class Model:
'''
Implements a simple two layer convLSTM architecture, no feedback.
'''
def __init__(self, sess, num_classes,
train_featurizer='FINE_TUNE', train_main_model='FROM_SCRATCH', train_fc='FROM_SCRATCH',
weights_filename=None, is_training=True):
self.sess = sess
self.weights = np.load(weights_filename) if weights_filename is not None else None
self.num_classes = num_classes
assert train_featurizer in ['NO', 'FINE_TUNE', 'FROM_SCRATCH'], 'train_featurizer must be either: NO, FINE_TUNE, or FROM_SCRATCH'
self.train_featurizer = train_featurizer if is_training else 'NO'
assert train_main_model in ['NO', 'FINE_TUNE', 'FROM_SCRATCH'], 'train_main_model must be either: NO, FINE_TUNE, or FROM_SCRATCH'
self.train_main_model = train_main_model if is_training else 'NO'
assert train_fc in ['NO', 'FINE_TUNE', 'FROM_SCRATCH'], 'train_fc must be either: NO, FINE_TUNE, or FROM_SCRATCH'
self.train_fc = train_fc if is_training else 'NO'
self.is_training = is_training
self.featurizer_variables = []
self.main_model_variables = []
self.fc_variables = []
self.vfeedbacknet_base = VFeedbackNetBase(sess, num_classes, is_training=is_training)
self._declare_variables()
def _declare_variables(self):
with tf.variable_scope('vfeedbacknet_model1'):
with tf.variable_scope('convlstm1'):
with tf.variable_scope('rnn'):
with tf.variable_scope('conv_lstm_cell'):
regularizer = None # tf.contrib.layers.l2_regularizer(scale=0.25)
initializer = tf.contrib.layers.xavier_initializer()
n = 512
m = 4*n
input_size = [7, 7, n]
kernel2d_size = [3, 3]
kernel_size = kernel2d_size + [2*n] + [m]
with tf.variable_scope('convlstm'):
kernel = tf.get_variable('kernel', kernel_size, initializer=initializer, regularizer=regularizer)
W_ci = tf.get_variable('W_ci', input_size, initializer=initializer, regularizer=regularizer)
W_cf = tf.get_variable('W_cf', input_size, initializer=initializer, regularizer=regularizer)
W_co = tf.get_variable('W_co', input_size, initializer=initializer, regularizer=regularizer)
bias = tf.get_variable('bias', [m], initializer=tf.zeros_initializer(), regularizer=regularizer)
self.convLSTMCell1 = ConvLSTMCell([7, 7], 512, [3, 3])
with tf.variable_scope('convlstm2'):
with tf.variable_scope('rnn'):
with tf.variable_scope('conv_lstm_cell'):
regularizer = None # tf.contrib.layers.l2_regularizer(scale=0.25)
initializer = tf.contrib.layers.xavier_initializer()
n = 512
m = 4*n
input_size = [7, 7, n]
kernel2d_size = [3, 3]
kernel_size = kernel2d_size + [2*n] + [m]
with tf.variable_scope('convlstm'):
kernel = tf.get_variable('kernel', kernel_size, initializer=initializer, regularizer=regularizer)
W_ci = tf.get_variable('W_ci', input_size, initializer=initializer, regularizer=regularizer)
W_cf = tf.get_variable('W_cf', input_size, initializer=initializer, regularizer=regularizer)
W_co = tf.get_variable('W_co', input_size, initializer=initializer, regularizer=regularizer)
bias = tf.get_variable('bias', [m], initializer=tf.zeros_initializer(), regularizer=regularizer)
self.convLSTMCell2 = ConvLSTMCell([7, 7], 512, [3, 3])
def get_variables(self):
return self.featurizer_variables + self.main_model_variables + self.fc_variables
def print_variables(self):
for var in self.get_variables():
print(var.name)
def initialize_variables(self):
logging.debug('--- begin variable initialization (vfeedbacknet) ---')
if self.train_featurizer == 'FROM_SCRATCH':
logging.debug('vgg16:FROM_SCRATCH; using random initialization')
for var in self.featurizer_variables:
self.sess.run(var.initializer)
else:
for var in self.featurizer_variables:
logging.debug('LOADING FROM WEIGHTS_FILE {}: {}'.format(var.name, var.shape))
assert self.weights is not None, 'Need to specify weights file to load from for featurizer_variables'
self.sess.run(var.assign(self.weights[var.name]))
if self.train_main_model == 'FROM_SCRATCH':
logging.debug('feedback: FROM_SCRATCH; using random initialization')
for var in self.main_model_variables:
self.sess.run(var.initializer)
else:
for var in self.main_model_variables:
logging.debug(' LOADING FROM WEIGHTS_FILE {}: {}'.format(var.name, var.shape))
assert self.weights is not None, 'Need to specify weights file to load from for main_model_variables'
self.sess.run(var.assign(self.weights[var.name]))
if self.train_fc == 'FROM_SCRATCH':
logging.debug('fc: FROM_SCRATCH; using random initialization')
for var in self.fc_variables:
self.sess.run(var.initializer)
else:
for var in self.fc_variables:
logging.debug('LOADING FROM WEIGHTS_FILE {}: {}'.format(var.name, var.shape))
assert self.weights is not None, 'Need to specify weights file to load from for fc_variables'
self.sess.run(var.assign(self.weights[var.name]))
logging.debug('--- end variable initialization (vfeedbacknet) ---')
def export_variables(self, export_filename):
VFeedbackNetBase.export_variables(self.sess, self.get_variables(), export_filename)
def __call__(self, inputs, inputs_sequence_length):
#assert inputs.shape[1:] == (20, 112, 112), 'expected input shape of (20, 112, 112) but got {}'.format(inputs.shape)
ModelLogger.log('raw_input', inputs)
inputs = self.vfeedbacknet_base.split_video(inputs)
ModelLogger.log('input', inputs)
## featurizer ##
inputs = [ self.vfeedbacknet_base.vgg16_layer1(inp, var_list=self.featurizer_variables) for inp in inputs ]
ModelLogger.log('vgg-layer1', inputs)
