x = tf.placeholder(tf.float32, [None, None, 3])
y = tf.placeholder(tf.float32, [1001])
x_preporcess = tf.reshape(preprocess_image(x, 224, 224, TRAIN),
[1, 224, 224, 3])
y_ = tf.reshape(y, [1, 1001])
train_flag = tf.placeholder(tf.bool)
with tf.device('/cpu:0'):
q = tf.FIFOQueue(BATCH_SIZE * 3, [tf.float32, tf.float32],
shapes=[[224, 224, 3], [1001]])
enqueue_op = q.enqueue_many([x_preporcess, y_])
x_b, y_b = q.dequeue_many(BATCH_SIZE)
out, restore_var = mobilenet_v2_224(x_b, train_flag, None, None, None)
with tf.name_scope('cross_entropy'):
cross_entropy = tf.reduce_mean(
tf.losses.softmax_cross_entropy(y_b,
out,
label_smoothing=0.1,
weights=0.4))
with tf.name_scope('total_loss'):
total_loss = cross_entropy + tf.add_n(
tf.losses.get_regularization_losses())
with tf.name_scope('optimizer'):
train = tf.train.RMSPropOptimizer(0.03, 0.9, 0.9)
[1, 224, 224, 3])
y_ = tf.reshape(y, [1, 1001])
train_flag = tf.placeholder(tf.bool)
with tf.device('/cpu:0'):
q = tf.FIFOQueue(BATCH_SIZE * 3, [tf.float32, tf.float32],
shapes=[[224, 224, 3], [1001]])
enqueue_op = q.enqueue_many([x_preporcess, y_])
x_b, y_b = q.dequeue_many(BATCH_SIZE)
statistic_builder = cl.statistic_flow_builder()
score_builder = cl.score_update_flow_builder()
gradient_builder = cl.average_gradients_flow_builder()
out, restore_var = mobilenet_v2_224(x_b, train_flag, statistic_builder,
score_builder, gradient_builder)
with tf.name_scope('cross_entropy'):
cross_entropy = tf.reduce_mean(
tf.losses.softmax_cross_entropy(y_b, out, label_smoothing=0.1))
with tf.name_scope('prune_statistic'):
gradient_builder.set_y(cross_entropy)
gradient_builder.build_flow_on_conv()
score_builder.set_gard_var_wise_list(
gradient_builder.post_process_and_get_tensor_list())
statistic_update_op_list = []
statistic_update_op_list += statistic_builder.get_update_op()
statistic_update_op_list += gradient_builder.get_update_op_list()
statistic_reset_op_list = []