = BatchNormalization(name='bn17')(x)
x = Activation('relu')(x)
x = Conv2DTranspose(512, (3, 3), padding='same', name='deconv5')(x)
x = BatchNormalization(name='bn18')(x)
x = Activation('relu')(x)
x = Conv2DTranspose(256, (3, 3), padding='same', name='deconv6')(x)
x = BatchNormalization(name='bn19')(x)
x = Activation('relu')(x)
x = Dropout(0.7)(x)
x = UpSampling2D()(x)
x = Conv2DTranspose(256, (3, 3), padding='same', name='deconv7')(x)
x = BatchNormalization(name='bn20')(x)
x = Activation('relu')(x)
x = Conv2DTranspose(256, (3, 3), padding='same', name='deconv8')(x)
x = BatchNormalization(name='bn21')(x)
x = Activation('relu')(x)
x = Conv2DTranspose(128, (3, 3), padding='same', name='deconv9')(x)
x = BatchNormalization(name='bn22')(x)
x = Activation('relu')(x)
x = Dropout(0.7)(x)
x = UpSampling2D()(x)
x = Conv2DTranspose(128, (3, 3), padding='same', name='deconv10')(x)
x = BatchNormalization(name='bn23')(x)
x = Activation('relu')(x)
x = Conv2DTranspose(64, (3, 3), padding='same', name='deconv11')(x)
x = BatchNormalization(name='bn24')(x)
x = Activation('relu')(x)
x = Dropout(0.7)(x)
x = UpSampling2D()(x)
x = Conv2DTranspose(64, (3, 3), padding='same', name='deconv12')(x)
x = BatchNormalization(name='bn25')(x)
x = Activation('relu')(x)
x = Conv2DTranspose(1, (3, 3), padding='same', name='deconv13')(x)
x = BatchNormalization(name='bn26')(x)
x = Activation('sigmoid')(x)
pred = Reshape((input_size[0], input_size[1]))(x)
return Model(inputs=img_input, outputs=pred)
I have used the same dataset for another modle UNet but there was no overfit for UNet. This is giving overfit only for SegNet model.
For more information :
model = segnet(input_size = (224, 224, INPUT_CHANNELS))
model.compile(optimizer= Adam(lr=1e-5), loss= [dice_coef_loss]
, metrics=[iou, dice_coe, precision, recall, accuracy])
model_checkpoint = ModelCheckpoint(str(j+1) + '_skin_leison.hdf5',
monitor='loss',
verbose=1,
save_best_only=True)
callbacks_list = [model_checkpoint]
history = model.fit(X_train_cv,
y_train_cv,
epochs= 70,
callbacks = callbacks_list,
batch_size= 8,
validation_data=(X_valid_cv, y_valid_cv))
I am currently trying to implement a Bagnet on keras following this article :https://www.lyrn.ai/2019/02/14/bagnet-imagenet-with-a-simple-bof-model/
I am building one model to evaluate a patch and another to evaluate the averaging of all my patchs. And I try to ensemble them with an input: (number of patchs,x_patch,y_patch,channel_patch). But it isnt working well. My main issues come from evaluating all of the patchs and average the result.
Here is my code. If anyone has suggestion on how to deal with that.
def build_patch_model(lr, l2,patch_number):
##############
# BRANCH MODEL
##############
regul = regularizers.l2(l2)
optim = Adam(lr=lr)
kwargs = {'padding':'same', 'kernel_regularizer':regul}
inp = Input(shape=patch_size) # patch qxqx3
#x = MaxPooling2D((2, 2), strides=(2, 2))(inp) # 24x24x128
#x = BatchNormalization()(x)
x = Conv2D(32, (1,1), activation='relu', **kwargs)(inp)
for _ in range(6): x = subblock(x, 32, **kwargs)
x = MaxPooling2D((2, 2), strides=(2, 2))(x) # 24x24x128
x = BatchNormalization()(x)
x = Conv2D(64, (1,1), activation='relu', **kwargs)(x)
for _ in range(6): x = subblock(x, 64, **kwargs)
x = MaxPooling2D((2, 2), strides=(2, 2))(x) # 24x24x128
x = BatchNormalization()(x)
x = Conv2D(128, (1,1), activation='relu', **kwargs)(x)
for _ in range(6): x = subblock(x, 128, **kwargs)
#x = MaxPooling2D((2, 2), strides=(2, 2))(x) # 24x24x128
#x = BatchNormalization()(x)
x = Conv2D(256, (1,1), activation='relu', **kwargs)(x)
for _ in range(6): x = subblock(x, 256, **kwargs)
x = MaxPooling2D((2, 2), strides=(2, 2))(x) # 24x24x128
x = BatchNormalization()(x)
x = Conv2D(512, (1,1), activation='relu', **kwargs)(x)
for _ in range(6): x = subblock(x, 512, **kwargs)
#x = MaxPooling2D((2, 2), strides=(2, 2))(x) # 24x24x128
#x = BatchNormalization()(x)
x = Conv2D(1080, (1,1), activation='relu', **kwargs)(x)
for _ in range(6): x = subblock(x, 1080, **kwargs)
x = MaxPooling2D((2, 2), strides=(2, 2))(x) # 24x24x128
x = BatchNormalization()(x)
x = Conv2D(2048, (1,1), activation='relu', **kwargs)(x)
for _ in range(12): x = subblock(x, 2048, **kwargs)
x = GlobalMaxPooling2D()(x) # 2048
y = Dense(5004, use_bias=False, activation="softmax", name='patch_heatmap')(x)
branch_model = Model(inp, y)
branch_model.compile(optim, loss='categorical_crossentropy', metrics=['categorical_crossentropy', 'acc'])
return branch_model
build_averaging_model():
##############
# HEAD MODEL #
##############
#x = Average()(input_head)
inp = Input(shape=(5004,))
x = Dense(5004,use_bias=False, activation="softmax", name='patch_sum')(inp)
head_model = Model(inp, x, name='head')
#x = head_model([xa, xb])
return head_model
def build_model(head_model,branch_model):
inp = Input(shape=(patch_number,patch_size[0],patch_size[1],patch_size[2]))
input_head = []
for i in range(patch_number):
input_head.append(branch_model(inp[i,:,:,:]))
x = Average()(input_head)
x = head_model(x)
model = Model(inp,x)
return model