def check_avg_divisor(self, place):
with fluid.dygraph.guard(place):
input_np = np.random.random([2, 3, 32, 32, 32]).astype("float32")
input = fluid.dygraph.to_variable(input_np)
padding = 0
result = avg_pool3d(input,
kernel_size=2,
stride=2,
padding=padding,
divisor_override=8)
result_np = pool3D_forward_naive(input_np,
ksize=[2, 2, 2],
strides=[2, 2, 2],
paddings=[0, 0, 0],
pool_type='avg')
self.assertTrue(np.allclose(result.numpy(), result_np))
avg_pool3d_dg = paddle.nn.layer.AvgPool3d(kernel_size=2,
stride=2,
padding=0)
result = avg_pool3d_dg(input)
self.assertTrue(np.allclose(result.numpy(), result_np))
padding = [0, 0, 0, 0, 0, 0]
result = avg_pool3d(input,
kernel_size=2,
stride=2,
padding=padding,
divisor_override=8)
self.assertTrue(np.allclose(result.numpy(), result_np))
def check_avg_dygraph_padding_results(self, place):
with fluid.dygraph.guard(place):
input_np = np.random.random([2, 3, 32, 32, 32]).astype("float32")
input = fluid.dygraph.to_variable(input_np)
result = avg_pool3d(input,
kernel_size=2,
stride=2,
padding=1,
ceil_mode=False,
count_include_pad=True)
result_np = avg_pool3D_forward_naive(input_np,
ksize=[2, 2, 2],
strides=[2, 2, 2],
paddings=[1, 1, 1],
ceil_mode=False,
exclusive=False)
self.assertTrue(np.allclose(result.numpy(), result_np))
avg_pool3d_dg = paddle.nn.layer.AvgPool3d(kernel_size=2,
stride=None,
padding=1,
ceil_mode=False,
count_include_pad=True)
result = avg_pool3d_dg(input)
self.assertTrue(np.allclose(result.numpy(), result_np))
def forward(self, inputs):
assert (len(inputs) == self.num_pathways
), "Input tensor does not contain {} pathway".format(
self.num_pathways)
pool_out = []
for pathway in range(self.num_pathways):
if self.pool_size[pathway] is None:
tmp_out = F.adaptive_avg_pool3d(x=inputs[pathway],
output_size=(1, 1, 1),
data_format="NCDHW")
else:
tmp_out = F.avg_pool3d(x=inputs[pathway],
kernel_size=self.pool_size[pathway],
stride=1,
data_format="NCDHW")
pool_out.append(tmp_out)
x = paddle.concat(x=pool_out, axis=1)
x = paddle.transpose(x=x, perm=(0, 2, 3, 4, 1))
# Perform dropout.
if self.dropout_rate > 0.0:
x = self.dropout(x)
x = self.projection(x)
# Performs fully convlutional inference.
if not self.training: # attr of base class
x = F.softmax(x, axis=4)
x = paddle.mean(x, axis=[1, 2, 3])
x = paddle.reshape(x, shape=(x.shape[0], -1))
return x
def run1():
with fluid.dygraph.guard():
input_np = np.random.uniform(
-1, 1, [2, 3, 32, 32, 32]).astype(np.float32)
input_pd = fluid.dygraph.to_variable(input_np)
padding = [[0, 1], [0, 0], [0, 0], [0, 0], [0, 0]]
res_pd = avg_pool3d(
input_pd, kernel_size=2, stride=2, padding=padding)
def run6():
with fluid.dygraph.guard():
input_np = np.random.uniform(-1, 1, [2, 3, 32, 32, 32]).astype(
np.float32)
input_pd = fluid.dygraph.to_variable(input_np)
res_pd = avg_pool3d(input_pd,
kernel_size=2,
stride=2,
padding="padding",
data_format='NNNN')
def run_size_out_of_range():
with fluid.dygraph.guard():
input_np = np.random.uniform(
-1, 1, [2, 3, 32, 32, 32]).astype(np.float32)
input_pd = fluid.dygraph.to_variable(input_np)
res_pd = avg_pool3d(
input_pd,
kernel_size=2,
stride=0,
padding="VALID",
ceil_mode=True)
def check_avg_static_results(self, place):
with fluid.program_guard(fluid.Program(), fluid.Program()):
input = fluid.data(name="input",
shape=[2, 3, 32, 32, 32],
dtype="float32")
result = avg_pool3d(input, kernel_size=2, stride=2, padding=0)
input_np = np.random.random([2, 3, 32, 32, 32]).astype("float32")
result_np = pool3D_forward_naive(input_np,
ksize=[2, 2, 2],
strides=[2, 2, 2],
paddings=[0, 0, 0],
pool_type='avg')
exe = fluid.Executor(place)
fetches = exe.run(fluid.default_main_program(),
feed={"input": input_np},
fetch_list=[result])
self.assertTrue(np.allclose(fetches[0], result_np))
def check_avg_dygraph_results(self, place):
with fluid.dygraph.guard(place):
input_np = np.random.random([2, 3, 32, 32, 32]).astype("float32")
input = fluid.dygraph.to_variable(input_np)
result = avg_pool3d(input, kernel_size=2, stride=2, padding="SAME")
result_np = pool3D_forward_naive(input_np,
ksize=[2, 2, 2],
strides=[2, 2, 2],
paddings=[0, 0, 0],
pool_type='avg',
padding_algorithm="SAME")
self.assertTrue(np.allclose(result.numpy(), result_np))
avg_pool3d_dg = paddle.nn.layer.AvgPool3d(kernel_size=2,
stride=None,
padding="SAME")
result = avg_pool3d_dg(input)
self.assertTrue(np.allclose(result.numpy(), result_np))
