Overfitting was observed in the Synthetic v3 region

[mooncakehub]

I used Pointct ，a net like Point transnformer ,I found that when I used Synthetic v3 as a test aera ,training losses are declining and validation losses are rising,it seems overfitting.I used Synthetic v1 as test aera,it seems normal.Is this a problem with the code or a problem with the dataset。

Thanks！