oom error

[Im-JimmyHu]

my device is win 10 with rtx3060(12gb) when i try to train with configs/oriented_rcnn/oriented_rcnn_r50_fpn_1x_dota_le90.py and dota v1.5 ,even i set the batch-szie to 1,it still run oom after 1000steps. after reading the faq,i set the with_cp=true.,but it not help .the other 2 parm like gpu_assign_th and fp16, i don't know where to set. can you give me some advice?

besides when i turn to hrsc, i find a typo and fix it https://github.com/open-mmlab/mmrotate/blob/c62f148fcc2c8253218e67f2277cb8770bcd7df0/mmrotate/datasets/hrsc.py#L98 you may need to remove the self.img_prefix,or you will find the path error.

[zytx121]

You can use fp16 by adding one line command to your config: https://github.com/open-mmlab/mmrotate/blob/c62f148fcc2c8253218e67f2277cb8770bcd7df0/configs/redet/redet_re50_refpn_fp16_1x_dota_le90.py#L3

Or, you can use
https://github.com/open-mmlab/mmrotate/blob/c62f148fcc2c8253218e67f2277cb8770bcd7df0/configs/redet/redet_re50_refpn_1x_dota_ms_rr_le90.py#L40

[Im-JimmyHu]

thanks,it works. and the typo i issue maybe you should check.besides, the the link of useful-tool in get-started is depreated,please remove the tutorial in the link.

[zytx121]

Thanks! I will check it.