Ze Liu

Hi @cuge1995 , I'm not familiar with the `Slurm Workload Manager`. But as the error message says, can you check the `CUDA_HOME` in your environment and maybe it is not...

Hi @a18700 , We have not tried this scenario, but in our implementation, you could set a appropriate `sampleDl` to first sub-sample the whole point cloud to a proper scale....

Sorry for my delay reply and thanks for your interest in our work. The problem may be that the GPU memory is not enough. Can you reduce the batch size...

Hi @cherry-ing. That's weird, can you tell me your environment like python, CUDA and pytorch version?

Hi @MingyeXu, This may cause by the lack of GPU memory, can you reduce the batch size and try again?

Hi @missbook520 , I'm sorry that the checkpoint provided here is not correspond to the latest code(we change all LeakyReLU to ReLU in the latest code), I'll re-train all models...

Hi @cuge1995, that's strange, can you show me more information about the error and your environment?

Hi @SilvioGiancola Thanks for your interest in our work! Of course, we will sharing the PartNet models soon and we are still working on it, since the training has not...

Hi @SilvioGiancola, We have released the pytorch models for PartNet. As for the problem you mentioned, may be the learning rate is too large, since in our initial implementation, the...

Hi @Luodian, the pretrained models can be found in: https://github.com/microsoft/Swin-Transformer/blob/main/MODELHUB.md#imagenet-22k-pretrained-swin-moe-models. We have also provided an instruction on how to run Swin-MoE, which can be found in: https://github.com/microsoft/Swin-Transformer/blob/main/get_started.md#mixture-of-experts-support