Matthew Tancik

It seems like it is not using the GPU for some reason. The "Train Rays / Sec" should be ~1000x higher. Do the other models work fine (specifically nerfacto)?

Odd. I don't have a great answer for you then. It seems like the issue may lie in nerfacc. You could take a look at the repo and associated issues...

Feel free to merge when ready.

This looks like a colab related error - https://github.com/googlecolab/colabtools/issues/364 Some are suggesting using the upload button - https://stackoverflow.com/questions/53630073/google-colaboratory-import-data-stack-size-exceeded

Since there is a `spatial_distortion` in the nerfacto model, the scene box is never used. https://github.com/nerfstudio-project/nerfstudio/blob/009e21be248418028e6303ecde593a20ef88fa87/nerfstudio/fields/nerfacto_field.py#L199-L204 It is worth noting that the nerfacto model isn't really setup to work with...

Which model are you running? Are you seeing this on all scenes, or just the blender scenes?

For lego, try also adding the following arguments `--pipeline.model.near-plane 2. --pipeline.model.far-plane 6.`

@snarb Do you have some data that you can share that reliably produces this error? Can you try changing the `x.clamp(-15,15)` to smaller values like `x.clamp(-12,12)`

Good find @XinyueZ. Do you find that these weight-decays effect the training speed or quality in the first 30K iters? If now, we should consider changing making these the default.

No longer relevant