inputs = [ self.vfeedbacknet_base.vgg16_layer2(inp, var_list=self.featurizer_variables) for inp in inputs ]
ModelLogger.log('vgg-layer2', inputs)
## main model ##
inputs = [ self.vfeedbacknet_base.vgg16_layer3(inp, var_list=self.featurizer_variables) for inp in inputs ]
ModelLogger.log('vgg-layer3', inputs)
inputs = [ self.vfeedbacknet_base.vgg16_layer4(inp, var_list=self.featurizer_variables) for inp in inputs ]
ModelLogger.log('vgg-layer4', inputs)
inputs = [ self.vfeedbacknet_base.vgg16_layer5(inp, var_list=self.featurizer_variables) for inp in inputs ]
ModelLogger.log('vgg-layer5', inputs)
inputs = self.convLSTM_layer1(inputs, inputs_sequence_length, var_list=self.main_model_variables)
ModelLogger.log('convLSTM1', inputs)
## ave_pool and fc ##
inputs = [ self.vfeedbacknet_base.ave_pool(inp) for inp in inputs ]
ModelLogger.log('ave_pool', inputs)
inputs = [ self.vfeedbacknet_base.fc_layer(inp, var_list=self.fc_variables) for inp in inputs ]
ModelLogger.log('fc', inputs)
logits = tf.stack(inputs, axis=1)
logits = tf.expand_dims(logits, axis=1)
ModelLogger.log('logits', logits)
return logits
def convLSTM_layer1(self, inputs, inputs_sequence_length, var_list=None):
with tf.variable_scope('vfeedbacknet_model1'):
with tf.variable_scope('convlstm1'):
with tf.variable_scope('rnn'):
with tf.variable_scope('conv_lstm_cell'):
with tf.variable_scope('convlstm', reuse=True):
kernel = tf.get_variable('kernel')
W_ci = tf.get_variable('W_ci')
W_cf = tf.get_variable('W_cf')
W_co = tf.get_variable('W_co')
bias = tf.get_variable('bias')
if var_list is not None and kernel not in var_list:
var_list.append(kernel)
if var_list is not None and W_ci not in var_list:
var_list.append(W_ci)
if var_list is not None and W_cf not in var_list:
var_list.append(W_cf)
if var_list is not None and W_co not in var_list:
var_list.append(W_co)
if var_list is not None and bias not in var_list:
var_list.append(bias)
inputs, state = tf.nn.dynamic_rnn(
self.convLSTMCell1,
tf.stack(inputs, axis=1),
dtype=tf.float32,
#sequence_length=inputs_sequence_length,
)
inputs = tf.unstack(inputs, axis=1)
return inputs
class Model:
'''
TODO(jremmons) add description
'''
model_name = 'model16'
def __init__(self,
sess,
num_classes,
batch_size,
train_featurizer='FINE_TUNE',
train_main_model='FINE_TUNE',
train_fc='FINE_TUNE',
weights_filename=None,
is_training=True):
self.sess = sess
self.weights = np.load(
weights_filename) if weights_filename is not None else None
self.num_classes = num_classes
self.batch_size = batch_size
assert train_featurizer in [
'NO', 'FINE_TUNE', 'FROM_SCRATCH'
], 'train_featurizer must be either: NO, FINE_TUNE, or FROM_SCRATCH'
self.train_featurizer = train_featurizer if is_training else 'NO'
assert train_main_model in [
'NO', 'FINE_TUNE', 'FROM_SCRATCH'
], 'train_main_model must be either: NO, FINE_TUNE, or FROM_SCRATCH'
self.train_main_model = train_main_model if is_training else 'NO'
assert train_fc in [
'NO', 'FINE_TUNE', 'FROM_SCRATCH'
], 'train_fc must be either: NO, FINE_TUNE, or FROM_SCRATCH'
self.train_fc = train_fc if is_training else 'NO'
self.is_training = is_training
self.featurizer_variables = []
self.main_model_variables = []
self.fc_variables = []
self.vfeedbacknet_base = VFeedbackNetBase(sess,
num_classes,
train_vgg16=train_featurizer,
is_training=is_training)
self._declare_variables()
def _declare_variables(self):
with tf.variable_scope('vfeedbacknet_{}'.format(Model.model_name)):
# with tf.variable_scope('process_featurizer_output'):
# with tf.variable_scope('conv1'):
# regularizer = None # tf.contrib.layers.l2_regularizer(scale=0.25)
# initializer = tf.contrib.layers.xavier_initializer()
# kernel = tf.get_variable('kernel', shape=[3, 3, 512, 512], dtype=tf.float32, regularizer=regularizer, initializer=initializer)
# biases = tf.get_variable('biases', shape=[512], dtype=tf.float32, regularizer=regularizer, initializer=initializer)
with tf.variable_scope('convlstm1'):
with tf.variable_scope('rnn'):
with tf.variable_scope('conv_lstm_cell'):
regularizer = None # tf.contrib.layers.l2_regularizer(scale=0.25)
initializer = tf.contrib.layers.xavier_initializer()
n = 512
m = 4 * n
input_size = [7, 7, n]
kernel2d_size = [3, 3]
kernel_size = kernel2d_size + [2 * n] + [m]
with tf.variable_scope('convlstm'):
kernel = tf.get_variable('kernel',
kernel_size,
initializer=initializer,
regularizer=regularizer)
W_ci = tf.get_variable('W_ci',
input_size,
initializer=initializer,
regularizer=regularizer)
W_cf = tf.get_variable('W_cf',
input_size,
initializer=initializer,
regularizer=regularizer)
W_co = tf.get_variable('W_co',
input_size,
initializer=initializer,
regularizer=regularizer)
bias = tf.get_variable(
'bias', [m],
initializer=tf.zeros_initializer(),
regularizer=regularizer)
self.convLSTMCell1 = ConvLSTMCell(input_size[:2], n, [3, 3])
# with tf.variable_scope('convlstm2'):
# with tf.variable_scope('rnn'):
# with tf.variable_scope('conv_lstm_cell'):
# regularizer = None # tf.contrib.layers.l2_regularizer(scale=0.25)
# initializer = tf.contrib.layers.xavier_initializer()
# n = 512
# m = 4*n
# input_size = [7, 7, n]
# kernel2d_size = [3, 3]
# kernel_size = kernel2d_size + [2*n] + [m]
# with tf.variable_scope('convlstm'):
# kernel = tf.get_variable('kernel', kernel_size, initializer=initializer, regularizer=regularizer)
# W_ci = tf.get_variable('W_ci', input_size, initializer=initializer, regularizer=regularizer)
# W_cf = tf.get_variable('W_cf', input_size, initializer=initializer, regularizer=regularizer)
# W_co = tf.get_variable('W_co', input_size, initializer=initializer, regularizer=regularizer)
# bias = tf.get_variable('bias', [m], initializer=tf.zeros_initializer(), regularizer=regularizer)
# self.convLSTMCell2 = ConvLSTMCell(input_size[:2], n, [3, 3])
with tf.variable_scope('reshape_convs'):
