import os
import sys
import tensorflow as tf
import scipy
import numpy as np
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.append(os.path.join(BASE_DIR, '../utils'))
import tf_util
import pointnet
from custom_layers import Scale
from keras.layers import Input, Dense, Convolution2D, MaxPooling2D, AveragePooling2D, ZeroPadding2D, Dropout, Flatten, add, Reshape, Activation
from keras.layers.normalization import BatchNormalization
from keras.models import Model
from keras import backend as K
K.set_learning_phase(1) #set learning phase
def placeholder_inputs(batch_size, img_rows=224, img_cols=224, points=16384, separately=False):
imgs_pl = tf.placeholder(tf.float32, shape=(batch_size, img_rows, img_cols, 3))
pts_pl = tf.placeholder(tf.float32, shape=(batch_size, points, 3))
if separately:
speeds_pl = tf.placeholder(tf.float32, shape=(batch_size))
angles_pl = tf.placeholder(tf.float32, shape=(batch_size))
labels_pl = [speeds_pl, angles_pl]
labels_pl = tf.placeholder(tf.float32, shape=(batch_size, 2))
return imgs_pl, pts_pl, labels_pl
def get_resnet(img_rows=224, img_cols=224, separately=False):
"""
Resnet 152 Model for Keras
Model Schema and layer naming follow that of the original Caffe implementation
https://github.com/KaimingHe/deep-residual-networks
ImageNet Pretrained Weights
Theano: https://drive.google.com/file/d/0Byy2AcGyEVxfZHhUT3lWVWxRN28/view?usp=sharing
TensorFlow: https://drive.google.com/file/d/0Byy2AcGyEVxfeXExMzNNOHpEODg/view?usp=sharing
Parameters:
img_rows, img_cols - resolution of inputs
channel - 1 for grayscale, 3 for color
"""
img_input = Input(shape=(img_rows, img_cols, 3), name='data')
eps = 1.1e-5
x = ZeroPadding2D((3, 3), name='conv1_zeropadding')(img_input)
x = Convolution2D(64, (7, 7), strides=(2, 2), name='conv1', use_bias=False)(x)
x = BatchNormalization(epsilon=eps, axis=3, name='bn_conv1')(x)
x = Scale(axis=3, name='scale_conv1')(x)
x = Activation('relu', name='conv1_relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2), name='pool1')(x)
x = conv_block(x, 3, [64, 64, 256], stage=2, block='a', strides=(1, 1))
x = identity_block(x, 3, [64, 64, 256], stage=2, block='b')
x = identity_block(x, 3, [64, 64, 256], stage=2, block='c')
x = conv_block(x, 3, [128, 128, 512], stage=3, block='a')
for i in range(1,8):
x = identity_block(x, 3, [128, 128, 512], stage=3, block='b'+str(i))
x = conv_block(x, 3, [256, 256, 1024], stage=4, block='a')
for i in range(1,36):
x = identity_block(x, 3, [256, 256, 1024], stage=4, block='b'+str(i))
x = conv_block(x, 3, [512, 512, 2048], stage=5, block='a')
x = identity_block(x, 3, [512, 512, 2048], stage=5, block='b')
x = identity_block(x, 3, [512, 512, 2048], stage=5, block='c')
x_fc = AveragePooling2D((7, 7), name='avg_pool')(x)
x_fc = Flatten()(x_fc)
x_fc = Dense(1000, activation='softmax', name='fc1000')(x_fc)
model = Model(img_input, x_fc)
# Use pre-trained weights for Tensorflow backend
weights_path = 'utils/weights/resnet152_weights_tf.h5'
assert (os.path.exists(weights_path))
model.load_weights(weights_path, by_name=True)
# Truncate and replace softmax layer for transfer learning
# Cannot use model.layers.pop() since model is not of Sequential() type
# The method below works since pre-trained weights are stored in layers but not in the model
x_newfc = AveragePooling2D((7, 7), name='avg_pool')(x)
