def func(self, place):
shape = [2, 2, 2, 2, 3]
eps = 0.005
dtype = np.float32 if fluid.core.is_compiled_with_rocm(
) else np.float64
x = layers.data('x', shape, False, dtype)
y = layers.conv3d(input=x,
num_filters=2,
filter_size=1,
padding=[1, 0, 1, 0, 1, 0],
bias_attr=False,
use_cudnn=True,
groups=1,
data_format="NDHWC")
x_arr = np.random.uniform(-1, 1, shape).astype(dtype)
w = fluid.default_main_program().global_block().all_parameters()
w_arr = []
for p in w:
w_arr.append(np.random.uniform(-1, 1, p.shape).astype(dtype))
gradient_checker.double_grad_check([x] + w,
y,
x_init=[x_arr] + w_arr,
place=place,
eps=eps)
def func(self, place):
shape = [2, 4, 5, 3, 2]
eps = 0.005
dtype = np.float64
x = layers.data('x', shape, False, dtype)
y = layers.conv3d(x, 2, 1, padding=1, bias_attr=False)
x_arr = np.random.uniform(-1, 1, shape).astype(dtype)
w = fluid.default_main_program().global_block().all_parameters()
w_arr = []
for p in w:
w_arr.append(np.random.uniform(-1, 1, p.shape).astype(dtype))
gradient_checker.double_grad_check(
[x] + w, y, x_init=[x_arr] + w_arr, place=place, eps=eps)
def _build_net(self):
# ConvLSTM2D
rnn_out, last_hidden = ConvLSTM.convlstm2d_rnn(
rnn_input=self.input,
init_hidden=None,
init_cell=None,
padding=1,
hidden_h=self.h,
hidden_w=self.w,
filters=self.filters,
filter_size=self.filter_size,
sequence_length=self.input_seqlen)
# Batch Norm
bn = layers.layer_norm(rnn_out, begin_norm_axis=4)
# ConvLSTM2D
rnn_out, last_hidden = ConvLSTM.convlstm2d_rnn(
rnn_input=bn,
init_hidden=None,
init_cell=None,
padding=1,
hidden_h=self.h,
hidden_w=self.w,
filters=self.filters,
filter_size=self.filter_size,
sequence_length=self.input_seqlen)
# Batch Norm
bn = layers.layer_norm(rnn_out, begin_norm_axis=4)
# ConvLSTM2D
rnn_out, last_hidden = ConvLSTM.convlstm2d_rnn(
rnn_input=bn,
init_hidden=None,
init_cell=None,
padding=1,
hidden_h=self.h,
hidden_w=self.w,
filters=self.filters,
filter_size=self.filter_size,
sequence_length=self.input_seqlen)
# Batch Norm
bn = layers.layer_norm(rnn_out, begin_norm_axis=4)
# ConvLSTM2D
rnn_out, last_hidden = ConvLSTM.convlstm2d_rnn(
rnn_input=bn,
init_hidden=None,
init_cell=None,
padding=1,
hidden_h=self.h,
hidden_w=self.w,
filters=self.filters,
filter_size=self.filter_size,
sequence_length=self.input_seqlen)
# Batch Norm
bn = layers.layer_norm(rnn_out, begin_norm_axis=4)
# Transpose : (batch x C x D x H x W)
tr = layers.transpose(bn, [0, 4, 1, 2, 3])
# Conv3D
conv3d = layers.conv3d(input=tr,
num_filters=2,
filter_size=3,
padding=1)
# conv3d : (batch x C x D x H x W)
conv3d = layers.transpose(conv3d, [0, 2, 3, 4, 1])
# conv3d: (batch x D x H x W x C)
return conv3d
