def __init__(self,
depth,
filter_size,
hidden_filter_size,
strides,
padding,
bias=LSTMBiasInit,
initializer=block_util.RsqrtInitializer(dims=(0, 1, 2)),
use_moving_average=False,
name=None):
super(Conv2DLSTM, self).__init__([None, None, depth], name)
self._iter = 0
with self._BlockScope():
self._input_conv = blocks_std.Conv2D(4 * depth,
filter_size,
strides,
padding,
bias=None,
act=None,
initializer=initializer,
name='input_conv2d')
self._hidden_conv = blocks_std.Conv2D(4 * depth,
hidden_filter_size, [1, 1],
'SAME',
bias=None,
act=None,
initializer=initializer,
name='hidden_conv2d')
if bias is not None:
self._bias = blocks_std.BiasAdd(bias, name='biases')
else:
self._bias = blocks_std.PassThrough()
def testConv2DReflect(self):
np.random.seed(768798)
x_shape = [4, 10, 12, 6]
f_shape = [3, 4, 6, 5]
strides = [1, 2, 2, 1]
padding = 'REFLECT'
conv = blocks_std.Conv2D(depth=f_shape[-1],
filter_size=f_shape[0:2],
strides=strides[1:3],
padding=padding,
act=None,
bias=None)
x_value = np.random.normal(size=x_shape)
x = tf.convert_to_tensor(x_value, dtype=tf.float32)
y = conv(x)
with self.test_session():
tf.global_variables_initializer().run()
f_value = conv._kernel.eval()
y_value = y.eval()
y_expected = _NumpyConv2D(x_value, f_value,
strides=strides, padding=padding)
self.assertAllClose(y_expected, y_value)
def __init__(self,
depth,
filter_size,
hidden_filter_size,
strides,
padding,
bias=blocks_lstm.LSTMBiasInit,
initializer=block_util.RsqrtInitializer(dims=(0, 1, 2)),
name=None):
super(RasterScanConv2DLSTM, self).__init__([None, None, depth], name)
with self._BlockScope():
self._input_conv = blocks_masked_conv2d.RasterScanConv2D(
4 * depth,
filter_size,
strides,
padding,
strict_order=False,
bias=None,
act=None,
initializer=initializer,
name='input_conv2d')
self._hidden_conv = blocks_std.Conv2D(4 * depth,
hidden_filter_size, [1, 1],
'SAME',
bias=None,
act=None,
initializer=initializer,
name='hidden_conv2d')
if bias is not None:
self._bias = blocks_std.BiasAdd(bias, name='biases')
else:
self._bias = blocks_std.PassThrough()
def testConv2DBias(self):
input_shape = [19, 14, 14, 64]
filter_shape = [3, 7, 64, 128]
strides = [1, 2, 2, 1]
output_shape = [19, 6, 4, 128]
conv = blocks_std.Conv2D(depth=filter_shape[-1],
filter_size=filter_shape[0:2],
strides=strides[1:3],
padding='VALID',
act=None,
bias=blocks_std.Bias(1))
x = tf.placeholder(dtype=tf.float32, shape=input_shape)
y = conv(x)
self.CheckBiasAdd(y, conv._bias)
self.assertEqual(output_shape, y.get_shape().as_list())
