Haozhi Huang

icon indicating an email [email protected]

icon company XVERSE icon location Shenzhen, China icon location I obtained my Ph.D. degree. from Tsinghua University in 2017. I used to be a researcher at Tencent AI Lab. I am currently the leader of XVERSE engine team.

Results 2 issues of Haozhi Huang

Why not calculating TVloss during forward pass?

1 comment

Hi. Comparing TVLoss:updateOutput(input) with TextureLoss:updateOutput(input), I found that the code didn't calculate TVLoss during forward pass but only calculate the gradient of TVLoss during backward pass. Any idea why?

why x_hat reqires_grad?

x_hat = Variable(alpha * X.data + (1 - alpha) * (X.data + 0.5 * X.data.std() * torch.rand(X.size())), requires_grad=True) I suppose we don't need the gradients x_hat right? Since x_hat is...