Open rracinskij opened 7 years ago
Hi,
just noticed that nn.SpatialLogSoftMax() outputs zeros if the batch size is 1.
E.g. the readme example
ii=torch.randn(4,8,16,16) -- batchSize x features x height x width m=nn.SpatialLogSoftMax() oo = m:forward(ii)
works properly, but torch.randn(4,8,16,16) will output zeros. Is it a bug or a feature?
torch.randn(4,8,16,16)
I have the same problem, which leads the output error of nn.ClassNLLCriterion to inf
Hi,
just noticed that nn.SpatialLogSoftMax() outputs zeros if the batch size is 1.
E.g. the readme example
works properly, but
torch.randn(4,8,16,16)
will output zeros. Is it a bug or a feature?