PyTorch之图像和Tensor填充的实例
在PyTorch中可以对图像和Tensor进行填充，如常量值填充，镜像填充和复制填充等。在图像预处理阶段设置图像边界填充的方式如下：
importvision.torchvision.transformsastransforms img_to_pad=transforms.Compose([ transforms.Pad(padding=2,padding_mode='symmetric'), transforms.ToTensor(), ])
对Tensor进行填充的方式如下：
importtorch.nn.functionalasF feature=feature.unsqueeze(0).unsqueeze(0) avg_feature=F.pad(feature,pad=[1,1,1,1],mode='replicate')
这里需要注意一点的是，transforms.Pad只能对PIL图像格式进行填充，而F.pad可以对Tensor进行填充，目前F.pad不支持对2DTensor进行填充，可以通过unsqueeze扩展为4DTensor进行填充。
F.pad的部分源码如下：
@torch._jit_internal.weak_script
defpad(input,pad,mode='constant',value=0):
#type:(Tensor,List[int],str,float)->Tensor
r"""Padstensor.
Padingsize:
Thenumberofdimensionstopadis:math:`\left\lfloor\frac{\text{len(pad)}}{2}\right\rfloor`
andthedimensionsthatgetpaddedbeginswiththelastdimensionandmovesforward.
Forexample,topadthelastdimensionoftheinputtensor,then`pad`hasform
`(padLeft,padRight)`;topadthelast2dimensionsoftheinputtensor,thenuse
`(padLeft,padRight,padTop,padBottom)`;topadthelast3dimensions,use
`(padLeft,padRight,padTop,padBottom,padFront,padBack)`.
Paddingmode:
See:class:`torch.nn.ConstantPad2d`,:class:`torch.nn.ReflectionPad2d`,and
:class:`torch.nn.ReplicationPad2d`forconcreteexamplesonhoweachofthe
paddingmodesworks.Constantpaddingisimplementedforarbitrarydimensions.
Replicatepaddingisimplementedforpaddingthelast3dimensionsof5Dinput
tensor,orthelast2dimensionsof4Dinputtensor,orthelastdimensionof
3Dinputtensor.Reflectpaddingisonlyimplementedforpaddingthelast2
dimensionsof4Dinputtensor,orthelastdimensionof3Dinputtensor.
..include::cuda_deterministic_backward.rst
Args:
input(Tensor):`Nd`tensor
pad(tuple):m-elemtuple,where:math:`\frac{m}{2}\leq`inputdimensionsand:math:`m`iseven.
mode:'constant','reflect'or'replicate'.Default:'constant'
value:fillvaluefor'constant'padding.Default:0
Examples::
>>>t4d=torch.empty(3,3,4,2)
>>>p1d=(1,1)#padlastdimby1oneachside
>>>out=F.pad(t4d,p1d,"constant",0)#effectivelyzeropadding
>>>print(out.data.size())
torch.Size([3,3,4,4])
>>>p2d=(1,1,2,2)#padlastdimby(1,1)and2ndtolastby(2,2)
>>>out=F.pad(t4d,p2d,"constant",0)
>>>print(out.data.size())
torch.Size([3,3,8,4])
>>>t4d=torch.empty(3,3,4,2)
>>>p3d=(0,1,2,1,3,3)#padby(0,1),(2,1),and(3,3)
>>>out=F.pad(t4d,p3d,"constant",0)
>>>print(out.data.size())
torch.Size([3,9,7,3])
"""
assertlen(pad)%2==0,'Paddinglengthmustbedivisibleby2'
assertlen(pad)//2<=input.dim(),'Paddinglengthtoolarge'
ifmode=='constant':
ret=_VF.constant_pad_nd(input,pad,value)
else:
assertvalue==0,'Paddingmode"{}""doesn\'ttakeinvalueargument'.format(mode)
ifinput.dim()==3:
assertlen(pad)==2,'3Dtensorsexpect2valuesforpadding'
ifmode=='reflect':
ret=torch._C._nn.reflection_pad1d(input,pad)
elifmode=='replicate':
ret=torch._C._nn.replication_pad1d(input,pad)
else:
ret=input#TODO:removethiswhenjitraisesupportscontrolflow
raiseNotImplementedError
elifinput.dim()==4:
assertlen(pad)==4,'4Dtensorsexpect4valuesforpadding'
ifmode=='reflect':
ret=torch._C._nn.reflection_pad2d(input,pad)
elifmode=='replicate':
ret=torch._C._nn.replication_pad2d(input,pad)
else:
ret=input#TODO:removethiswhenjitraisesupportscontrolflow
raiseNotImplementedError
elifinput.dim()==5:
assertlen(pad)==6,'5Dtensorsexpect6valuesforpadding'
ifmode=='reflect':
ret=input#TODO:removethiswhenjitraisesupportscontrolflow
raiseNotImplementedError
elifmode=='replicate':
ret=torch._C._nn.replication_pad3d(input,pad)
else:
ret=input#TODO:removethiswhenjitraisesupportscontrolflow
raiseNotImplementedError
else:
ret=input#TODO:removethiswhenjitraisesupportscontrolflow
raiseNotImplementedError("Only3D,4D,5Dpaddingwithnon-constantpaddingaresupportedfornow")
returnret
以上这篇PyTorch之图像和Tensor填充的实例就是小编分享给大家的全部内容了，希望能给大家一个参考，也希望大家多多支持毛票票。