search

sniklaus / softmax-splatting

an implementation of softmax splatting for differentiable forward warping using PyTorch
469 stars 58 forks source link

Why use torch.cat([tenEncone[intLevel], tenMetricone] as softsplat input tenIn #63

Closed Candy-Crusher closed 1 month ago

Candy-Crusher commented 1 month ago
1729073895277

Hi dear authors,

Thanks for the amazing work. I have a confusion about softsplat input. According to my understanding, the tenIn shoud be image features solely. But in the code, I notice that you also concate tenMetricone. May I ask the reason behind it?

sniklaus commented 1 month ago

Thank you for the kind words! It has been a while, but if I remember correctly then providing this metric to the synthesis network slightly improves the interpolation quality. So it doesn't have anything to do with the softmax splattig itself, just with providing more information to the synthesis network. Hope that helps!