The predicted labels and Accuracy Layer

[Douglas Almonfrey]

Hi all, I'm new with Caffe. Does the Accuracy layer perform ArgMax operation? I ask this because in the MNIST example, at the test stage, the last layer is the Accuracy layer and it receives data from an Inner Product layer (IP2)... At the training stage, the last layer is the SoftmaxLoss layer and ok, this layer performs the Argmax operation and computes the loss, for sure... But at the test stage I'm confuse. Additionally, if the Accuracy layer performs ArgMax operation, how can I access the predicted labels?

Thank you very much.