use resnet50 as the backbone training COCO, the map is only 42.5，cannot reach 46.4

[wssywh]

The configs file used is pycenternet_r50_fpn_3_levels_giou_fast_mstrain_6x_coco.py，others haven't changed anything，the result of map is only 42.5，why？？？？？

[Duankaiwen]

@wssywh use pycenternet_r50_fpn_3_levels_giou_fast_adamw_mstrain_6x_coco.py instead

[wssywh]

@Duankaiwen use pycenternet_r50_fpn_3_levels_giou_fast_adamw_mstrain_6x_coco.py,the final map is 42.9

[Duankaiwen]

@wssywh can i see your training log?

[wssywh]

@Duankaiwen this is the training log : https://docs.qq.com/document/DVVpNZXdpT1RkZmdL

[Duankaiwen]

@wssywh I see, to obtain the 46.4 AP, you should use pycenternet_r50_fpn_giou_mstrain_2x_coco.py. The pycenternet_r50_fpn_3_levels_giou_fast_adamw_mstrain_6x_coco.py is for CenetrNert-RT, which achieves 43.2 AP in my machine.