Danila Rukhovich
Danila Rukhovich
hard to say anything specific... are you training on all our 6 datasets, downloaded from huggingface? do all python package versions match our dockerfile?
I believe training is quite noisy and the difference between last checkpoints is within statistical error.
I think our vizualization produces .obj files, that we recommend to load in meshlab. But you can load them in open3d.
Please follow installation process from our Dockerfile including this [line](https://github.com/filaPro/oneformer3d/blob/main/Dockerfile#L29) to install `segmentor`.
Unfortunately, there is no single best checkpoint due to randomness in training and inference. However, the provided checkpoint achieves almost all *maximum* metrics reported in the paper, and even outperforming...
This looks like some bug with your installation. The provided checkpoint should reproduce metrics from our readme.
Yes should be fine. Can you privide the complete log of your test run?
Looks like package versions (torch, spconv, cuda) don't match our Dockerfile. Can you try with exact same versions or run it in docker?
Our checkpoint in corresponding to the 6-dataset config. Please use 6-dataset config and comment non-necessary datasets from validation.
Please check comment about RAM memory during evaluation in our README.