with tf.variable_scope('conv1'):
regularizer = None # tf.contrib.layers.l2_regularizer(scale=0.25)
initializer = tf.contrib.layers.xavier_initializer()
kernel = tf.get_variable('kernel',
shape=[3, 3, 256, 512],
dtype=tf.float32,
regularizer=regularizer,
initializer=initializer)
biases = tf.get_variable('biases',
shape=[512],
dtype=tf.float32,
regularizer=regularizer,
initializer=initializer)
# with tf.variable_scope('conv2'):
# regularizer = None # tf.contrib.layers.l2_regularizer(scale=0.25)
# initializer = tf.contrib.layers.xavier_initializer()
# kernel = tf.get_variable('kernel', shape=[3, 3, 512, 1024], dtype=tf.float32, regularizer=regularizer, initializer=initializer)
# biases = tf.get_variable('biases', shape=[1024], dtype=tf.float32, regularizer=regularizer, initializer=initializer)
# with tf.variable_scope('conv3'):
# regularizer = None # tf.contrib.layers.l2_regularizer(scale=0.25)
# initializer = tf.contrib.layers.xavier_initializer()
# kernel = tf.get_variable('kernel', shape=[3, 3, 512, 1024], dtype=tf.float32, regularizer=regularizer, initializer=initializer)
# biases = tf.get_variable('biases', shape=[1024], dtype=tf.float32, regularizer=regularizer, initializer=initializer)
with tf.variable_scope('feedback_block1'):
regularizer = None # tf.contrib.layers.l2_regularizer(scale=0.25)
initializer = tf.contrib.layers.xavier_initializer()
input_size = [7, 7]
kernel_size = [3, 3, 512, 512]
W_xf = tf.get_variable('W_xf',
kernel_size,
dtype=tf.float32,
initializer=initializer,
regularizer=regularizer)
W_xi = tf.get_variable('W_xi',
kernel_size,
dtype=tf.float32,
initializer=initializer,
regularizer=regularizer)
W_xc = tf.get_variable('W_xc',
kernel_size,
dtype=tf.float32,
initializer=initializer,
regularizer=regularizer)
W_xo = tf.get_variable('W_xo',
kernel_size,
dtype=tf.float32,
initializer=initializer,
regularizer=regularizer)
W_hf = tf.get_variable('W_hf',
kernel_size,
dtype=tf.float32,
initializer=initializer,
regularizer=regularizer)
W_hi = tf.get_variable('W_hi',
kernel_size,
dtype=tf.float32,
initializer=initializer,
regularizer=regularizer)
W_hc = tf.get_variable('W_hc',
kernel_size,
dtype=tf.float32,
initializer=initializer,
regularizer=regularizer)
W_ho = tf.get_variable('W_ho',
kernel_size,
dtype=tf.float32,
initializer=initializer,
regularizer=regularizer)
W_cf = tf.get_variable(
'W_cf', [input_size[0], input_size[1], kernel_size[-1]],
dtype=tf.float32,
initializer=initializer,
regularizer=regularizer)
W_ci = tf.get_variable(
'W_ci', [input_size[0], input_size[1], kernel_size[-1]],
dtype=tf.float32,
initializer=initializer,
regularizer=regularizer)
W_co = tf.get_variable(
'W_co', [input_size[0], input_size[1], kernel_size[-1]],
dtype=tf.float32,
initializer=initializer,
regularizer=regularizer)
b_f = tf.get_variable('b_f', [kernel_size[-1]],
dtype=tf.float32,
initializer=tf.zeros_initializer(),
regularizer=regularizer)
b_i = tf.get_variable('b_i', [kernel_size[-1]],
dtype=tf.float32,
initializer=tf.zeros_initializer(),
regularizer=regularizer)
b_c = tf.get_variable('b_c', [kernel_size[-1]],
dtype=tf.float32,
initializer=tf.zeros_initializer(),
regularizer=regularizer)
b_o = tf.get_variable('b_o', [kernel_size[-1]],
dtype=tf.float32,
initializer=tf.zeros_initializer(),
regularizer=regularizer)
# with tf.variable_scope('feedback_block2'):
# regularizer = None # tf.contrib.layers.l2_regularizer(scale=0.25)
# initializer = tf.contrib.layers.xavier_initializer()
# input_size = [14, 14]
# kernel_size = [3, 3, 256, 256]
# W_xf = tf.get_variable('W_xf', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_xi = tf.get_variable('W_xi', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_xc = tf.get_variable('W_xc', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_xo = tf.get_variable('W_xo', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_hf = tf.get_variable('W_hf', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_hi = tf.get_variable('W_hi', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_hc = tf.get_variable('W_hc', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_ho = tf.get_variable('W_ho', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_cf = tf.get_variable('W_cf', [input_size[0],input_size[1],kernel_size[-1]], dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_ci = tf.get_variable('W_ci', [input_size[0],input_size[1],kernel_size[-1]], dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_co = tf.get_variable('W_co', [input_size[0],input_size[1],kernel_size[-1]], dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# b_f = tf.get_variable('b_f', [kernel_size[-1]], dtype=tf.float32, initializer=tf.zeros_initializer(), regularizer=regularizer)
# b_i = tf.get_variable('b_i', [kernel_size[-1]], dtype=tf.float32, initializer=tf.zeros_initializer(), regularizer=regularizer)
# b_c = tf.get_variable('b_c', [kernel_size[-1]], dtype=tf.float32, initializer=tf.zeros_initializer(), regularizer=regularizer)
# b_o = tf.get_variable('b_o', [kernel_size[-1]], dtype=tf.float32, initializer=tf.zeros_initializer(), regularizer=regularizer)
# with tf.variable_scope('feedback_block3'):
# regularizer = None # tf.contrib.layers.l2_regularizer(scale=0.25)
# initializer = tf.contrib.layers.xavier_initializer()
# input_size = [7, 7]
# kernel_size = [3, 3, 512, 512]
# W_xf = tf.get_variable('W_xf', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_xi = tf.get_variable('W_xi', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_xc = tf.get_variable('W_xc', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_xo = tf.get_variable('W_xo', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_hf = tf.get_variable('W_hf', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_hi = tf.get_variable('W_hi', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_hc = tf.get_variable('W_hc', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_ho = tf.get_variable('W_ho', kernel_size, dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_cf = tf.get_variable('W_cf', [input_size[0],input_size[1],kernel_size[-1]], dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_ci = tf.get_variable('W_ci', [input_size[0],input_size[1],kernel_size[-1]], dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# W_co = tf.get_variable('W_co', [input_size[0],input_size[1],kernel_size[-1]], dtype=tf.float32, initializer=initializer, regularizer=regularizer)
# b_f = tf.get_variable('b_f', [kernel_size[-1]], dtype=tf.float32, initializer=tf.zeros_initializer(), regularizer=regularizer)
# b_i = tf.get_variable('b_i', [kernel_size[-1]], dtype=tf.float32, initializer=tf.zeros_initializer(), regularizer=regularizer)
# b_c = tf.get_variable('b_c', [kernel_size[-1]], dtype=tf.float32, initializer=tf.zeros_initializer(), regularizer=regularizer)
# b_o = tf.get_variable('b_o', [kernel_size[-1]], dtype=tf.float32, initializer=tf.zeros_initializer(), regularizer=regularizer)
with tf.variable_scope('fc'):
regularizer = None # tf.contrib.layers.l2_regularizer(scale=0.25)
initializer = tf.contrib.layers.xavier_initializer()
trainable = False if self.train_fc == 'NO' else True
weight = tf.get_variable('weights',
shape=[512, self.num_classes],
dtype=tf.float32,
initializer=initializer,
regularizer=regularizer,
trainable=trainable)
biases = tf.get_variable('biases',
shape=[self.num_classes],
dtype=tf.float32,
initializer=initializer,
regularizer=regularizer,
trainable=trainable)
def get_variables(self):
return self.featurizer_variables + self.main_model_variables + self.fc_variables
def print_variables(self):
var_list = self.get_variables()
var_list_len = len(var_list)
for var, idx in zip(var_list, range(var_list_len)):
print(str(idx).zfill(3), var.name)
def initialize_variables(self):
logging.debug('--- begin variable initialization (vfeedbacknet) ---')
var_list = self.get_variables()
self.print_variables()
print('Number of variables in model: {}'.format(len(var_list)))
if self.train_featurizer == 'FROM_SCRATCH':
logging.debug('vgg16:FROM_SCRATCH; using random initialization')
for var in self.featurizer_variables:
self.sess.run(var.initializer)
else:
for var in self.featurizer_variables:
logging.debug('LOADING FROM WEIGHTS_FILE {}: {}'.format(
var.name, var.shape))
assert self.weights is not None, 'Need to specify weights file to load from for featurizer_variables'
self.sess.run(var.assign(self.weights[var.name]))
if self.train_main_model == 'FROM_SCRATCH':
logging.debug(
'feedback: FROM_SCRATCH; using random initialization')
for var in self.main_model_variables:
self.sess.run(var.initializer)
else:
for var in self.main_model_variables:
logging.debug(' LOADING FROM WEIGHTS_FILE {}: {}'.format(
var.name, var.shape))
assert self.weights is not None, 'Need to specify weights file to load from for main_model_variables'
self.sess.run(var.assign(self.weights[var.name]))
if self.train_fc == 'FROM_SCRATCH':
logging.debug('fc: FROM_SCRATCH; using random initialization')
for var in self.fc_variables:
self.sess.run(var.initializer)
else:
for var in self.fc_variables:
logging.debug('LOADING FROM WEIGHTS_FILE {}: {}'.format(
var.name, var.shape))
assert self.weights is not None, 'Need to specify weights file to load from for fc_variables'
self.sess.run(var.assign(self.weights[var.name]))
logging.debug('--- end variable initialization (vfeedbacknet) ---')
def export_variables(self, export_filename):
VFeedbackNetBase.export_variables(self.sess, self.get_variables(),
export_filename)
def __call__(self, inputs, inputs_sequence_length):
#assert inputs.shape[1:] == (20, 112, 112), 'expected input shape of (20, 112, 112) but got {}'.format(inputs.shape)
ModelLogger.log('raw_input', inputs)
inputs = self.vfeedbacknet_base.split_video(inputs)
ModelLogger.log('input', inputs)
## featurizer ##
inputs = [
self.vfeedbacknet_base.vgg16_layer1(
inp, var_list=self.featurizer_variables) for inp in inputs
]
ModelLogger.log('vgg-layer1', inputs)
inputs = [
self.vfeedbacknet_base.vgg16_layer2(
inp, var_list=self.featurizer_variables) for inp in inputs
]
ModelLogger.log('vgg-layer2', inputs)
inputs = [
self.vfeedbacknet_base.vgg16_layer3(
inp, var_list=self.featurizer_variables) for inp in inputs
]
ModelLogger.log('vgg-layer3', inputs)
inputs = [
