Pytorch中torch.cat()函數(shù)舉例解析

更新時間：2022年12月22日 16:12:16 作者：cv_lhp

一般torch.cat()是為了把多個tensor進行拼接而存在的,下面這篇文章主要給大家介紹了關于Pytorch中torch.cat()函數(shù)舉例解析的相關資料,文中通過示例代碼介紹的非常詳細,需要的朋友可以參考下

一. torch.cat（）函數(shù)解析

1. 函數(shù)說明

1.1 官網(wǎng)：torch.cat()，函數(shù)定義及參數(shù)說明如下圖所示：

1.2 函數(shù)功能

函數(shù)將兩個張量（tensor）按指定維度拼接在一起，注意：除拼接維數(shù)dim數(shù)值可不同外其余維數(shù)數(shù)值需相同，方能對齊，如下面例子所示。torch.cat()函數(shù)不會新增維度，而torch.stack()函數(shù)會新增一個維度，相同的是兩個都是對張量進行拼接

2. 代碼舉例

2.1 輸入兩個二維張量（dim=0）：dim=0對行進行拼接

a = torch.randn(2,3)
b =  torch.randn(3,3)
c = torch.cat((a,b),dim=0)
a,b,c

輸出結果如下：
(tensor([[-0.90, -0.37, 1.96],
[-2.65, -0.60, 0.05]]),
tensor([[ 1.30, 0.24, 0.27],
[-1.99, -1.09, 1.67],
[-1.62, 1.54, -0.14]]),
tensor([[-0.90, -0.37, 1.96],
[-2.65, -0.60, 0.05],
[ 1.30, 0.24, 0.27],
[-1.99, -1.09, 1.67],
[-1.62, 1.54, -0.14]]))

2.2 輸入兩個二維張量（dim=1）: dim=1對列進行拼接

a = torch.randn(2,3)
b =  torch.randn(2,4)
c = torch.cat((a,b),dim=1)
a,b,c

輸出結果如下：
(tensor([[-0.55, -0.84, -1.60],
[ 0.39, -0.96, 1.02]]),
tensor([[-0.83, -0.09, 0.05, 0.17],
[ 0.28, -0.74, -0.27, -0.85]]),
tensor([[-0.55, -0.84, -1.60, -0.83, -0.09, 0.05, 0.17],
[ 0.39, -0.96, 1.02, 0.28, -0.74, -0.27, -0.85]]))

2.3 輸入兩個三維張量：dim=0 對通道進行拼接

a = torch.randn(2,3,4)
b =  torch.randn(1,3,4)
c = torch.cat((a,b),dim=0)
a,b,c

輸出結果如下：
(tensor([[[ 0.51, -0.72, -0.02, 0.76],
[ 0.72, 1.01, 0.39, -0.13],
[ 0.37, -0.63, -2.69, 0.74]],

[[ 0.72, -0.31, -0.27, 0.10],
[ 1.66, -0.06, 1.91, -0.66],
[ 0.34, -0.23, -0.18, -1.22]]]),
tensor([[[ 0.94, 0.77, -0.41, -1.20],
[-0.23, -1.03, -0.25, 1.67],
[-1.00, -0.68, -0.35, -0.50]]]),
tensor([[[ 0.51, -0.72, -0.02, 0.76],
[ 0.72, 1.01, 0.39, -0.13],
[ 0.37, -0.63, -2.69, 0.74]],

[[ 0.72, -0.31, -0.27, 0.10],
[ 1.66, -0.06, 1.91, -0.66],
[ 0.34, -0.23, -0.18, -1.22]],

[[ 0.94, 0.77, -0.41, -1.20],
[-0.23, -1.03, -0.25, 1.67],
[-1.00, -0.68, -0.35, -0.50]]]))

2.4 輸入兩個三維張量：dim=1對行進行拼接

a = torch.randn(2,3,4)
b =  torch.randn(2,4,4)
c = torch.cat((a,b),dim=1)
a,b,c

輸出結果如下：
(tensor([[[-0.86, 0.00, -1.26, 1.20],
[-0.46, -1.08, -0.82, 2.03],
[-0.89, 0.43, 1.92, 0.49]],

[[ 0.24, -0.02, 0.32, 0.97],
[ 0.33, -1.34, 0.76, -1.55],
[ 0.38, 1.45, 0.27, -0.64]]]),
tensor([[[ 0.82, 0.85, -0.30, -0.58],
[-0.09, 0.40, 0.02, 0.75],
[-0.70, 0.67, -0.88, -0.50],
[-0.62, -1.65, -1.10, -1.39]],

[[-0.85, -1.61, -0.35, -0.56],
[ 0.00, 1.40, 0.41, 0.39],
[-0.01, 0.04, 0.80, 0.41],
[-1.21, -0.64, 1.14, 1.64]]]),
tensor([[[-0.86, 0.00, -1.26, 1.20],
[-0.46, -1.08, -0.82, 2.03],
[-0.89, 0.43, 1.92, 0.49],
[ 0.82, 0.85, -0.30, -0.58],
[-0.09, 0.40, 0.02, 0.75],
[-0.70, 0.67, -0.88, -0.50],
[-0.62, -1.65, -1.10, -1.39]],

[[ 0.24, -0.02, 0.32, 0.97],
[ 0.33, -1.34, 0.76, -1.55],
[ 0.38, 1.45, 0.27, -0.64],
[-0.85, -1.61, -0.35, -0.56],
[ 0.00, 1.40, 0.41, 0.39],
[-0.01, 0.04, 0.80, 0.41],
[-1.21, -0.64, 1.14, 1.64]]]))

2.5 輸入兩個三維張量：dim=2對列進行拼接

a = torch.randn(2,3,4)
b =  torch.randn(2,3,5)
c = torch.cat((a,b),dim=2)
a,b,c

輸出結果如下：
(tensor([[[ 0.13, -0.02, 0.13, -0.25],
[ 1.42, -0.22, -0.87, 0.27],
[-0.07, 1.04, -0.06, 0.91]],

[[ 0.88, -1.46, 0.04, 0.35],
[ 1.36, 0.64, 0.75, 0.39],
[ 0.36, 1.13, 0.83, 0.56]]]),
tensor([[[-0.47, -2.30, -0.49, -1.02, 1.74],
[ 0.71, 0.89, 0.80, -0.05, -1.35],
[-0.40, 0.26, -0.78, -1.50, -0.92]],

[[-0.77, -0.01, 1.23, 0.70, -0.66],
[ 0.28, -0.18, -0.91, 2.23, 1.14],
[-1.93, -0.17, 0.15, 0.40, 0.32]]]),
tensor([[[ 0.13, -0.02, 0.13, -0.25, -0.47, -2.30, -0.49, -1.02, 1.74],
[ 1.42, -0.22, -0.87, 0.27, 0.71, 0.89, 0.80, -0.05, -1.35],
[-0.07, 1.04, -0.06, 0.91, -0.40, 0.26, -0.78, -1.50, -0.92]],

[[ 0.88, -1.46, 0.04, 0.35, -0.77, -0.01, 1.23, 0.70, -0.66],
[ 1.36, 0.64, 0.75, 0.39, 0.28, -0.18, -0.91, 2.23, 1.14],
[ 0.36, 1.13, 0.83, 0.56, -1.93, -0.17, 0.15, 0.40, 0.32]]]))