torch.nonzero一维、二维、三维

torch.nonzero

一维

import torch
input = torch.randint(15, 17, (6, 2))
rels_mask = torch.nonzero(input[:, 0] - 15)
print(input)
print(input[:, 0].shape)
print('----------------------')
print(rels_mask)
print(rels_mask.shape)

输出

tensor([[15, 15],
        [16, 15],
        [16, 16],
        [16, 16],
        [15, 15],
        [16, 16]])
torch.Size([6])
----------------------
tensor([[1],
        [2],
        [3],
        [5]])
torch.Size([4, 1])

二维、三维

参考

https://blog.csdn.net/monchin/article/details/79750216