Nonnegative Probability Map

[tsuijenk]

Hello,

For the count loss, the input, which is a probability map, ranges from [0,1], and should be a result of applying the sigmoid activation function after the 2nd final convolutional layer. Is that right?

I got confused by the wording here: 0: probabilities (without sigmoid), line72.

Link to code: https://github.com/TuragaLab/DECODE/blob/master/decode/neuralfitter/loss.py

Thank you!

[ASpeiser]

Yes exactly. The Comment you mention is part of a different loss function that is not in use in the current version.