My question doesn’t refer straightforward to the PyTorch library, but vision problem.
I try to assess image quality with the Spearman rank-order coefficient metric(SROCC). The loss function combines few losses and my model train on two labels (input itself and additional label extracted from original image). So I wonder how I am supposed to assess the quality of the result when I have a combination of two different labels that influence the result. So I thought to separately measure with SROCC metric between output and first label (the image itself) and between output and the second label, then to average both results.
I am not sure about the correctness of my solution, any other suggestions?