def run7():
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 = max_pool3d(input_pd,
kernel_size=2,
stride=2,
padding="padding",
data_format='NNNN')
def run10():
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 = max_pool3d(input_pd,
kernel_size=2,
stride=2,
padding=0,
data_format='NDHWC',
return_indices=True)
def forward(self, x):
x = self._conv(x)
x = self._bn(x)
x = F.relu(x)
x = F.max_pool3d(x=x,
kernel_size=[1, 3, 3],
stride=[1, 2, 2],
padding=[0, 1, 1],
data_format="NCDHW")
return x
def check_max_dygraph_padding(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, 0], [0, 0], [0, 0], [0, 0], [0, 0]]
result = max_pool3d(input, kernel_size=2, stride=2, padding=padding)
result_np = pool3D_forward_naive(
input_np,
ksize=[2, 2, 2],
strides=[2, 2, 2],
paddings=[0, 0, 0],
pool_type='max')
self.assertTrue(np.allclose(result.numpy(), result_np))
max_pool3d_dg = paddle.nn.layer.MaxPool3D(
kernel_size=2, stride=2, padding=0)
result = max_pool3d_dg(input)
self.assertTrue(np.allclose(result.numpy(), result_np))
padding = [0, 0, 0, 0, 0, 0]
result = max_pool3d(input, kernel_size=2, stride=2, padding=padding)
self.assertTrue(np.allclose(result.numpy(), result_np))
def check_max_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 = max_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='max')
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 forward(self, x):
x = self.s1(x) #VideoModelStem
x = self.s1_fuse(x) #FuseFastToSlow
x = self.s2(x) #ResStage
x = self.s2_fuse(x)
for pathway in range(self.num_pathways):
x[pathway] = F.max_pool3d(
x=x[pathway],
kernel_size=self.pool_size_ratio[pathway],
stride=self.pool_size_ratio[pathway],
padding=[0, 0, 0],
data_format="NCDHW")
x = self.s3(x)
x = self.s3_fuse(x)
x = self.s4(x)
x = self.s4_fuse(x)
x = self.s5(x)
return x
def check_max_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 = max_pool3d(
input, kernel_size=2, stride=2, padding=1, ceil_mode=False)
result_np = max_pool3D_forward_naive(
input_np,
ksize=[2, 2, 2],
strides=[2, 2, 2],
paddings=[1, 1, 1],
ceil_mode=False)
self.assertTrue(np.allclose(result.numpy(), result_np))
max_pool3d_dg = paddle.nn.layer.MaxPool3D(
kernel_size=2, stride=None, padding=1, ceil_mode=False)
result = max_pool3d_dg(input)
self.assertTrue(np.allclose(result.numpy(), result_np))
def check_max_dygraph_ndhwc_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(
np.transpose(input_np, [0, 2, 3, 4, 1]))
result = max_pool3d(input,
kernel_size=2,
stride=2,
padding=0,
data_format="NDHWC",
return_indices=False)
result_np = pool3D_forward_naive(input_np,
ksize=[2, 2, 2],
strides=[2, 2, 2],
paddings=[0, 0, 0],
pool_type='max')
self.assertTrue(
np.allclose(np.transpose(result.numpy(), [0, 4, 1, 2, 3]),
result_np))
def check_max_dygraph_stride_is_none(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, indices = max_pool3d(input,
kernel_size=2,
stride=None,
padding="SAME",
return_indices=True)
result_np = pool3D_forward_naive(input_np,
ksize=[2, 2, 2],
strides=[2, 2, 2],
paddings=[0, 0, 0],
pool_type='max',
padding_algorithm="SAME")
self.assertTrue(np.allclose(result.numpy(), result_np))
max_pool3d_dg = paddle.nn.layer.MaxPool3d(kernel_size=2,
stride=2,
padding=0)
result = max_pool3d_dg(input)
self.assertTrue(np.allclose(result.numpy(), result_np))
