Closed RoyAmoyal closed 3 months ago
train_joint.py
train_joint.py
It's the exactly same Loss as the original Gaussian Splatting. I don't see where they optimize the camera poses (extrinsic's of the cameras, T's in their equation) along with the 3DGS. It requires additional code that I am probably missing..
Hey, thanks for sharing the code for your cool work!
I can't find the code section related to equation number 6, where you jointly optimize the gaussians together with the camera parameters. (3.3. Joint Optimization for Alignment)
I can only find the original training of Gaussian Splatting in train_joint.py that optimizes the gaussians like in the original 3DGS, but not the camera poses. (I can't find the difference in the cuda backward section too)
thanks!