def func(self, place):
shape = [2, 4, 3, 3]
eps = 0.005
dtype = np.float64
if core.is_compiled_with_rocm():
dtype = np.float32
x = layers.data('x', shape, False, dtype)
y = layers.conv2d_transpose(
x, 2, filter_size=1, groups=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))
if core.is_compiled_with_rocm():
# HIP will sometimes fail if no atol
gradient_checker.double_grad_check(
[x] + w,
y,
x_init=[x_arr] + w_arr,
place=place,
eps=eps,
atol=1e-4)
else:
gradient_checker.double_grad_check(
[x] + w, y, x_init=[x_arr] + w_arr, place=place, eps=eps)
gradient_checker.double_grad_check_for_dygraph(
self.conv_transpose_wrapper, [x] + w,
y,
x_init=[x_arr] + w_arr,
place=place)
def func(self, place):
shape = [2, 3, 3, 2]
eps = 0.005
dtype = np.float64
if core.is_compiled_with_rocm():
dtype = np.float32
x = layers.data('x', shape, False, dtype)
y = layers.conv2d_transpose(input=x,
num_filters=2,
filter_size=1,
padding=[1, 1],
bias_attr=False,
use_cudnn=True,
groups=1,
data_format="NHWC")
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))
if core.is_compiled_with_rocm():
# HIP will sometimes fail if no atol
gradient_checker.double_grad_check([x] + w,
y,
x_init=[x_arr] + w_arr,
place=place,
eps=eps,
atol=1e-4)
else:
gradient_checker.double_grad_check([x] + w,
y,
x_init=[x_arr] + w_arr,
place=place,
eps=eps)
def net(self, input):
c1, c2, c3, c4, c5 = self.backbone(input)
channels = [64, 128, 256, 512]
x = ConvUp(c5, c4, channels[2], name='up5')
x = ConvUp(x, c3, channels[1], name='up6')
x = ConvUp(x, c2, channels[0], name='up7')
x = ConvUp(x, c1, channels[0], name='up8')
x = FL.conv2d_transpose(x, num_filters=self.out_channels, filter_size=2, stride=2)
return x
def __call__(self, input):
return layers.conv2d_transpose(
input=input,
num_filters=num_filters,
output_size=output_size,
filter_size=filter_size,
padding=padding,
stride=stride,
dilation=dilation,
groups=groups,
param_attr=self.attr_holder.param_attr,
bias_attr=self.attr_holder.bias_attr,
use_cudnn=use_cudnn,
act=act)
def func(self, place):
shape = [2, 4, 3, 3]
eps = 0.005
dtype = np.float64
x = layers.data('x', shape, False, dtype)
y = layers.conv2d_transpose(x,
2,
filter_size=1,
groups=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 test_conv2d_transpose(self):
program = Program()
with program_guard(program):
img = layers.data(name='pixel', shape=[3, 2, 2], dtype='float32')
layers.conv2d_transpose(input=img, num_filters=10, output_size=28)
print(str(program))
def ConvUp(x1, x2, out_channels, name=None):
x1 = FL.conv2d_transpose(x1, num_filters=x1.shape[1] // 2, filter_size=2, stride=2)
x = FL.concat([x1,x2], axis=1)
x = DoubleConv_up(x, out_channels, name=name+"_doubleconv")
return x
def test_conv2d_transpose(self):
program = Program()
with program_guard(program):
img = layers.data(name='pixel', shape=[3, 2, 2], dtype='float32')
layers.conv2d_transpose(input=img, num_filters=10, output_size=28)
print(str(program))
