CNN, do pretrained cnn models use softmax during training?

[sanjayk]

Does pretrained model that we used for occlusion analysis( (vgg16 in this case) use 'SoftMax during training as well?

In Occlusion Analysis (0511_CNNVisualisation), SoftMax function is applied on the output and results interpreted as probability. If softmax is not used during training than I think softmax output cannot be interpreted as probability.

[Vinay_Suyal]

Cross Entropy Loss consists of both Softmax function as well as Negative loss likelihood.