search

lizhe00 / AnimatableGaussians

Code of [CVPR 2024] "Animatable Gaussians: Learning Pose-dependent Gaussian Maps for High-fidelity Human Avatar Modeling"
https://animatable-gaussians.github.io/
Other
887 stars 58 forks source link

question about cano_weight_volume #44

Open liwenssss opened 1 month ago

liwenssss commented 1 month ago

I followed the process in gen_weight_volume.py to generate cano_weight_volume, and I referenced the code at https://github.com/lizhe00/AnimatableGaussians/blob/77992646a9112fc04e34acf5aa9bf6748ae91143/gen_data/gen_weight_volume.py#L179 to compare pts_w_query with pts_w_val. I found that some values have discrepancies reaching the order of 0.1. Is this normal, and how significant is the impact on the final results? Additionally, in your process, do you not optimize cano_weight_volume, but rather directly use the weight values from cano_weight_volume?

lizhe00 commented 4 weeks ago

The error is acceptable because the solver cannot guarantee exactly the same LBS values of the sampled points. We do not optimize the LBS weight volume.