Xiaoyang Wu

Hi, could you provide a more detailed train.log?

Hi, I checked the log, 1. It seem that background should be treated as a ignore segment? 2. Also to ensure that trainining augmentation is aligned with validation alignment, maybe...

Hi, I think you will get an answer after reading our preprocessing code for ScanNet as an example.

Hi, when flash_attention is disabled, please reduce patch_size to 128 or 256.

Hi, we only had a 100 during our experiments, but I think PTv3 is efficient enough to satisfy your request. Just try more, if you encounter some unexpected issue, tell...

It should be available soon!!! I also wish I could have more time! But CVPR is coming soon, too many things need to be finished before that.

> > Hi, when flash_attention is disabled, please reduce patch_size to 128 or 256. > > I was using two RTX 4090 datasets to train the semantics_kitti dataset (without using...

Hi, could you provide the config exported by Pointcept (in your exp folder) and more detailed error output for reference?

Hi, the discussion here might be helpful (https://discuss.pytorch.org/t/runtimeerror-unable-to-open-shared-memory-object-depending-on-the-model/116090). It might cause by system share memory limitation.

> but in the final Precise Evaluation, it got stuck for a long time Hi, is your custom point cloud very dense? Could you provide a more detailed test log...