x_newfc = Flatten()(x_newfc)
x_newfc = Dense(256, name='fc8')(x_newfc)
model = Model(img_input, x_newfc)
return model
def get_model(net, is_training, add_lstm=False, bn_decay=None, separately=False):
""" ResNet152 regression model, input is BxWxHx3, output Bx2"""
batch_size = net[0].get_shape()[0].value
img_net, pt_net = net[0], net[1]
img_net = get_resnet(224, 224)(img_net)
with tf.variable_scope('pointnet'):
pt_net = pointnet.get_model(pt_net, tf.constant(True))
net = tf.reshape(tf.stack([img_net, pt_net], axis=2), [batch_size, -1])
if not add_lstm:
for i, dim in enumerate([256, 128, 16]):
fc_scope = "fc" + str(i + 1)
dp_scope = "dp" + str(i + 1)
net = tf_util.fully_connected(net, dim, bn=True,
is_training=is_training,
scope=fc_scope,
bn_decay=bn_decay)
net = tf_util.dropout(net, keep_prob=0.7,
is_training=is_training,
scope=dp_scope)
net = tf_util.fully_connected(net, 2, activation_fn=None, scope='fc4')
else:
fc_scope = "fc1"
net = tf_util.fully_connected(net, 784, bn=True,
is_training=is_training,
scope=fc_scope,
bn_decay=bn_decay)
net = tf_util.dropout(net, keep_prob=0.7,
is_training=is_training,
scope="dp1")
net = cnn_lstm_block(net)
return net
def cnn_lstm_block(input_tensor):
lstm_in = tf.reshape(input_tensor, [-1, 28, 28])
lstm_out = tf_util.stacked_lstm(lstm_in,
num_outputs=10,
time_steps=28,
scope="cnn_lstm")
W_final = tf.Variable(tf.truncated_normal([10, 2], stddev=0.1))
b_final = tf.Variable(tf.truncated_normal([2], stddev=0.1))
return tf.multiply(tf.atan(tf.matmul(lstm_out, W_final) + b_final), 2)
def identity_block(input_tensor, kernel_size, filters, stage, block):
'''The identity_block is the block that has no conv layer at shortcut
# Arguments
input_tensor: input tensor
kernel_size: defualt 3, the kernel size of middle conv layer at main path
filters: list of integers, the nb_filters of 3 conv layer at main path
stage: integer, current stage label, used for generating layer names
block: 'a','b'..., current block label, used for generating layer names
'''
eps = 1.1e-5
nb_filter1, nb_filter2, nb_filter3 = filters
conv_name_base = 'res' + str(stage) + block + '_branch'
bn_name_base = 'bn' + str(stage) + block + '_branch'
scale_name_base = 'scale' + str(stage) + block + '_branch'
x = Convolution2D(nb_filter1, (1, 1), name=conv_name_base + '2a', use_bias=False)(input_tensor)
x = BatchNormalization(epsilon=eps, axis=3, name=bn_name_base + '2a')(x)
x = Scale(axis=3, name=scale_name_base + '2a')(x)
x = Activation('relu', name=conv_name_base + '2a_relu')(x)
x = ZeroPadding2D((1, 1), name=conv_name_base + '2b_zeropadding')(x)
x = Convolution2D(nb_filter2, (kernel_size, kernel_size),
name=conv_name_base + '2b', use_bias=False)(x)
x = BatchNormalization(epsilon=eps, axis=3, name=bn_name_base + '2b')(x)
x = Scale(axis=3, name=scale_name_base + '2b')(x)
x = Activation('relu', name=conv_name_base + '2b_relu')(x)
x = Convolution2D(nb_filter3, (1, 1), name=conv_name_base + '2c', use_bias=False)(x)
x = BatchNormalization(epsilon=eps, axis=3, name=bn_name_base + '2c')(x)
x = Scale(axis=3, name=scale_name_base + '2c')(x)
x = add([x, input_tensor], name='res' + str(stage) + block)
x = Activation('relu', name='res' + str(stage) + block + '_relu')(x)
return x
def conv_block(input_tensor, kernel_size, filters, stage, block, strides=(2, 2)):
'''conv_block is the block that has a conv layer at shortcut
