The training of bevfusion only modifies the POINT-CLOUD-RANGE algorithm, resulting in significant differences in effectiveness

[zzwforever]

May I ask for your advice? When training the bevfusion model with the same data and configuration, the algorithm performs well when the XY value of POINT-CLOUD-RANGE is the same, but when the XY value is twice that of X or the difference between XY is significant, the algorithm performs worse, only changing the value of XY. What is the reason for this