# Copyright 2019 Google LLC
#
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""QOctave example."""
import numpy as np
import sys
from tensorflow.keras import activations
from tensorflow.keras import initializers
import tensorflow.keras.backend as K
from tensorflow.keras.layers import Input
from tensorflow.keras.models import Model
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.utils import to_categorical
from functools import partial
from qkeras import * # pylint: disable=wildcard-import
def create_model():
"""use qocatve in network."""
kernel_initializer=initializers.he_normal(seed=42)
x = x_in = Input(shape=(256, 256, 3))
# Block 1
high, low = QOctaveConv2D(
32, (3, 3),
alpha=0.5,
strides=(2, 2),
padding='valid',
kernel_initializer=kernel_initializer,
bias_initializer="zeros",
bias_quantizer="quantized_bits(4,1)",
depthwise_quantizer="quantized_bits(4,1)",
depthwise_activation="quantized_bits(6,2,1)",
pointwise_quantizer="quantized_bits(4,1)",
acc_quantizer="quantized_bits(16,7,1)",
activation="quantized_relu(6,2)",
use_separable=True,
name='block1_conv1')([x, None])
# Block 2
high, low = QOctaveConv2D(
64, (3, 3),
alpha=0.4,
strides=(2, 2),
padding='same',
kernel_initializer=kernel_initializer,
bias_initializer="zeros",
bias_quantizer="quantized_bits(4,1)",
depthwise_quantizer="quantized_bits(4,1)",
depthwise_activation="quantized_bits(6,2,1)",
pointwise_quantizer="quantized_bits(4,1)",
acc_quantizer="quantized_bits(16,7,1)",
activation="quantized_relu(6,2)",
use_separable=True,
name='block2_conv1')([high, low])
# Block 3
high, low = QOctaveConv2D(
64, (3, 3),
alpha=0.4,
strides=(2, 2),
padding='same',
kernel_initializer=kernel_initializer,
bias_initializer="zeros",
bias_quantizer="quantized_bits(4,1)",
depthwise_quantizer="quantized_bits(4,1)",
depthwise_activation="quantized_bits(6,2,1)",
pointwise_quantizer="quantized_bits(4,1)",
acc_quantizer="quantized_bits(16,7,1)",
activation="quantized_relu(6,2)",
use_separable=True,
name='block3_conv1')([high, low])
high, low = QOctaveConv2D(
32, (3, 3),
alpha=0.4,
strides=(1, 1),
padding='same',
kernel_initializer=kernel_initializer,
bias_initializer='zeros',
bias_quantizer="quantized_bits(4,1)",
depthwise_quantizer="quantized_bits(4,1)",
depthwise_activation="quantized_bits(6,2,1)",
pointwise_quantizer="quantized_bits(4,1)",
acc_quantizer="quantized_bits(16,7,1)",
activation="quantized_relu(6,2)",
use_separable=True,
name='block3_conv2')([high, low])
high, low = QOctaveConv2D(
32, (3, 3),
alpha=0.3,
strides=(1, 1),
padding='same',
kernel_initializer=kernel_initializer,
bias_initializer='zeros',
bias_quantizer="quantized_bits(4,1)",
depthwise_quantizer="quantized_bits(4,1)",
depthwise_activation="quantized_bits(6,2,1)",
pointwise_quantizer="quantized_bits(4,1)",
acc_quantizer="quantized_bits(16,7,1)",
activation="quantized_relu(6,2)",
use_separable=True,
name='block3_conv3')([high, low])
x, _ = QOctaveConv2D(
32, (3, 3),
alpha=0.0,
strides=(2, 2),
padding='same',
kernel_initializer=kernel_initializer,
bias_initializer='zeros',
bias_quantizer="quantized_bits(4,1)",
depthwise_quantizer="quantized_bits(4,1)",
depthwise_activation="quantized_bits(6,2,1)",
pointwise_quantizer="quantized_bits(4,1)",
acc_quantizer="quantized_bits(16,7,1)",
activation="quantized_relu(6,2)",
use_separable=True,
name='block3_conv_down')([high, low])
# Upsample
x = UpSampling2D(size=(2, 2), data_format="channels_last")(x)
x = QConv2D(
2, (2, 2),
strides=(1, 1),
kernel_initializer=kernel_initializer,
bias_initializer="ones",
kernel_quantizer=quantized_bits(4, 0, 1),
bias_quantizer=quantized_bits(4, 0, 1),
padding="same",
name="conv_up")(
x)
x = Activation("softmax", name="softmax")(x)
output = x
model = Model(x_in, output, name='qoctave_network')
return model
# Create the model
def customLoss(y_true,y_pred):
log1 = 1.5 * y_true * K.log(y_pred + 1e-9) * K.pow(1-y_pred, 2)
log0 = 0.5 * (1 - y_true) * K.log((1 - y_pred) + 1e-9) * K.pow(y_pred, 2)
return (- K.sum(K.mean(log0 + log1, axis = 0)))
if __name__ == '__main__':
model = create_model()
model.compile(optimizer="Adam", loss=customLoss, metrics=['acc'])
model.summary(line_length=100)
print_qstats(model)
