def __init__(self, num_classes=60):
super().__init__()
graph = Graph()
self.A = tf.Variable(graph.A,
dtype=tf.float32,
trainable=False,
name='adjacency_matrix')
self.data_bn = tf.keras.layers.BatchNormalization(axis=1)
self.STGCN_layers = []
self.STGCN_layers.append(SpatioTemporalGraphConv(64, residual=False))
self.STGCN_layers.append(SpatioTemporalGraphConv(64))
self.STGCN_layers.append(SpatioTemporalGraphConv(64))
self.STGCN_layers.append(SpatioTemporalGraphConv(64))
self.STGCN_layers.append(SpatioTemporalGraphConv(128, stride=2))
self.STGCN_layers.append(SpatioTemporalGraphConv(128))
self.STGCN_layers.append(SpatioTemporalGraphConv(128))
self.STGCN_layers.append(SpatioTemporalGraphConv(256, stride=2))
self.STGCN_layers.append(SpatioTemporalGraphConv(256))
self.STGCN_layers.append(SpatioTemporalGraphConv(256))
self.pool = tf.keras.layers.GlobalAveragePooling2D(
data_format='channels_first')
self.logits = tf.keras.layers.Conv2D(num_classes,
kernel_size=1,
padding='same',
kernel_initializer=INITIALIZER,
kernel_regularizer=REGULARIZER,
data_format='channels_first')
def __init__(self, num_classes=60):
super().__init__()
graph = Graph()
A = graph.A.astype(np.float32)
self.data_bn = tf.keras.layers.BatchNormalization(axis=1)
self.STGCN_layers = []
self.STGCN_layers.append(STGCN(64, A, 300, residual=False))
self.STGCN_layers.append(STGCN(64, A, 300))
self.STGCN_layers.append(STGCN(64, A, 300))
self.STGCN_layers.append(STGCN(64, A, 300))
self.STGCN_layers.append(STGCN(128, A, 300, stride=2, downsample=True))
self.STGCN_layers.append(STGCN(128, A, 150))
self.STGCN_layers.append(STGCN(128, A, 150))
self.STGCN_layers.append(STGCN(256, A, 150, stride=2, downsample=True))
self.STGCN_layers.append(STGCN(256, A, 75))
self.STGCN_layers.append(STGCN(256, A, 75))
self.pool = tf.keras.layers.GlobalAveragePooling2D(
data_format='channels_first')
self.logits = tf.keras.layers.Conv2D(num_classes,
kernel_size=1,
padding='same',
kernel_initializer=INITIALIZER,
data_format='channels_first',
kernel_regularizer=REGULARIZER)
def __init__(self, num_classes=60):
super().__init__()
self.graph = Graph()
A = self.graph.A.astype(np.float32)
self.data_bn = tf.keras.layers.BatchNormalization(axis=1)
self.GTC_layers = []
self.GTC_layers.append(GraphTemporalConv(64, A, down=False, residual=False))
self.GTC_layers.append(GraphTemporalConv(64, A))
self.GTC_layers.append(GraphTemporalConv(64, A))
self.GTC_layers.append(GraphTemporalConv(64, A))
self.GTC_layers.append(GraphTemporalConv(128, A, down=True, stride=2))
self.GTC_layers.append(GraphTemporalConv(128, A))
self.GTC_layers.append(GraphTemporalConv(128, A))
self.GTC_layers.append(GraphTemporalConv(256, A, down=True, stride=2))
self.GTC_layers.append(GraphTemporalConv(256, A))
self.GTC_layers.append(GraphTemporalConv(256, A))
self.fc = tf.keras.layers.Dense(num_classes,
kernel_initializer=INITIALIZER,
kernel_regularizer=REGULARIZER)