Open lzqsd opened 6 months ago
The current volsdf alpha value is computed as:
alpha = torch.abs(dists.detach()) * volsdf_density(sdf, inv_std)
I believe the actual formula should be:
alpha = 1 - torch.exp(-dists * volsdf_density(sdf, inv_std) )
Also it is weird to clamp the inv_std value to be smaller than 80. I guess the reason is to prevent the wrong alpha to be larger than 1? In other works, the inv_std for vol_sdf can goes to 1000.
Thanks and looking forward to hearing you reply!
The current volsdf alpha value is computed as:
alpha = torch.abs(dists.detach()) * volsdf_density(sdf, inv_std)
I believe the actual formula should be:
alpha = 1 - torch.exp(-dists * volsdf_density(sdf, inv_std) )
Also it is weird to clamp the inv_std value to be smaller than 80. I guess the reason is to prevent the wrong alpha to be larger than 1? In other works, the inv_std for vol_sdf can goes to 1000.
Thanks and looking forward to hearing you reply!