def __init__(self, batch_size, is_training=True, is_fine_tune=False):
""" Init for Pipeline """
self.batch_size = batch_size
self.part_batch = (self.batch_size / 2)
self.is_training = is_training
### Create a new network model
self.net_scope = "netmodel_prime"
self.net_model = NetModel(self.net_scope,
batch_size=self.batch_size,
is_training=is_training,
is_fine_tune=is_fine_tune)
### Get suitable params
self.net_params = utils.get_network_params(self.net_scope)
self.resnet_params = utils.get_net_train_params("resnet_v2_50")
### Define all trainable params
self.trainable_params = self.net_params + self.resnet_params
self.var_list = tf.global_variables()
### Weight save & restore
self.bestwt_saver = tf.train.Saver(self.var_list, max_to_keep=5)
self.iterwt_saver = tf.train.Saver(self.var_list, max_to_keep=5)
### Scope/Name
self.scope = "composer"
self.name = "model_composer"
### Get critical network nodes
self.nodes_net = self.net_model.get_network_nodes()
## Nodes
self.in_gt_nodes = self.nodes_net['inputs_and_gt']
self.cnn_nodes = self.nodes_net['cnn_layer']
self.smpl_nodes = self.nodes_net['smpl_layer']
self.cam_mesh_nodes = self.nodes_net['cam_mesh_module']
self.ren_nodes = self.nodes_net['renderer_layer']
help="File to save weights")
parser.add_argument('--inference',
type=str,
default=None,
help="File to load weights")
parser.add_argument('--resume',
type=str,
default=None,
help="Weights file to resume training")
args = parser.parse_args()
# Initialize seed before any use
np.random.seed(args.rng_seed)
# Get network parameters
nw_params = get_network_params(args.dataset, args.size, args.batch_size)
metric_names = nw_params['metric_names']
en_top5 = True
num_resnet_mods = nw_params['num_resnet_mods']
args.iter_interval = nw_params['iter_interval']
learning_schedule = nw_params['learning_schedule']
en_bottleneck = nw_params['en_bottleneck']
ax.Y.length = nw_params['num_classes']
# Set batch size
ax.N.length = args.batch_size
# Create training and validation set objects
train_set, valid_set = make_aeon_loaders(args.data_dir,
args.batch_size,
args.num_iterations,