Repeatable loss, is exp(), or softmax()

[wuhanfeng-000]

Thank you for great job!!!
i have a small question about repeatable loss.In ALIKE paper, the matching probability map is obtained by normalizing the similarity map.but use exp() while in decriptor loss use softmax().
it is a careless mistake?or i misunderstood.
Thanks