Question about Dual Task Loss notation in the paper

[hhhyyeee]

Hi. I've come up with a question about the loss notation in the GSCNN paper.

The paper is telling us that since GSCNN is built on Joint Multi-Task Learning, the overall loss for the network is composed like Eq. 3:

However, later in the paper, the same notation of overall loss is used, but with the Dual Task Regularizer, in Eq. 7:

So the question is:

1. Is this typo? When we look into the official pytorch code, we can easily recognize that all the losses (seg loss, bce loss, attention loss, dualtask loss) are just added together (loss.py)
2. If not, does the regularizer of Eq. 7 work as if it limits the magnitude of the loss of Eq.3?

Thank you.