将 3d 矩阵列表转换为 4d 矩阵

Question

我正在尝试编写一个函数，它接收 3d 矩阵列表。

所以.. list 中的每个元素都有 shape (rows,cols, some_scalar).。我正在尝试将其重塑为 4d 矩阵..所以output = (number_of_elements_in_matrix, rows,cols,some_scalar)

到目前为止我有

output = np.zeros((len(list_of_matrices), list_of_matrices[0].shape[0], list_of_matrices[0].shape[1],
                      list_of_matrices[0].shape[2]), dtype=np.uint8)

我怎么知道用这些值填充这个输出 4d 张量..

def reshape_matrix(list_of_matrices):
   output = np.zeros((len(list_of_matrices), list_of_matrices[0].shape[0], list_of_matrices[0].shape[1],
                          list_of_matrices[0].shape[2]), dtype=np.uint8)


   return output

Answer 1

您可以使用np.stack沿第一个轴（轴 = 0）堆叠，如下所示 -

np.stack(list_of_matrices,axis=0)

样品运行 -

In [22]: # Create an input list of arrays
    ...: arr1 = np.random.rand(4,5,2)
    ...: arr2 = np.random.rand(4,5,2)
    ...: arr3 = np.random.rand(4,5,2)
    ...: list_of_matrices = [arr1,arr2,arr3]
    ...: 

In [23]: np.stack(list_of_matrices,axis=0).shape
Out[23]: (3, 4, 5, 2)