net = tf_utils.fully_connected(y_tilde,
1024,
bn=True,
is_training=is_training,
scope='fc1',
bn_decay=bn_decay)
net = tf_utils.fully_connected(net,
1024,
bn=True,
is_training=is_training,
scope='fc2',
bn_decay=bn_decay)
net = tf_utils.fully_connected(net,
2048 * 3,
activation_fn=None,
scope='fc3')
net = tf.reshape(net, (batch_size, 2048, 3))
return net
if __name__ == '__main__':
tf.enable_eager_execution()
TRAIN_DATASET = part_dataset.PartDataset(
root='/data/dataset/shapenetcore_partanno_segmentation_benchmark_v0',
npoints=2048,
classification=False,
class_choice=None,
split='trainval')
print('=============')
print(input_fn(TRAIN_DATASET, 2, 8, repeat=True, prefetch_size=6))
os.system('cp %s %s' % (MODEL_FILE, LOG_DIR)) # bkp of model def
os.system('cp train.py %s' % (LOG_DIR)) # bkp of train procedure
LOG_FOUT = open(os.path.join(LOG_DIR, 'log_train.txt'), 'w')
LOG_FOUT.write(str(FLAGS)+'\n')
BN_INIT_DECAY = 0.5
BN_DECAY_DECAY_RATE = 0.5
BN_DECAY_DECAY_STEP = float(DECAY_STEP)
BN_DECAY_CLIP = 0.99
HOSTNAME = socket.gethostname()
# Shapenet official train/test split
#DATA_PATH = os.path.join(BASE_DIR, 'data/shapenetcore_partanno_segmentation_benchmark_v0')
DATA_PATH = '/home/sohee/code/dataset/shapenet/shapenetcore_partanno_segmentation_benchmark_v0'
TRAIN_DATASET = part_dataset.PartDataset(root=DATA_PATH, npoints=NUM_POINT, classification=False, class_choice=FLAGS.category, split='trainval')
TEST_DATASET = part_dataset.PartDataset(root=DATA_PATH, npoints=NUM_POINT, classification=False, class_choice=FLAGS.category, split='test')
def log_string(out_str):
LOG_FOUT.write(out_str+'\n')
LOG_FOUT.flush()
print(out_str)
def get_learning_rate(batch):
learning_rate = tf.train.exponential_decay(
BASE_LEARNING_RATE, # Base learning rate.
batch * BATCH_SIZE, # Current index into the dataset.
DECAY_STEP, # Decay step.
DECAY_RATE, # Decay rate.
staircase=True)
learing_rate = tf.maximum(learning_rate, 0.00001) # CLIP THE LEARNING RATE!
type=int,
default=1,
help=
'Number of groups of generated points -- used for hierarchical FC decoder. [default: 1]'
)
FLAGS = parser.parse_args()
MODEL_PATH = FLAGS.model_path
GPU_INDEX = FLAGS.gpu
NUM_POINT = FLAGS.num_point
MODEL = importlib.import_module(FLAGS.model) # import network module
DATA_PATH = os.path.join(
BASE_DIR, 'data/shapenetcore_partanno_segmentation_benchmark_v0')
TEST_DATASET = part_dataset.PartDataset(root=DATA_PATH,
npoints=NUM_POINT,
classification=False,
class_choice=FLAGS.category,
split='test',
normalize=True)
print(len(TEST_DATASET))
def get_model(batch_size, num_point):
with tf.Graph().as_default():
with tf.device('/gpu:' + str(GPU_INDEX)):
pointclouds_pl, labels_pl = MODEL.placeholder_inputs(
batch_size, num_point)
is_training_pl = tf.placeholder(tf.bool, shape=())
pred, end_points = MODEL.get_model(pointclouds_pl, is_training_pl)
loss = MODEL.get_loss(pred, labels_pl, end_points)
saver = tf.train.Saver()
# Create a session
