nerfstudio-project
Adjustable camera positions
Is there a reason why nerf.studio, instant-ngp, dust3r etc don't allow users to manually correct the camera positions in the software itself visually? To me it seems like this would make sense for refinement? But at the moment we have to manually adjust a transforms.json file every time which just seems absurd.
Hey I just think that these kind of features are not supported in the visualization tools provided by these software and implementing this kind of user adjustments is not trivial. The tools built around nerfstudio are meant for visualizing the training and reconstruction of nerfs/3dgs scenes in real time, and not for editing point clouds, camera positions, or other attributes of the training data. Also, pose refinement is treated more as an additional "learning stage" which should not require manual adjustment, but I understand this kind of feature could be useful. I think improving the visualization tools is a must, but it is quite a challenge and you need good knowledge of this kind of development, and viser is already quite successful trying to solve some of these issues, but its development group is very small compared to dedicated software like blender etc who really focus on the manual editing side of things.
I imagine it is possible to adjust the positions using the current GUI. All the attributes of the cameras are indeed exposed to python and viser, so adding some toggles etc sounds feasible. I just think that no-one is currently looking into this.