“[...]“、“[::]“与torch.cat的解析（YOLOv5 Focus common.py）

"[...]"、"[::]"与torch.cat的解析（YOLOv5 common.py）

YOLOv5中common.py的函数Focus.forward，返回为self.conv(torch.cat([x[..., ::2, ::2], x[..., 1::2, ::2], x[..., ::2, 1::2], x[..., 1::2, 1::2]], 1))，这句话的解析。

def forward(self, x):  # x(b,c,w,h) -> y(b,4c,w/2,h/2)
    return self.conv(torch.cat([x[..., ::2, ::2], x[..., 1::2, ::2], x[..., ::2, 1::2], x[..., 1::2, 1::2]], 1))

1. "[...]"简单的使用

A = np.arange(25).reshape(5,5)
array([[ 0,  1,  2,  3,  4],
       [ 5,  6,  7,  8,  9],
       [10, 11, 12, 13, 14],
       [15, 16, 17, 18, 19],
       [20, 21, 22, 23, 24]])

  (1) 遍历每行，2表示索引为2的所在列

A[...,2]
array([ 2,  7, 12, 17, 22])

（2）遍历每行，:2表示索引<2的0，1所在的列

A[...,:2]
array([[ 0,  1],
       [ 5,  6],
       [10, 11],
       [15, 16],
       [20, 21]])

（3）遍历每行，::2表示步长为2，即0，2，4列

A[...,::2]
array([[ 0,  2,  4],
       [ 5,  7,  9],
       [10, 12, 14],
       [15, 17, 19],
       [20, 22, 24]])

（4）::2表示取间隔步长为2的取行数据，遍历每列

A[::2,...]
array([[ 0,  1,  2,  3,  4],
       [10, 11, 12, 13, 14],
       [20, 21, 22, 23, 24]])

（5）相当于插入维度，即reshape(A,[1,4,4])

A[None,...]
array([[[ 0,  1,  2,  3,  4],
        [ 5,  6,  7,  8,  9],
        [10, 11, 12, 13, 14],
        [15, 16, 17, 18, 19],
        [20, 21, 22, 23, 24]]])

总结："..."表示所有，"::x"表示x个步长

2. torch.cat

A = np.arange(9).reshape(3,3)
B = np.arange(1,10).reshape(3,3)
AA = torch.from_numpy(A)
BB = torch.from_numpy(B)

（1）按维度0拼接，即竖着拼接

C1 = torch.cat([AA,BB],0)
tensor([[0, 1, 2],
        [3, 4, 5],
        [6, 7, 8],
        [1, 2, 3],
        [4, 5, 6],
        [7, 8, 9]], dtype=torch.int32)

（2）按维度1拼接，即横着拼接

C2 = torch.cat([AA,BB],1)
tensor([[0, 1, 2, 1, 2, 3],
        [3, 4, 5, 4, 5, 6],
        [6, 7, 8, 7, 8, 9]], dtype=torch.int32)

3. torch.cat([x[..., ::2, ::2], x[..., 1::2, ::2], x[..., ::2, 1::2], x[..., 1::2, 1::2]], 1)

X = np.arange(16*3).reshape(3,4,4)
array([[[ 0,  1,  2,  3],
        [ 4,  5,  6,  7],
        [ 8,  9, 10, 11],
        [12, 13, 14, 15]],

       [[16, 17, 18, 19],
        [20, 21, 22, 23],
        [24, 25, 26, 27],
        [28, 29, 30, 31]],

       [[32, 33, 34, 35],
        [36, 37, 38, 39],
        [40, 41, 42, 43],
        [44, 45, 46, 47]]])

（1）行和列步长间隔都为2

X[...,::2,::2]
array([[[ 0,  2],
        [ 8, 10]],

       [[16, 18],
        [24, 26]],

       [[32, 34],
        [40, 42]]])

（2）行和列步长间隔都为2，行的起始位置为1

X[..., 1::2, ::2]
array([[[ 4,  6],
        [12, 14]],

       [[20, 22],
        [28, 30]],

       [[36, 38],
        [44, 46]]])

（3）行和列步长间隔都为2，列的起始位置为1

X[..., ::2, 1::2]
array([[[ 1,  3],
        [ 9, 11]],

       [[17, 19],
        [25, 27]],

       [[33, 35],
        [41, 43]]])

（4）torch.cat([x[..., ::2, ::2], x[..., 1::2, ::2], x[..., ::2, 1::2], x[..., 1::2, 1::2]], 1)

X = torch.from_numpy(X)
tensor([[[ 0,  1,  2,  3],
         [ 4,  5,  6,  7],
         [ 8,  9, 10, 11],
         [12, 13, 14, 15]],

        [[16, 17, 18, 19],
         [20, 21, 22, 23],
         [24, 25, 26, 27],
         [28, 29, 30, 31]],

        [[32, 33, 34, 35],
         [36, 37, 38, 39],
         [40, 41, 42, 43],
         [44, 45, 46, 47]]], dtype=torch.int32)


torch.cat([X[..., ::2, ::2], X[..., 1::2, ::2], X[..., ::2, 1::2], X[..., 1::2, 1::2]], 1)
tensor([[[ 0,  2],
         [ 8, 10],
         [ 4,  6],
         [12, 14],
         [ 1,  3],
         [ 9, 11],
         [ 5,  7],
         [13, 15]],

        [[16, 18],
         [24, 26],
         [20, 22],
         [28, 30],
         [17, 19],
         [25, 27],
         [21, 23],
         [29, 31]],

        [[32, 34],
         [40, 42],
         [36, 38],
         [44, 46],
         [33, 35],
         [41, 43],
         [37, 39],
         [45, 47]]], dtype=torch.int32)