The question about conditional MI loss

[PANPANKK]

Thanks for your exciting work! But what's the difference between self.infonce_x1y(self.linears_infonce_x1y(x1_embed), y) and self.infonce_x1x2_cond(torch.cat([self.linears_infonce_x1x2_cond0, y_ohe], dim=1), torch.cat([self.linears_infonce_x1x2_cond1, y_ohe], dim=1)). I mean why we need to use y but y_ohe?