Jonas Schult comments

Results 21 comments of


                                            Jonas Schult

Cross Attention with attention_step1_v2 using differently sized key and query sets

Thanks for your hints! :) As it turned out, the training loss with `pointops.attention_step1` didn't decrease as well as my initial implementation using padding. So, it is possible that I...

how to do the experiment without RGB?

Hi Warsin! Thank you very much for being interested in our work. We released a wiki where you will find information about how to start the training/testing with or without...

Inference code used on https://mix3d-demo.nekrasov.dev/mask3d/

Hi Pablo, that is a great suggestion! I'll prepare an easy to use python interface soon. Best, Jonas

support multi-gpus ?

Hi! Right now, the codebase does not support multi-gpu training. It sounds like a great extension! :) Best, Jonas

How to prepare split the train and validation dataset of STPLS3D Dataset

Hi, thanks for checking this. I will add some more detailed instructions about STPLS3D soon. Best, Jonas

How to process the data for Scannet200?

Hi, please use the preprocessing script from our repository. See here: https://github.com/JonasSchult/Mask3D#scannet--scannet200 Best, Jonas

I have some question about the version of detectron2 hydra-core and omegaconf

Hi! I can confirm. I did it the same way like @lyzhenshuai and @Wenhua98 . Pip complains about this, however, this ensures that the right versions are installed. Best, Jonas

The performance of pretrained model on S3DIS (area1)

Hi, I just reran the inference script on a newly processed s3dis dataset. I could reproduce the claimed scores. Did you use the **non** aligned version of S3DIS? Moreover, Area...

The performance of pretrained model on S3DIS (area1)

Hi! Yes exactly please set `filter_out_instances=true` and `score_threshold=0.8`. We give some more explanations on page 5 in our paper. Best, Jonas

The performance of pretrained model on S3DIS (area1)

Hi! I used a single A40 GPU with 48GB RAM.