Xiaoyang Wu

Yes. So increase in parameters does not make the model slower and does not make the memory cost larger. So in my opinion, it is worth introducing it (also discussed...

This is an extremely rare issue and is usually caused by different reasons. It may caused by local code env, data issues, or some parameter tuning. So my suggestion is...

> I finally found the error in my code! It seems like pointops works by default with cuda:0, which caused that I had some tensors on this gpu, while others...

Hi, Readme here might be helpful (https://github.com/Pointcept/Pointcept?tab=readme-ov-file#offset) The point structure is introduced to organize the components of a batched point cloud. Assume we have a batched point cloud containing 3...

Hi, try this modification:

> Sorry to bother you with my problems again. Thank you for your answer. I have been dealing with the problem for several days, so I thought it would be...

Hi, releasing config for semantic_kitti is in our schedule. Typically I release a config after verifying the performance. Yet, recently, I have been fully devoted to a new research project,...

Hi, releasing config for semantic_kitti is in our schedule. Typically I release a config after verifying the performance. Yet, recently, I have been fully devoted to a new research project,...

Hi, the constant time and memory, regardless of patch size, sourced from FlashAttention, which fully utilized the L1 and L2 cache of GPU to operate attention. More details are available...

The meaning of sequence length in NLP means token number feed to network each forward. It is close to the concept of numbers of points in 3D point clouds. It...