def build_model(weight_path):
#base_model = ResNet50(input_shape=(224,224,3), weights=None, include_top=False)
#base_model.load_weights(weight_path, by_name=True)
#X = base_model.output
#predictions = Dense(10, activation='softmax')(X)
model = ResNet18(Input((224, 224, 3)), classes=10)
model.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=["accuracy"])
return model
train_labels_categorical = keras.utils.to_categorical(train_labels,
num_classes=10)
vali_labels_categorical = keras.utils.to_categorical(vali_labels,
num_classes=10)
train_images_2d = train_images.values.reshape(train_images.shape[0], 28,
28, 1)
vali_images_2d = vali_images.values.reshape(vali_images.shape[0], 28, 28,
1)
test_images_2d = X_test.values.reshape(X_test.shape[0], 28, 28, 1)
shape, classes = (28, 28, 1), 10
x = keras.layers.Input(shape)
model = ResNet18(x, classes=classes)
RMS = RMSprop(lr=0.0001)
model.compile("adam", "categorical_crossentropy", ["accuracy"])
# model.compile(optimizer=RMS, loss="binary_crossentropy",metrics=['accuracy'])
# model = Sequential()
# model.add(Conv2D(32, (3, 3),padding='same', activation='relu', input_shape=(28, 28, 1)))
# model.add(Conv2D(32, (3, 3),padding='same', activation='relu'))
# model.add(MaxPooling2D(pool_size=(2, 2)))
# model.add(Dropout(0.25))
#
# model.add(Conv2D(64, (3, 3),padding='same', activation='relu'))
# model.add(Conv2D(64, (3, 3),padding='same', activation='relu'))
# model.add(MaxPooling2D(pool_size=(2, 2)))
# model.add(Dropout(0.25))
def centernet(num_classes, backbone='resnet50', input_size=512, max_objects=100, score_threshold=0.1, nms=True):
assert backbone in ['resnet18', 'resnet34', 'resnet50']
output_size = input_size // 4
image_input = Input(shape=(None, None, 3))
hm_input = Input(shape=(output_size, output_size, num_classes))
wh_input = Input(shape=(max_objects, 2))
reg_input = Input(shape=(max_objects, 2))
reg_mask_input = Input(shape=(max_objects,))
index_input = Input(shape=(max_objects,))
if backbone == 'resnet18':
resnet = ResNet18(image_input, include_top=False, freeze_bn=True)
elif backbone == 'resnet34':
resnet = ResNet34(image_input, include_top=False, freeze_bn=True)
else:
resnet = ResNet50(image_input, include_top=False, freeze_bn=True)
# C5 (b, 16, 16, 512)
C2, C3, C4, C5 = resnet.outputs
C5 = Dropout(rate=0.5)(C5)
C4 = Dropout(rate=0.4)(C4)
C3 = Dropout(rate=0.3)(C3)
C2 = Dropout(rate=0.2)(C2)
x = C5
# decoder
x = Conv2D(256, 1, padding='same', use_bias=False,
kernel_initializer='he_normal',
kernel_regularizer=l2(5e-4))(UpSampling2D()(x))
x = BatchNormalization()(x)
x = ReLU()(x)
x = Concatenate()([C4, x])
x = Conv2D(256, 3, padding='same', use_bias=False,
kernel_initializer='he_normal',
kernel_regularizer=l2(5e-4))(x)
x = BatchNormalization()(x)
# (b, 32, 32, 512)
x = ReLU()(x)
x = Conv2D(128, 1, padding='same', use_bias=False,
kernel_initializer='he_normal',
kernel_regularizer=l2(5e-4))(UpSampling2D()(x))
x = BatchNormalization()(x)
x = ReLU()(x)
x = Concatenate()([C3, x])
x = Conv2D(128, 3, padding='same', use_bias=False,
kernel_initializer='he_normal',
kernel_regularizer=l2(5e-4))(x)
x = BatchNormalization()(x)
# (b, 64, 64, 128)
x = ReLU()(x)
x = Conv2D(64, 1, padding='same', use_bias=False,
kernel_initializer='he_normal',
kernel_regularizer=l2(5e-4))(UpSampling2D()(x))
x = BatchNormalization()(x)
x = ReLU()(x)
x = Concatenate()([C2, x])
x = Conv2D(64, 3, padding='same', use_bias=False,
kernel_initializer='he_normal',
kernel_regularizer=l2(5e-4))(x)
x = BatchNormalization()(x)
# (b, 128, 128, 512)
x = ReLU()(x)
# hm header
y1 = Conv2D(64, 3, padding='same', use_bias=False, kernel_initializer='he_normal', kernel_regularizer=l2(5e-4))(x)
y1 = BatchNormalization()(y1)
y1 = ReLU()(y1)
y1 = Conv2D(num_classes, 1, kernel_initializer='he_normal', kernel_regularizer=l2(5e-4), activation='sigmoid')(y1)
# wh header
y2 = Conv2D(64, 3, padding='same', use_bias=False, kernel_initializer='he_normal', kernel_regularizer=l2(5e-4))(x)
y2 = BatchNormalization()(y2)
y2 = ReLU()(y2)
y2 = Conv2D(2, 1, kernel_initializer='he_normal', kernel_regularizer=l2(5e-4))(y2)
# reg header
y3 = Conv2D(64, 3, padding='same', use_bias=False, kernel_initializer='he_normal', kernel_regularizer=l2(5e-4))(x)
y3 = BatchNormalization()(y3)
y3 = ReLU()(y3)
y3 = Conv2D(2, 1, kernel_initializer='he_normal', kernel_regularizer=l2(5e-4))(y3)
loss_ = Lambda(loss, name='centernet_loss')(
[y1, y2, y3, hm_input, wh_input, reg_input, reg_mask_input, index_input])
model = Model(inputs=[image_input, hm_input, wh_input, reg_input, reg_mask_input, index_input], outputs=[loss_])
# detections = decode(y1, y2, y3)
detections = Lambda(lambda x: decode(*x,
max_objects=max_objects,
score_threshold=score_threshold,
nms=nms,
num_classes=num_classes))([y1, y2, y3])
prediction_model = Model(inputs=image_input, outputs=detections)
debug_model = Model(inputs=image_input, outputs=[y1, y2, y3])
return model, prediction_model, debug_model
# Model version
n = 34
assert n in (18, 34), 'N must be 18 or 34'
depth = n
version = 1
# model name, depth and version
model_type = 'ResNet%dv%d' % (depth, version)
# Load the cifar10 data
# Input image dimensions
# Normalize data
# If subtract pixel mean is enabled
# Convert class vectors to binary class matrices
def lr_schedule(epoch):
pass
if n==18:
model = ResNet18(input_shape=input_shape,depth=depth)
else:
model = ResNet34(input_shape = input_shape,depth=depth)
model.compile(loss='categorical_crossentropy',optimizer=AdaBound())