Documentation for SoftmaxWithLoss about label shape

[dashesy]

Issue summary

The documentation for SoftmaxWithLoss is

@param bottom input Blob vector (length 2)

* -#

 *      the labels @f$ l @f$, an integer-valued Blob with values

But here in the code this is what is checked:

  CHECK_EQ(outer_num_ * inner_num_, bottom[1]->count())
      << "Number of labels must match number of predictions; "
      << "e.g., if softmax axis == 1 and prediction shape is (N, C, H, W), "
      << "label count (number of labels) must be N*H*W, "
      << "with integer values in {0, 1, ..., C-1}.";

So the documentation should have been:

* -#

[shaibagon]

it is a little bit more complicated than that... The label shape depends on the axis parameter as well...

[dashesy]

axis parameter should point to C. The only way that could be correct is if axis is -1 (last one). Even then, having C == 1 is not correct.