self.vfeedbacknet_base.vgg16_layer4(
inp, var_list=self.featurizer_variables) for inp in inputs
]
ModelLogger.log('vgg-layer4', inputs)
inputs = self.convLSTM_layer1(inputs,
inputs_sequence_length,
var_list=self.main_model_variables)
ModelLogger.log('convLSTM1', inputs)
# inputs = [ self.vfeedbacknet_base.vgg16_layer5(inp, var_list=self.featurizer_variables) for inp in inputs ]
# ModelLogger.log('vgg-layer5', inputs)
## main model ##
logits = []
featurizer_outputs = inputs
feedback_outputs = None
# "feedback" 1
feedback_outputs11 = [
self.feedback_block1(inp, var_list=self.main_model_variables)
for inp in featurizer_outputs
]
inputs = list(map(lambda x: x['hidden_state'], feedback_outputs11))
ModelLogger.log('feedback_block1', inputs)
# feedback_outputs21 = [ self.feedback_block2(inp, var_list=self.main_model_variables) for inp in inputs ]
# inputs = list(map(lambda x : x['hidden_state'], feedback_outputs21))
# ModelLogger.log('feedback_block2', inputs)
# inputs = [ self.reshape_conv_layer(inp, 2, var_list=self.main_model_variables) for inp in inputs ]
# ModelLogger.log('reshape_conv_layer2', inputs)
# feedback_outputs31 = [ self.feedback_block3(inp, var_list=self.main_model_variables) for inp in inputs ]
# inputs = list(map(lambda x : x['hidden_state'], feedback_outputs31))
# ModelLogger.log('feedback_block3', inputs)
# inputs = [ self.reshape_conv_layer(inp, 3, var_list=self.main_model_variables) for inp in inputs ]
# ModelLogger.log('reshape_conv_layer3', inputs)
inputs = [self.vfeedbacknet_base.ave_pool(inp) for inp in inputs]
ModelLogger.log('ave_pool', inputs)
inputs = [
self.fc_layer(inp, var_list=self.fc_variables) for inp in inputs
]
ModelLogger.log('fc', inputs)
logits.append(tf.stack(inputs, axis=1))
# "feedback" 2
feedback_outputs12 = [
self.feedback_block1(inp,
state=state,
var_list=self.main_model_variables)
for inp, state in zip(featurizer_outputs, feedback_outputs11)
]
inputs = list(map(lambda x: x['hidden_state'], feedback_outputs12))
ModelLogger.log('feedback_block1', inputs)
# feedback_outputs22 = [ self.feedback_block2(inp, state=state, var_list=self.main_model_variables) for inp,state in zip(inputs, feedback_outputs21) ]
# inputs = list(map(lambda x : x['hidden_state'], feedback_outputs22))
# ModelLogger.log('feedback_block2', inputs)
# inputs = [ self.reshape_conv_layer(inp, 2, var_list=self.main_model_variables) for inp in inputs ]
# ModelLogger.log('reshape_conv_layer2', inputs)
# feedback_outputs32 = [ self.feedback_block3(inp, state=state, var_list=self.main_model_variables) for inp,state in zip(inputs, feedback_outputs31) ]
# inputs = list(map(lambda x : x['hidden_state'], feedback_outputs32))
# ModelLogger.log('feedback_block3', inputs)
# inputs = [ self.reshape_conv_layer(inp, 3, var_list=self.main_model_variables) for inp in inputs ]
# ModelLogger.log('reshape_conv_layer3', inputs)
inputs = [self.vfeedbacknet_base.ave_pool(inp) for inp in inputs]
ModelLogger.log('ave_pool', inputs)
inputs = [
self.fc_layer(inp, var_list=self.fc_variables) for inp in inputs
]
ModelLogger.log('fc', inputs)
logits.append(tf.stack(inputs, axis=1))
# "feedback" 3
feedback_outputs13 = [
self.feedback_block1(inp,
state=state,
var_list=self.main_model_variables)
for inp, state in zip(featurizer_outputs, feedback_outputs12)
]
inputs = list(map(lambda x: x['hidden_state'], feedback_outputs13))
ModelLogger.log('feedback_block1', inputs)
# feedback_outputs23 = [ self.feedback_block2(inp, state=state, var_list=self.main_model_variables) for inp,state in zip(inputs, feedback_outputs22) ]
# inputs = list(map(lambda x : x['hidden_state'], feedback_outputs23))
# ModelLogger.log('feedback_block2', inputs)
# inputs = [ self.reshape_conv_layer(inp, 2, var_list=self.main_model_variables) for inp in inputs ]
# ModelLogger.log('reshape_conv_layer2', inputs)
# feedback_outputs33 = [ self.feedback_block3(inp, state=state, var_list=self.main_model_variables) for inp,state in zip(inputs, feedback_outputs32) ]
# inputs = list(map(lambda x : x['hidden_state'], feedback_outputs33))
# ModelLogger.log('feedback_block3', inputs)
# inputs = [ self.reshape_conv_layer(inp, 3, var_list=self.main_model_variables) for inp in inputs ]
# ModelLogger.log('reshape_conv_layer3', inputs)
inputs = [self.vfeedbacknet_base.ave_pool(inp) for inp in inputs]
ModelLogger.log('ave_pool', inputs)
inputs = [
self.fc_layer(inp, var_list=self.fc_variables) for inp in inputs
]
ModelLogger.log('fc', inputs)
logits.append(tf.stack(inputs, axis=1))
# output
logits = tf.stack(logits, axis=1)
ModelLogger.log('combined-feedback-logits', logits)
return logits
def feedback_block1(self, inputs, state=None, var_list=None):
return self.feedback_block(1, inputs, state=state, var_list=var_list)
def feedback_block2(self, inputs, state=None, var_list=None):
return self.feedback_block(2, inputs, state=state, var_list=var_list)
def feedback_block3(self, inputs, state=None, var_list=None):
return self.feedback_block(3, inputs, state=state, var_list=var_list)
def feedback_block(self, block_num, inputs, state=None, var_list=None):
hidden_state = None
cell_state = None
if state is not None:
hidden_state = state['hidden_state']
cell_state = state['cell_state']
assert (cell_state is None) == (
hidden_state is None
), 'cell_state and hidden_state must BOTH be supplied as arguments.'
