Xiaoyang Wu

Hi, I didn't try to run our codebase directly on a window server (with wsl2 ubuntu), yet here are some solutions to which you can refer: > Use wsl2 ubuntu...

Hi, I think preprocessing your custom data to the same format as one of the datasets supported by Pointcept and following the readme for the testing process is a good...

Because Neighborhood Attention still needs KNN to determine the kernel. This is not for the pooling layer.

> @Gofinge Is the lidar beam lines effect the results a lot? My own data is collected by an 128-beam lidar Yes, this is one known factor that avoids models...

Yes. And in our PTv3, KNN is fully removed from the pipeline.

I will push a DefaultClassifier in next version.

Hi, discussion here might be helpful (https://github.com/Pointcept/Pointcept/issues/108#issuecomment-1893709522). We do have a solution to these large point clouds.

Hi, thanks for your effort in running PTv3 on the CPU. I think disabled bias won't have a negative influence on the model performance. Looking for your further feedback on...

> When I run test on multi gpu I have less samples then exists. In my test set I have 7 samples, also I see this in the logs when...

> Do you have similar issue like this: #97 > > Check your evaluation log whether it is completed or not. > > In my case, validation and test using...