Training error

[HZXu-526]

Hi guys, I met a problem that says "Q = torch.quantile(grads_abs.reshape(-1), 1 - ratio), RuntimeError: quantile() input tensor is too large" when I was training the bicycle scene. It seems like the version of the function "density and prune" is different from the version in Feb.

[niujinshuchong]

Hi, yes, we added a new metric for densification proposed in gaussian opacity field https://github.com/autonomousvision/gaussian-opacity-fields. You can try to disable it by removing this line https://github.com/autonomousvision/mip-splatting/blob/main/scene/gaussian_model.py#L504 and this line https://github.com/autonomousvision/mip-splatting/blob/main/scene/gaussian_model.py#L480