def create_pcn_encoder(inputs, args):
with tf.variable_scope('encoder_0', reuse=tf.AUTO_REUSE):
features = mlp_conv(inputs, [128, 256], args.phase)
features_global = tf.reduce_max(features, axis=1, keep_dims=True, name='maxpool_0')
features = tf.concat([features, tf.tile(features_global, [1, tf.shape(inputs)[1], 1])], axis=2)
with tf.variable_scope('encoder_1', reuse=tf.AUTO_REUSE):
features = mlp_conv(features, [512, args.code_nfts], args.phase)
features = tf.reduce_max(features, axis=1, name='maxpool_1')
return features
def create_decoder(self, features):
with tf.variable_scope('decoder', reuse=tf.AUTO_REUSE):
coarse = mlp(features, [1024, 1024, self.num_coarse * 3],
self.args.phase)
coarse = tf.reshape(coarse, [-1, self.num_coarse, 3])
with tf.variable_scope('folding', reuse=tf.AUTO_REUSE):
grid = tf.meshgrid(tf.linspace(-0.05, 0.05, self.grid_size),
tf.linspace(-0.05, 0.05, self.grid_size))
grid = tf.expand_dims(tf.reshape(tf.stack(grid, axis=2), [-1, 2]),
0)
grid_feat = tf.tile(grid,
[tf.shape(features)[0], self.num_coarse, 1])
point_feat = tf.tile(tf.expand_dims(coarse, 2),
[1, 1, self.grid_size**2, 1])
point_feat = tf.reshape(point_feat, [-1, self.num_fine, 3])
global_feat = tf.tile(tf.expand_dims(features, 1),
[1, self.num_fine, 1])
feat = tf.concat([grid_feat, point_feat, global_feat], axis=2)
center = tf.tile(tf.expand_dims(coarse, 2),
[1, 1, self.grid_size**2, 1])
center = tf.reshape(center, [-1, self.num_fine, 3])
fine = mlp_conv(feat, [512, 512, 3], self.args.phase)
fine = fine + center
print(coarse.shape, fine.shape)
return coarse, fine
def create_decoder(self, features):
with tf.variable_scope('decoder', reuse=tf.AUTO_REUSE):
x = tf.linspace(-self.grid_scale, self.grid_scale, self.grid_size)
y = tf.linspace(-self.grid_scale, self.grid_scale, self.grid_size)
grid = tf.meshgrid(x, y)
grid = tf.reshape(tf.stack(grid, axis=2), [-1, 2])
grid = tf.tile(tf.expand_dims(grid, 0),
[tf.shape(features)[0], 1, 1])
features = tf.tile(tf.expand_dims(features, 1),
[1, self.num_output_points, 1])
with tf.variable_scope('folding_1'):
fold1 = mlp_conv(tf.concat([features, grid], axis=2),
[512, 512, 3], self.args.phase)
with tf.variable_scope('folding_2'):
fold2 = mlp_conv(tf.concat([features, fold1], axis=2),
[512, 512, 3], self.args.phase)
print(fold1.shape, fold2.shape)
return fold1, fold2
def create_pointnet_encoder(inputs, args):
with tf.variable_scope('encoder_0', reuse=tf.AUTO_REUSE):
features = mlp_conv(inputs, [64, 128, args.code_nfts], args.phase, bn=True)
code = tf.reduce_max(features, axis=1, name='maxpool_0')
with tf.variable_scope('encoder_1', reuse=tf.AUTO_REUSE):
code = mlp(code, [args.code_nfts,], args.phase)
with tf.variable_scope('bn', reuse=tf.AUTO_REUSE):
code = tf.layers.batch_normalization(code, training=args.phase)
code = tf.nn.relu(code, 'relu')
return code
def create_encoder(self, inputs):
with tf.variable_scope('encoder_0', reuse=tf.AUTO_REUSE):
features = mlp_conv(inputs, [128, 256])
features_global = tf.reduce_max(features,
axis=1,
keep_dims=True,
name='maxpool_0')
print('features_global', features_global)
features = tf.concat([
features,
tf.tile(features_global, [1, tf.shape(inputs)[1], 1])
],
axis=2)
print('features', features)
with tf.variable_scope('encoder_1', reuse=tf.AUTO_REUSE):
features = mlp_conv(features, [512, 1024])
print('features', features)
features = tf.reduce_max(features, axis=1, name='maxpool_1')
print('features', features)
return features
def create_level(self, level, input_channels, output_channels, inputs, bn):
with tf.variable_scope('level_%d' % (level), reuse=tf.AUTO_REUSE):
features = mlp_conv(inputs, [
input_channels,
int(input_channels / 2),
int(input_channels / 4),
int(input_channels / 8),
output_channels * int(self.args.tarch[level])
], self.args.phase, bn)
features = tf.reshape(features,
[tf.shape(features)[0], -1, output_channels])
return features