Does tf.model.metrics or compile use the maximum output of softmax?

Hello! I am very new to programming and not from the field so I would greatly appreciate any help.

I made a U-Net model to segment tumors and have used the softmax function in the last convolutional layer. I have a multi-class segmentation, meaning that my output has four channels, and each element has a value of 0-1 in each of those channels due to softmax.

I then used the tf.keras.metrics to calculate my metrics and this function was inputed as an argument into the tf.model.compile function.
My question is - how does tf.model.metrics work on decimal values from 0 to 1? Doesn’t it need that the values first be passed through the argmax function to get either 0 or 1 in each individual channel?

I hope my question is clear and I am not asking too much. Thank you in advance.