with tf.variable_scope('vfeedbacknet_{}'.format(Model.model_name)):
with tf.variable_scope('feedback_block{}'.format(block_num),
reuse=True):
W_xf = tf.get_variable('W_xf')
W_xi = tf.get_variable('W_xi')
W_xc = tf.get_variable('W_xc')
W_xo = tf.get_variable('W_xo')
W_hf = tf.get_variable('W_hf')
W_hi = tf.get_variable('W_hi')
W_hc = tf.get_variable('W_hc')
W_ho = tf.get_variable('W_ho')
W_cf = tf.get_variable('W_cf')
W_ci = tf.get_variable('W_ci')
W_co = tf.get_variable('W_co')
b_f = tf.get_variable('b_f')
b_i = tf.get_variable('b_i')
b_c = tf.get_variable('b_c')
b_o = tf.get_variable('b_o')
i_t = tf.sigmoid(
tf.nn.bias_add(
tf.nn.conv2d(
inputs, W_xi, [1, 1, 1, 1], padding='SAME') +
(tf.nn.conv2d(
hidden_state, W_hi, [1, 1, 1, 1], padding='SAME')
if hidden_state is not None else tf.to_float(0)) +
(tf.multiply(
cell_state, W_ci, name='element_wise_multipy')
if cell_state is not None else tf.to_float(0)), b_i))
#i_t = tf.contrib.layers.layer_norm(i_t)
f_t = tf.sigmoid(
tf.nn.bias_add(
tf.nn.conv2d(
inputs, W_xf, [1, 1, 1, 1], padding='SAME') +
(tf.nn.conv2d(
hidden_state, W_hf, [1, 1, 1, 1], padding='SAME')
if hidden_state is not None else tf.to_float(0)) +
(tf.multiply(
cell_state, W_cf, name='element_wise_multipy_ft')
if cell_state is not None else tf.to_float(0)), b_f))
#f_t = tf.contrib.layers.layer_norm(f_t)
j = tf.nn.bias_add(
tf.nn.conv2d(inputs, W_xc, [1, 1, 1, 1], padding='SAME') +
(tf.nn.conv2d(
hidden_state, W_hc, [1, 1, 1, 1], padding='SAME')
if hidden_state is not None else tf.to_float(0)), b_c)
#j = tf.contrib.layers.layer_norm(j)
new_cell_state = (tf.multiply(f_t, cell_state, name='element_wise_multipy_ct1') if cell_state is not None else tf.to_float(0)) + \
tf.multiply(i_t, tf.tanh( j ), name='element_wise_multipy_ct2')
#new_cell_state = tf.contrib.layers.layer_norm(new_cell_state)
o_t = tf.sigmoid(
tf.nn.bias_add(
tf.nn.conv2d(
inputs, W_xo, [1, 1, 1, 1], padding='SAME') +
(tf.nn.conv2d(
hidden_state, W_ho, [1, 1, 1, 1], padding='SAME')
if hidden_state is not None else tf.to_float(0)) +
tf.multiply(new_cell_state,
W_co,
name='element_wise_multipy_ot'), b_o))
#o_t = tf.contrib.layers.layer_norm(o_t)
new_hidden_state = tf.multiply(o_t,
tf.tanh(new_cell_state),
name='element_wise_multipy_it')
if var_list is not None:
for var in [
W_xf, W_xi, W_xc, W_xo, W_hf, W_hi, W_hc, W_ho,
W_cf, W_ci, W_co, b_f, b_i, b_c, b_o
]:
if var not in var_list:
var_list.append(var)
return {
'hidden_state': new_hidden_state,
'cell_state': new_cell_state
}
def convLSTM_layer1(self, inputs, inputs_sequence_length, var_list=None):
with tf.variable_scope('vfeedbacknet_{}'.format(Model.model_name)):
with tf.variable_scope('convlstm1'):
with tf.variable_scope('rnn'):
with tf.variable_scope('conv_lstm_cell'):
with tf.variable_scope('convlstm', reuse=True):
kernel = tf.get_variable('kernel')
W_ci = tf.get_variable('W_ci')
W_cf = tf.get_variable('W_cf')
W_co = tf.get_variable('W_co')
bias = tf.get_variable('bias')
if var_list is not None and kernel not in var_list:
var_list.append(kernel)
if var_list is not None and W_ci not in var_list:
var_list.append(W_ci)
if var_list is not None and W_cf not in var_list:
var_list.append(W_cf)
if var_list is not None and W_co not in var_list:
var_list.append(W_co)
if var_list is not None and bias not in var_list:
var_list.append(bias)
inputs, state = tf.nn.dynamic_rnn(
self.convLSTMCell1,
tf.stack(inputs, axis=1),
dtype=tf.float32,
sequence_length=None,
)
inputs = tf.unstack(inputs, axis=1)
return inputs
def convLSTM_layer2(self, inputs, inputs_sequence_length, var_list=None):
with tf.variable_scope('vfeedbacknet_{}'.format(Model.model_name)):
with tf.variable_scope('convlstm2'):
with tf.variable_scope('rnn'):
with tf.variable_scope('conv_lstm_cell'):
with tf.variable_scope('convlstm', reuse=True):
kernel = tf.get_variable('kernel')
W_ci = tf.get_variable('W_ci')
W_cf = tf.get_variable('W_cf')
W_co = tf.get_variable('W_co')
bias = tf.get_variable('bias')
if var_list is not None and kernel not in var_list:
var_list.append(kernel)
if var_list is not None and W_ci not in var_list:
var_list.append(W_ci)
if var_list is not None and W_cf not in var_list:
var_list.append(W_cf)
if var_list is not None and W_co not in var_list:
var_list.append(W_co)
if var_list is not None and bias not in var_list:
var_list.append(bias)
inputs, state = tf.nn.dynamic_rnn(
self.convLSTMCell2,
