def get_model(point_cloud, is_training, num_class, bn_decay=None, feature=None):
""" Semantic segmentation PointNet, input is B x N x 3, output B x num_class """
end_points = {}
l0_xyz = point_cloud
l0_points = feature
end_points['l0_xyz'] = l0_xyz
# c0
c0_l0_xyz, c0_l0_points, c0_l0_indices = DFCN_res_module(l0_xyz, l0_points, radius=2, out_channel=64, is_training=is_training, bn_decay=bn_decay, scope='layer0_c0', merge='concat')
l1_xyz, l1_points, l1_indices = pointnet_sa_module(c0_l0_xyz, c0_l0_points, npoint=1024, radius=2, nsample=32, mlp=[64,128], mlp2=None, group_all=False, is_training=is_training, bn_decay=bn_decay, scope='layer1')
# c1
c0_l1_xyz, c0_l1_points, c0_l1_indices = DFCN_res_module(l1_xyz, l1_points, radius=5, out_channel=128, is_training=is_training, bn_decay=bn_decay, scope='layer1_c0')
l2_xyz, l2_points, l2_indices = pointnet_sa_module(c0_l1_xyz, c0_l1_points, npoint=256, radius=4, nsample=32, mlp=[128,256], mlp2=None, group_all=False, is_training=is_training, bn_decay=bn_decay, scope='layer2')
# c2
c0_l2_xyz, c0_l2_points, c0_l2_indices = DFCN_res_module(l2_xyz, l2_points, radius=10, out_channel=256, is_training=is_training, bn_decay=bn_decay, scope='layer2_c0')
c1_l2_xyz, c1_l2_points, c1_l2_indices = DFCN_res_module(c0_l2_xyz, c0_l2_points, radius=10, out_channel=512, is_training=is_training, bn_decay=bn_decay, scope='layer2_c1', same_dim=True)
l2_cat_points = tf.concat([c0_l2_points, c1_l2_points], axis=-1)
fc_l2_points = tf_util.conv1d(l2_cat_points, 512, 1, padding='VALID', bn=True, is_training=is_training, scope='conv_2_fc', bn_decay=bn_decay)
# c3
l3_xyz, l3_points, l3_indices = pointnet_sa_module(c1_l2_xyz, fc_l2_points, npoint=64, radius=8, nsample=32, mlp=[512,512], mlp2=None, group_all=False, is_training=is_training, bn_decay=bn_decay, scope='layer3')
l2_points = pointnet_fp_module(l2_xyz, l3_xyz, l2_points, l3_points, [512,512], is_training, bn_decay, scope='fa_layer2')
_, l2_points_1, _ = DFCN_module(l2_xyz, l2_points, radius=10, out_channel=512, is_training=is_training, bn_decay=bn_decay, scope='fa_layer2_c0')
_, l2_points_2, _ = DFCN_module(l2_xyz, l2_points, radius=10, out_channel=512, is_training=is_training, bn_decay=bn_decay, scope='fa_layer2_c1')
_, l2_points_3, _ = DFCN_module(l2_xyz, l2_points, radius=10, out_channel=512, is_training=is_training, bn_decay=bn_decay, scope='fa_layer2_c2')
l2_points = tf.concat([l2_points_1, l2_points_2, l2_points_3], axis=-1)
l2_points = tf_util.conv1d(l2_points, 512, 1, padding='VALID', bn=True, is_training=is_training, scope='fa_2_fc', bn_decay=bn_decay)
l1_points = pointnet_fp_module(l1_xyz, l2_xyz, l1_points, l2_points, [256,256], is_training, bn_decay, scope='fa_layer3')
_, l1_points_1, _ = DFCN_module(l1_xyz, l1_points, radius=5, out_channel=256, is_training=is_training, bn_decay=bn_decay, scope='fa_layer3_c0')
_, l1_points_2, _ = DFCN_module(l1_xyz, l1_points_1, radius=5, out_channel=256, is_training=is_training, bn_decay=bn_decay, scope='fa_layer3_c1')
l1_points = tf.concat([l1_points_1, l1_points_2], axis=-1)
l1_points = tf_util.conv1d(l1_points, 256, 1, padding='VALID', bn=True, is_training=is_training, scope='fa_1_fc', bn_decay=bn_decay)
l0_points = pointnet_fp_module(l0_xyz, l1_xyz, l0_points, l1_points, [128,128,128], is_training, bn_decay, scope='fa_layer4')
_, l0_points, _ = DFCN_module(l0_xyz, l0_points, radius=2, out_channel=128, is_training=is_training, bn_decay=bn_decay, scope='fa_layer4_c0')
# FC layers
net = tf_util.conv1d(l0_points, 128, 1, padding='VALID', bn=True, is_training=is_training, scope='fc1', bn_decay=bn_decay)
net = tf_util.dropout(net, keep_prob=0.5, is_training=is_training, scope='dp1')
net = tf_util.conv1d(net, num_class, 1, padding='VALID', activation_fn=None, scope='fc2')
return net, end_points
def pointSIFT_res_module(xyz, points, radius, out_channel, is_training, bn_decay, scope='point_sift', bn=True, use_xyz=True, same_dim=False, merge='add'):
data_format = 'NHWC'
with tf.compat.v1.variable_scope(scope) as sc:
# conv1
_, new_points, idx, _ = pointSIFT_group(radius, xyz, points, use_xyz=use_xyz)
for i in range(3):
new_points = tf_util.conv2d(new_points, out_channel, [1, 2],
padding='VALID', stride=[1, 2],
bn=bn, is_training=is_training,
scope='c0_conv%d' % (i), bn_decay=bn_decay,
data_format=data_format)
new_points = tf.squeeze(new_points, [2])
# conv2
_, new_points, idx, _ = pointSIFT_group_with_idx(xyz, idx=idx, points=new_points, use_xyz=use_xyz)
for i in range(3):
if i == 2:
act = None
else:
act = tf.nn.relu
new_points = tf_util.conv2d(new_points, out_channel, [1, 2],
padding='VALID', stride=[1, 2],
bn=bn, is_training=is_training,
scope='c1_conv%d' % (i), bn_decay=bn_decay,
activation_fn=act,
data_format=data_format)
new_points = tf.squeeze(new_points, [2])
# residual part..
if points is not None:
if same_dim is True:
points = tf_util.conv1d(points, out_channel, 1, padding='VALID', bn=bn, is_training=is_training, scope='merge_channel_fc', bn_decay=bn_decay)
if merge == 'add':
new_points = new_points + points
elif merge == 'concat':
new_points = tf.concat([new_points, points], axis=-1)
else:
print("ways not found!!!")
new_points = tf.nn.relu(new_points)
return xyz, new_points, idx