according to the paper, HgWgDg is matched up with HxWxDx using trilinear interpolation, which is also called "Resampler" in the figure 2.
but in the code of this repo, there is no this step, instead you use the feature map after the Up_conv step. So the code can be runned, But it is clearly different from what proposed in the figure 2.
So I just wonder whether you implemented this way on purpose? And if yes, please enlighten me why it's better this way.
BTW, thank you so much for this repo, it really helps me a lot. :)
according to the paper, HgWgDg is matched up with HxWxDx using trilinear interpolation, which is also called "Resampler" in the figure 2. but in the code of this repo, there is no this step, instead you use the feature map after the Up_conv step. So the code can be runned, But it is clearly different from what proposed in the figure 2. So I just wonder whether you implemented this way on purpose? And if yes, please enlighten me why it's better this way. BTW, thank you so much for this repo, it really helps me a lot. :)