Which dimension in prediction array is the final output on U2-Net and U-Net 3+?

[vincentallen852]

I am trying figure out which dimension of the model output array on U2-Net and U-Net3+ is the final output.

The models have multiple activation layers and the outputs from the model are multidimensional. I am wondering which number in the predictions is the final output.

E.g. 1

A U2-Net prediction from 10 test images of 512px*512px has the shape (7, 10, 512,512). Is predictions[0] or predictions [6] the final activation output?

E.g. 2

Similarly, a U-Net3+ prediction from 10 test images of 512px*512px has the shape (4, 10, 512,512). Is predictions[0] or predictions[3] the final activation output?

I would appreciate any help.