import tensorflow as tf
import math
from models.slim.datasets import dataset_factory
from models.slim.preprocessing import preprocessing_factory
import squeezenet
import arg_parsing
slim = tf.contrib.slim
args = arg_parsing.parse_args(training=False)
tf.logging.set_verbosity(tf.logging.INFO)
with tf.Graph().as_default() as g:
with g.device(args.eval_device):
dataset = dataset_factory.get_dataset('cifar10', 'test', args.data_dir)
tf_global_step = slim.get_or_create_global_step()
provider = slim.dataset_data_provider.DatasetDataProvider(
dataset,
shuffle=False,
common_queue_capacity=2 * args.batch_size,
common_queue_min=args.batch_size)
[image, label] = provider.get(['image', 'label'])
image_preprocessing_fn = preprocessing_factory.get_preprocessing(
'cifarnet', is_training=False)
image = image_preprocessing_fn(image, 32, 32)
import vgg_cifar10
slim = tf.contrib.slim
EXPERIMENT_NAME = 'VGG14_long_train'
BATCH_SIZE = 512
CHECKPOINT_DIR = '/mnt/data1/vgg_results/' + EXPERIMENT_NAME + '/train'
EVAL_DIR = CHECKPOINT_DIR[:-5] + 'test'
DATA_DIR = '/mnt/data1/cifar'
EVAL_DEVICE = '/gpu:2'
tf.logging.set_verbosity(tf.logging.INFO)
with tf.Graph().as_default() as g:
with g.device(EVAL_DEVICE):
dataset = dataset_factory.get_dataset('cifar10', 'test', DATA_DIR)
tf_global_step = slim.get_or_create_global_step()
provider = slim.dataset_data_provider.DatasetDataProvider(
dataset,
shuffle=False,
common_queue_capacity=2 * BATCH_SIZE,
common_queue_min=BATCH_SIZE)
[image, label] = provider.get(['image', 'label'])
image_preprocessing_fn = preprocessing_factory.get_preprocessing(
'cifarnet', is_training=False)
image = image_preprocessing_fn(image, 32, 32)
DATA_DIR = '/mnt/data1/cifar'
TRAIN_DIR = '/mnt/data1/squeezenet_results/LR_01_95_DR_BN/train'
BATCH_SIZE = 256
INIT_LEARNING_RATE = 0.01
LR_DECAY = 0.95
NUM_EPOCHS_PER_DECAY = 2
MAX_STEPS = 8000
NUM_CLONES = 3
tf.logging.set_verbosity(tf.logging.INFO)
deploy_config = model_deploy.DeploymentConfig(num_clones=NUM_CLONES)
with tf.device(deploy_config.variables_device()):
global_step = slim.create_global_step()
dataset = dataset_factory.get_dataset('cifar10', 'train', '/mnt/data1/cifar')
network_fn = squeezenet.inference
image_preprocessing_fn = preprocessing_factory.get_preprocessing(
'cifarnet', is_training=True)
with tf.device(deploy_config.inputs_device()):
with tf.name_scope('inputs'):
provider = slim.dataset_data_provider.DatasetDataProvider(
dataset,
num_readers=7,
common_queue_capacity=20 * BATCH_SIZE,
common_queue_min=10 * BATCH_SIZE)
[image, label] = provider.get(['image', 'label'])