You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi, I've tested on bicycle scene of mipnerf 360, using the configs in the README.MD file. But the gaussians seemed not able to fit the scene. The loss will not decrease throughout the training. And there're not any increment on the gaussian numbers.
The text was updated successfully, but these errors were encountered:
Hi, thanks for your interest in LPM. How is the rendered image? Is it still in a blurred state? The reason for the Gaussian not increasing is that we are performing point calibration based on the multi-view prior, which resets some low-confidence Gaussian points. While this approach limits Gaussian growth, it does come at the cost of some performance compromise.
Hi, I've tested on bicycle scene of mipnerf 360, using the configs in the README.MD file. But the gaussians seemed not able to fit the scene. The loss will not decrease throughout the training. And there're not any increment on the gaussian numbers.
The text was updated successfully, but these errors were encountered: