Hi,
I wanted to ask that to have a fair comparison with GANSPace method in Fig.6, you need to shift your method in S space and GANSpace method in W space with a shift magnitude (sigma) which leads to equal change in the logit value of pretrained models.
Could you please let me know what are the proportion of shift magnitudes in W space of GANSpace and S space in your proposed method? I am interested to know the proportion of shift magnitudes for all attributes in your paper, if they are different for various directions you found.
Thanks a lot in advance!
Hi,
I wanted to ask that to have a fair comparison with GANSPace method in Fig.6, you need to shift your method in S space and GANSpace method in W space with a shift magnitude (sigma) which leads to equal change in the logit value of pretrained models.
Could you please let me know what are the proportion of shift magnitudes in W space of GANSpace and S space in your proposed method? I am interested to know the proportion of shift magnitudes for all attributes in your paper, if they are different for various directions you found.
Thanks a lot in advance!