Xin Lai

Thanks for your interest in our work. I don't really understand your problem. Can you clarify your question more clearly?

Do you meet the problem only in validation? If yes, I recommend you to adjust the ```batch_size``` in val_loader to be the same as ```n_gpu```. Since during validation, the size...

Which results do you want to visualize?

The `pointops_api.cpp` is in the `src/` directory. Can you try this: `cd /content/Stratified-Transformer/lib/pointops2/ && python setup.py install` ?

You environment may lack THC.h. I haven't met that before, so you can search which package contain it and install it.

which version do you install?

Can you try to reinstall with the alternative cammands? ``` pip install torch_sparse==0.6.12 pip install torch_points3d==1.3.0 pip install tensorboard timm termcolor tensorboardX ```

I haven't met that before. May I ask which version of numpy do you use?

Thanks for your interest in our work. Yes, `pointops.attention_step1_v2` is much faster than `pointops.attention_step1`. To use `pointops.attention_step1_v2`, you need to make sure that the `index_0` is in ascending order, and...

It seems like the gcc and g++ have some errors.