Danila Rukhovich
Danila Rukhovich
You can try to sample point with fixed random seed, and then use the same seed to sample annotation points during metric calculation.
Is it fine when training on all 6 datasets? Also I noticed that you modified learning rate. Not sure if it is a good idea.
Yes, I think should be fine. The released model is trained on 6 datasets, but can be tested on either all 6, each subset of them, or a new one....
Is it ok that your `test_evaluator` has different lengths of `datasets` and `datasets_classes` in line 766?
1. Yes, looks like something is missing in our preprocessing code. Did you try to download our preprocessed data from [huggingface](https://huggingface.co/datasets/maksimko123/UniDet3D/tree/main)? 2. Yes, should be the same.
This means mmcv is not properly built for your GPU. Can you check if TORCH_CUDA_ARCH_LIST [here](https://github.com/filaPro/unidet3d/blob/master/Dockerfile#L26C8-L26C28) needs to be updated for your GPU?
Our checkpoint in corresponding to the 6-dataset config. Please use 6-dataset config and comment non-necessary datasets from validation.
Hi. Don't remember such an error. Did you try with out provided checkpoint?
Looks like some issue of torch_scatter. Are using the same version as in our dockerfile?
should be around 1 day on a single GPU