tf.stack(inputs, axis=1),
dtype=tf.float32,
sequence_length=None,
)
inputs = tf.unstack(inputs, axis=1)
return inputs
def reshape_conv_layer(self, inputs, conv_num, var_list=None):
with tf.variable_scope('vfeedbacknet_{}'.format(Model.model_name)):
with tf.variable_scope('reshape_convs'):
with tf.variable_scope('conv{}'.format(conv_num), reuse=True):
kernel = tf.get_variable('kernel')
biases = tf.get_variable('biases')
inputs = tf.nn.conv2d(inputs,
kernel, [1, 1, 1, 1],
padding='SAME')
inputs = tf.nn.bias_add(inputs, biases)
inputs = tf.nn.relu(inputs)
inputs = tf.nn.max_pool(inputs,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='VALID')
if var_list is not None and kernel not in var_list:
var_list.append(kernel)
if var_list is not None and biases not in var_list:
var_list.append(biases)
return inputs
def conv_layer(self, inputs, var_list=None):
with tf.variable_scope('vfeedbacknet_{}'.format(Model.model_name)):
with tf.variable_scope('process_featurizer_output'):
with tf.variable_scope('conv1', reuse=True):
kernel = tf.get_variable('kernel')
biases = tf.get_variable('biases')
inputs = tf.nn.conv2d(inputs,
kernel, [1, 1, 1, 1],
padding='SAME')
inputs = tf.nn.bias_add(inputs, biases)
inputs = tf.nn.relu(inputs)
inputs = tf.nn.max_pool(inputs,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='VALID')
if var_list is not None and kernel not in var_list:
var_list.append(kernel)
if var_list is not None and biases not in var_list:
var_list.append(biases)
return inputs
def fc_layer(self, inputs, var_list=None):
with tf.variable_scope('vfeedbacknet_{}'.format(Model.model_name),
reuse=True):
with tf.variable_scope('fc'):
weights = tf.get_variable('weights')
biases = tf.get_variable('biases')
h, w, c, = inputs.shape[1:]
size = int(h) * int(w) * int(c)
inputs = tf.reshape(inputs, [-1, size])
inputs = tf.matmul(inputs, weights)
inputs = tf.nn.bias_add(inputs, biases)
if var_list is not None and weights not in var_list:
var_list.append(weights)
if var_list is not None and biases not in var_list:
var_list.append(biases)
return inputs
class Model:
'''
Basic imagenet vgg-16 (slightly modified)
'''
def __init__(self,
sess,
num_classes,
batch_size,
train_featurizer='FINE_TUNE',
train_main_model='FINE_TUNE',
train_fc='FINE_TUNE',
weights_filename=None,
is_training=True):
self.sess = sess
self.weights = np.load(
weights_filename) if weights_filename is not None else None
self.num_classes = num_classes
assert train_featurizer in [
'NO', 'FINE_TUNE', 'FROM_SCRATCH'
], 'train_featurizer must be either: NO, FINE_TUNE, or FROM_SCRATCH'
self.train_featurizer = train_featurizer if is_training else 'NO'
assert train_main_model in [
'NO', 'FINE_TUNE', 'FROM_SCRATCH'
], 'train_main_model must be either: NO, FINE_TUNE, or FROM_SCRATCH'
self.train_main_model = train_main_model if is_training else 'NO'
assert train_fc in [
'NO', 'FINE_TUNE', 'FROM_SCRATCH'
], 'train_fc must be either: NO, FINE_TUNE, or FROM_SCRATCH'
self.train_fc = train_fc if is_training else 'NO'
self.is_training = is_training
self.featurizer_variables = []
self.main_model_variables = []
self.fc_variables = []
self.vfeedbacknet_base = VFeedbackNetBase(sess,
num_classes,
is_training=is_training)
self._declare_variables()
def _declare_variables(self):
pass
# with tf.variable_scope('vfeedbacknet_model1'):
# with tf.variable_scope('convlstm1'):
# with tf.variable_scope('rnn'):
# with tf.variable_scope('conv_lstm_cell'):
# regularizer = None # tf.contrib.layers.l2_regularizer(scale=0.25)
# initializer = tf.contrib.layers.xavier_initializer()
# n = 512
# m = 4*n
# input_size = [7, 7, n]
# kernel2d_size = [3, 3]
# kernel_size = kernel2d_size + [2*n] + [m]
# with tf.variable_scope('convlstm'):
# kernel = tf.get_variable('kernel', kernel_size, initializer=initializer, regularizer=regularizer)
# W_ci = tf.get_variable('W_ci', input_size, initializer=initializer, regularizer=regularizer)
# W_cf = tf.get_variable('W_cf', input_size, initializer=initializer, regularizer=regularizer)
# W_co = tf.get_variable('W_co', input_size, initializer=initializer, regularizer=regularizer)
# bias = tf.get_variable('bias', [m], initializer=tf.zeros_initializer(), regularizer=regularizer)
# self.convLSTMCell1 = ConvLSTMCell([7, 7], 512, [3, 3])
# with tf.variable_scope('convlstm2'):
# with tf.variable_scope('rnn'):
# with tf.variable_scope('conv_lstm_cell'):