# Arguments
input_tensor: input tensor
kernel_size: defualt 3, the kernel size of middle conv layer at main path
filters: list of integers, the nb_filters of 3 conv layer at main path
stage: integer, current stage label, used for generating layer names
block: 'a','b'..., current block label, used for generating layer names
Note that from stage 3, the first conv layer at main path is with subsample=(2,2)
And the shortcut should have subsample=(2,2) as well
'''
eps = 1.1e-5
nb_filter1, nb_filter2, nb_filter3 = filters
conv_name_base = 'res' + str(stage) + block + '_branch'
bn_name_base = 'bn' + str(stage) + block + '_branch'
scale_name_base = 'scale' + str(stage) + block + '_branch'
x = Convolution2D(nb_filter1, (1, 1), strides=strides,
name=conv_name_base + '2a', use_bias=False)(input_tensor)
x = BatchNormalization(epsilon=eps, axis=3, name=bn_name_base + '2a')(x)
x = Scale(axis=3, name=scale_name_base + '2a')(x)
x = Activation('relu', name=conv_name_base + '2a_relu')(x)
x = ZeroPadding2D((1, 1), name=conv_name_base + '2b_zeropadding')(x)
x = Convolution2D(nb_filter2, (kernel_size, kernel_size),
name=conv_name_base + '2b', use_bias=False)(x)
x = BatchNormalization(epsilon=eps, axis=3, name=bn_name_base + '2b')(x)
x = Scale(axis=3, name=scale_name_base + '2b')(x)
x = Activation('relu', name=conv_name_base + '2b_relu')(x)
x = Convolution2D(nb_filter3, (1, 1), name=conv_name_base + '2c', use_bias=False)(x)
x = BatchNormalization(epsilon=eps, axis=3, name=bn_name_base + '2c')(x)
x = Scale(axis=3, name=scale_name_base + '2c')(x)
shortcut = Convolution2D(nb_filter3, (1, 1), strides=strides,
name=conv_name_base + '1', use_bias=False)(input_tensor)
shortcut = BatchNormalization(epsilon=eps, axis=3, name=bn_name_base + '1')(shortcut)
shortcut = Scale(axis=3, name=scale_name_base + '1')(shortcut)
x = add([x, shortcut], name='res' + str(stage) + block)
x = Activation('relu', name='res' + str(stage) + block + '_relu')(x)
return x
def get_loss(pred, label, l2_weight=0.0001):
diff = tf.square(tf.subtract(pred, label))
train_vars = tf.trainable_variables()
l2_loss = tf.add_n([tf.nn.l2_loss(v) for v in train_vars[1:]]) * l2_weight
loss = tf.reduce_mean(diff + l2_loss)
tf.summary.scalar('l2 loss', l2_loss * l2_weight)
tf.summary.scalar('loss', loss)
return loss
def summary_scalar(pred, label):
threholds = [5, 4, 3, 2, 1, 0.5]
angles = [float(t) / 180 * scipy.pi for t in threholds]
speeds = [float(t) / 20 for t in threholds]
for i in range(len(threholds)):
scalar_angle = "angle(" + str(angles[i]) + ")"
scalar_speed = "speed(" + str(speeds[i]) + ")"
ac_angle = tf.abs(tf.subtract(pred[:, 1], label[:, 1])) < threholds[i]
ac_speed = tf.abs(tf.subtract(pred[:, 0], label[:, 0])) < threholds[i]
ac_angle = tf.reduce_mean(tf.cast(ac_angle, tf.float32))
ac_speed = tf.reduce_mean(tf.cast(ac_speed, tf.float32))
tf.summary.scalar(scalar_angle, ac_angle)
tf.summary.scalar(scalar_speed, ac_speed)
def resize(imgs):
batch_size = imgs.shape[0]
imgs_new = []
for j in range(batch_size):
img = imgs[j,:,:,:]
new = scipy.misc.imresize(img, (224, 224))
imgs_new.append(new)
imgs_new = np.stack(imgs_new, axis=0)
return imgs_new
if __name__ == '__main__':
with tf.Graph().as_default():
imgs = tf.zeros((32, 224, 224, 3))
pts = tf.zeros((32, 16384, 3))
outputs = get_model([imgs, pts], tf.constant(True))
print(outputs)
