Haoxi Ran

Thanks for your interest in our work! Since the RepSurf codebase for the task of obejct detection borrowed heavily from the work Group-Free 3D Object Detection via Transformers, we have...

Hi, we will try our best to release the code of object detection. For training settings, we followed the work of GroupFree3D. For more details of RepSurf on the backbone,...

Sorry for the late reply. We have uploaded the implementation on segmentation tasks. And if you are interested in the repsurf model on modelnet, we will plan to upload it...

If necessary, we will publish it with the pretrained weights soon. However, if you can't wait for it, it could be quite simple to implement the experiment on scannet, as...

We use 4 gpus for training here.

Can you provide some code snippets for this problem? Do you use the cls's UmbrellaSurfaceConstructor or the seg's? I may consider a reason that if you do not use cuda-ops...

Can you provide more details while debugging? I've never come across anything like this. The difference between the training/validation parts is the evaluation code starting [here](https://github.com/hancyran/RepSurf/blob/29bacd52fe61ffb034c076be13163d415de6684a/segmentation/tool/train.py#L380). I recommend debugging here...

Thank you for the detailed info. Here are two things I think that may cause this prblem: 1. #workers for validation dataloader. The default value you have set to is...

你好， 我在3090上尝试过，可以跑的。你可以尝试改[这一行](https://github.com/hancyran/RepSurf/blob/main/segmentation/scripts/s3dis/train_pointnet2.sh#L11)去实现单卡的运行。感谢你的关注。 Hi, I have tried the implementation on 3090, and it works as well. You could try changing [this line](https://github.com/hancyran/RepSurf/blob/main/segmentation/scripts/s3dis/train_pointnet2.sh#L11) in the script for segmentation, and so as for...

I am not familiar with ISPRS dataset, but I think you can follow some common protocals: to scale the batch size by #gpus, e.g., 8 for 4 gpus and 2...