Question signed sqrt normalization

[zhujiem]

Is it correct to use torch.sqrt(outputs + 1e-5) since outputs may have negative values? Thanks!

https://github.com/dasguptar/bcnn.pytorch/blob/master/bcnn/model.py#L32

outputs = torch.sign(outputs) * torch.sqrt(outputs + 1e-5) # signed square root normalization

[dasguptar]

@jimzhu We are taking the output of VGG16 from a RELU layer, so it will be positive. After that, the bilinear product is also going to positive, and we can do regular sqrt.