def __init__(self, points, features, is_training, setting):
PointCNN.__init__(self, points, features, is_training, setting)
self.point_features = self.layer_fts[1]
fc_mean = tf.reduce_mean(self.fc_layers[-1], axis=1, keepdims=True, name='fc_mean')
self.fc_layers[-1] = tf.cond(is_training, lambda: self.fc_layers[-1], lambda: fc_mean)
self.logits = pf.dense(self.fc_layers[-1], setting.num_class, 'logits',
is_training, with_bn=False, activation=None)
def __init__(self, points, features, is_training, setting):
PointCNN.__init__(self, points, features, is_training, setting)
self.logits = pf.dense(self.fc_layers[-1],
setting.num_class,
'logits',
is_training,
with_bn=False,
activation=None)
def __init__(self, points, features, is_training, setting):
print("points", points.get_shape())
PointCNN.__init__(self, points, features, is_training, setting)
per_loc_bin_num = int(setting.LOC_SCOPE / setting.LOC_BIN_SIZE) * 2
loc_y_bin_num = int(setting.LOC_Y_SCOPE / setting.LOC_Y_BIN_SIZE) * 2
reg_channel = per_loc_bin_num * 4 + setting.NUM_HEAD_BIN * 2 + 3
reg_channel += (1 if not setting.LOC_Y_BY_BIN else loc_y_bin_num * 2)
channel_in = (self.layer_fts[-1]
if not setting.IS_FC_INPUT else self.fc_layers[-1])
#Normalizing dimension for 1D
channel_in = tf.reduce_mean(channel_in,
axis=1,
keep_dims=True,
name='fc_mean')
print("channel_in", channel_in.shape)
pre_channel = channel_in
REG_FC = [256, 256]
for k in range(0, len(setting.REG_FC)):
pre_channel = pf.conv1d(pre_channel,
setting.REG_FC[k],
"CLS_LAYERS_" + str(k),
is_training,
with_bn=True)
#pre_channel = pf.dense(pre_channel, reg_channel, "CLS_LAYERS_FC", is_training, with_bn=False, activation=None)
pre_channel = pf.conv1d(pre_channel,
reg_channel,
"CLS_LAYERS_FC",
is_training,
activation=None)
if setting.DP_RATIO >= 0:
pre_channel = tf.layers.dropout(pre_channel,
setting.DP_RATIO,
training=is_training,
name='fc_reg_drop')
#print("pre_channel", reg_channel, pre_channel.get_shape())
self.logits = tf.transpose(pre_channel, perm=(0, 2, 1))
self.logits = tf.squeeze(self.logits, axis=-1)
def __init__(self, points, features, is_training, setting):
PointCNN.__init__(self, points, features, is_training, setting)
with tf.variable_scope(setting.network_name):
batch_size = points.get_shape()[0].value
fc_flatten = tf.reshape(self.fc_layers[-1], [batch_size, -1])
fc_flatten = tf.concat([fc_flatten, features[:, 0, :]], axis=1)
fc1 = pf.dense(fc_flatten, 512, 'extra_fc_1', is_training)
# fc1_drop = tf.layers.dropout(fc1, 0.0, training=is_training, name='extra_fc_1_drop')
# self.fc_layers.append(fc1_drop)
fc2 = pf.dense(fc1, 256, 'extra_fc_2', is_training)
self.output = pf.dense(fc2,
3 + NUM_HEADING_BIN * 2 +
NUM_SIZE_CLUSTER * 4,
'output',
is_training,
with_bn=False,
activation=None)
def __init__(self, points, features, is_training, setting):
PointCNN.__init__(self, points, features, is_training, setting)
fc_mean = tf.reduce_mean(self.fc_layers[-1],
axis=1,
keep_dims=True,
name='fc_mean')
self.feature_list = tf.reshape(self.feature_list, [128, 61440])
self.feature_list_A = self.feature_list[0:64]
self.feature_list_B = self.feature_list[64:128]
self.fc_layers[-1] = tf.cond(is_training, lambda: self.fc_layers[-1],
lambda: fc_mean) #最后一层连接
self.logits = pf.dense(self.fc_layers[-1],
setting.num_class,
'logits',
is_training,
with_bn=False,
activation=None)
def __init__(self, points, features, num_class, is_training, setting):
PointCNN.__init__(self, points, features, num_class, is_training,
setting, 'segmentation')
'./mnist/train_files.txt', './mnist/test_files.txt')
nd_iter = mx.io.NDArrayIter(data={'data': data_train},
label={'softmax_label': label_train},
batch_size=setting.batch_size)
num_train = data_train.shape[0]
point_num = data_train.shape[1]
batch_num_per_epoch = int(math.ceil(num_train / setting.batch_size))
batch_num = batch_num_per_epoch * setting.num_epochs
batch_size_train = setting.batch_size
ctx = [mx.gpu(0)]
net = PointCNN(setting,
'classification',
with_feature=False,
prefix="PointCNN_")
net.hybridize()
sym_max_points = point_num
var = mx.sym.var('data',
shape=(batch_size_train // len(ctx), sym_max_points, 3))
probs = net(var)
probs_shape = get_shape(probs)
label_var = mx.sym.var('softmax_label',
shape=(batch_size_train // len(ctx), probs_shape[1]))
loss = get_loss_sym(probs, label_var)