Kalomaze equates rpGAN loss to Bradley-Terry ranking models.
Cool to see that on vision problems with unpaired data, relativistic GAN is still SOTA
https://openaccess.thecvf.com/content/CVPR2026W/NTIRE/papers/Perevozchikov_NTIRE_2026_Challenge_on_Learned_Smartphone_ISP_with_Unpaired_Data_CVPRW_2026_paper.pdf
Users call Relativistic GAN genuinely impressive for retaining SOTA on unpaired vision tasks because it succeeds despite unpaired data making everything harder.
classic/hinge GAN loss == binary cross entropy rpGAN/relative GAN loss == bradley terry mode collapse from the old GAN literature is primarily an artifact of the fact that pointwise scalars aren't grounded in the relative gap between the real and fake samples ranking > rating
Cool to see that on vision problems with unpaired data, relativistic GAN is still SOTA
https://openaccess.thecvf.com/content/CVPR2026W/NTIRE/papers/Perevozchikov_NTIRE_2026_Challenge_on_Learned_Smartphone_ISP_with_Unpaired_Data_CVPRW_2026_paper.pdf
@jm_alexia unpaired data makes everything harder but RGAN holding it down is genuinely impressive
Kalomaze equates rpGAN loss to Bradley-Terry ranking models.
Cool to see that on vision problems with unpaired data, relativistic GAN is still SOTA
https://openaccess.thecvf.com/content/CVPR2026W/NTIRE/papers/Perevozchikov_NTIRE_2026_Challenge_on_Learned_Smartphone_ISP_with_Unpaired_Data_CVPRW_2026_paper.pdf