Closed Shunichi09 closed 5 years ago
I dont understand the question.
This part of the codes is exactly about implementation of eq(5).
Where it generates best_k
number of samples, and computes the error for each one,
and takes the min error:
_g_l2_loss_rel = torch.min(_g_l2_loss_rel) / torch.sum(
loss_mask[start:end])
and then this value is added to the loss function:
loss += g_l2_loss_sum_rel
I'm sorry that I missed the things. I really appreciate your help.
Hi! Thank you for your great paper and your code.
I have a question about the diverse loss referred as eq (5) in your paper. In the part of the generator step function of your code,
It seems that you did not implement that kind of loss function. I thought that this program was for 1V-1 or 1V-20.
It would really help if you have any ideas to implement that loss function. @amiryanj Thank you!!