# regularizer = None # tf.contrib.layers.l2_regularizer(scale=0.25)
# initializer = tf.contrib.layers.xavier_initializer()
# n = 512
# m = 4*n
# input_size = [7, 7, n]
# kernel2d_size = [3, 3]
# kernel_size = kernel2d_size + [2*n] + [m]
# with tf.variable_scope('convlstm'):
# kernel = tf.get_variable('kernel', kernel_size, initializer=initializer, regularizer=regularizer)
# W_ci = tf.get_variable('W_ci', input_size, initializer=initializer, regularizer=regularizer)
# W_cf = tf.get_variable('W_cf', input_size, initializer=initializer, regularizer=regularizer)
# W_co = tf.get_variable('W_co', input_size, initializer=initializer, regularizer=regularizer)
# bias = tf.get_variable('bias', [m], initializer=tf.zeros_initializer(), regularizer=regularizer)
# self.convLSTMCell2 = ConvLSTMCell([7, 7], 512, [3, 3])
def get_variables(self):
return self.featurizer_variables + self.main_model_variables + self.fc_variables
def print_variables(self):
for var in self.get_variables():
print(var.name)
def initialize_variables(self):
logging.debug('--- begin variable initialization (vfeedbacknet) ---')
if self.train_featurizer == 'FROM_SCRATCH':
logging.debug('vgg16:FROM_SCRATCH; using random initialization')
for var in self.featurizer_variables:
self.sess.run(var.initializer)
else:
for var in self.featurizer_variables:
logging.debug('LOADING FROM WEIGHTS_FILE {}: {}'.format(
var.name, var.shape))
assert self.weights is not None, 'Need to specify weights file to load from for featurizer_variables'
self.sess.run(var.assign(self.weights[var.name]))
if self.train_main_model == 'FROM_SCRATCH':
logging.debug(
'feedback: FROM_SCRATCH; using random initialization')
for var in self.main_model_variables:
self.sess.run(var.initializer)
else:
for var in self.main_model_variables:
logging.debug(' LOADING FROM WEIGHTS_FILE {}: {}'.format(
var.name, var.shape))
assert self.weights is not None, 'Need to specify weights file to load from for main_model_variables'
self.sess.run(var.assign(self.weights[var.name]))
if self.train_fc == 'FROM_SCRATCH':
logging.debug('fc: FROM_SCRATCH; using random initialization')
for var in self.fc_variables:
self.sess.run(var.initializer)
else:
for var in self.fc_variables:
logging.debug('LOADING FROM WEIGHTS_FILE {}: {}'.format(
var.name, var.shape))
assert self.weights is not None, 'Need to specify weights file to load from for fc_variables'
self.sess.run(var.assign(self.weights[var.name]))
logging.debug('--- end variable initialization (vfeedbacknet) ---')
def export_variables(self, export_filename):
VFeedbackNetBase.export_variables(self.sess, self.get_variables(),
export_filename)
def __call__(self, inputs, inputs_sequence_length):
ModelLogger.log('raw_input', inputs)
inputs = self.vfeedbacknet_base.split_video(inputs)
ModelLogger.log('input', inputs)
## featurizer ##
inputs = [
self.vfeedbacknet_base.vgg16_layer1(
inp, var_list=self.featurizer_variables) for inp in inputs
]
ModelLogger.log('vgg-layer1', inputs)
inputs = [
self.vfeedbacknet_base.vgg16_layer2(
inp, var_list=self.featurizer_variables) for inp in inputs
]
ModelLogger.log('vgg-layer2', inputs)
## main model ##
inputs = [
self.vfeedbacknet_base.vgg16_layer3(
inp, var_list=self.featurizer_variables) for inp in inputs
]
ModelLogger.log('vgg-layer3', inputs)
inputs = [
self.vfeedbacknet_base.vgg16_layer4(
inp, var_list=self.featurizer_variables) for inp in inputs
]
ModelLogger.log('vgg-layer4', inputs)
inputs = [
self.vfeedbacknet_base.vgg16_layer5(
inp, var_list=self.featurizer_variables) for inp in inputs
]
ModelLogger.log('vgg-layer5', inputs)
## ave_pool and fc ##
# inputs = [ self.vfeedbacknet_base.ave_pool(inp) for inp in inputs ]
# ModelLogger.log('ave_pool', inputs)
inputs = [
self.vfeedbacknet_base.vgg16_fc_layer1(inp,
var_list=self.fc_variables)
for inp in inputs
]
ModelLogger.log('fc6', inputs)
inputs = [
self.vfeedbacknet_base.vgg16_fc_layer2(inp,
var_list=self.fc_variables)
for inp in inputs
]
ModelLogger.log('fc7', inputs)
inputs = [
self.vfeedbacknet_base.vgg16_fc_layer3(inp,
var_list=self.fc_variables)
for inp in inputs
]
ModelLogger.log('fc8', inputs)
# inputs = [ self.vfeedbacknet_base.fc_layer(inp, var_list=self.fc_variables) for inp in inputs ]
# ModelLogger.log('fc', inputs)
logits = tf.stack(inputs, axis=1)
logits = tf.expand_dims(logits, axis=1)
ModelLogger.log('logits', logits)
return logits