Justin Kerr

I think these changes look reasonable to me, delaying pose optimization until partial convergence makes sense. @jh-surh have you had a chance to take a look?

This is a good idea, but its quite a messy interface since there's so many mutable effects within the get_outputs. I wonder if we should just return them as a...

Interesting, does initializing gaussians randomly also work normally? Something to try would be launching training while paused, then inspect the initialized gaussians in the viewer to see if they line...

transform.json can sometimes automatically apply a dataparser transform to shift and scale the poses, which is saved as `'applied_transform'` in the json. Maybe if this pose is not identity, you...

Closing this for now, the version will be bumped/handled when https://github.com/nerfstudio-project/nerfstudio/pull/2888 is merged

It's possible you need to consider the dataparser transform too, which scales and shifts the poses to be centered at the origin with a scale of 1.

Cool! maybe we wait until the PR in the other repo is merged before merging this one?

Hi! gradients already propagate through depth when you use `rasterize_gaussians` with depths passed in as the "rgb" (as splatfacto already does, you would just need to render depth when `self.training`...

It would also be good to explore both metric losses, i.e. MSE with ground truth depth, and relative losses similar to [SparseNeRF](https://sparsenerf.github.io/), or [SparseGS](https://arxiv.org/abs/2312.00206). We've messed with these a bit...

There currently isn't great support for single image rendering, you can either - save the viewer canvas as a PNG (in viser, click the control knobs icon->export canvas) - run...