Shape misalignment in Networks code

[Tijmenremijnse]

I read the paper with great interest and was testing out the code to potentially apply it on some own data. When running the networks.py code, an error occurs due to a misalignment in the Tensor sizes. The error occurs in line 297, but some further tracking of tensor sizes seems to indicate the issues starts when applying the LocalAttention module in during the Upblock in the Up_Projection_Unit (line 232). I have not been able to overcome the issue myself so I was wondering if there's a fix for this?

The error can be reproduced by simply running the networks.py file as the example code is already included.

[ciande-narbit]

这作者就是瞎写的，我参照PUGAN将LocalAttention 换成了attention_unit，同时去掉了pos参数，上采样后的点云根本不可能有对应pos嘛