Xue Yang

A successful solution: set smaller `nms_pre` ``` test_cfg=dict( nms_pre=1000, min_bbox_size=0, score_thr=0.05, nms=dict(iou_thr=0.4), max_per_img=2000)) ```

Please run python `mmrotate/utils/collect_env.py` to collect necessary environment information and paste it here. @LiWentomng Any suggestions？

I guess it is because some sub-image in DOTA2.0 contain many objects, which causes some cuda operators to take up a lot of memory.

A successful solution: set smaller `nms_pre` ``` test_cfg=dict( nms_pre=1000, min_bbox_size=0, score_thr=0.05, nms=dict(iou_thr=0.4), max_per_img=2000)) ```

Thanks for your contribution. Please follow https://github.com/open-mmlab/mmcv/blob/master/CONTRIBUTING.md to set pre-commit hook and fix the lint error. It is best to provide a configuration file that works on this dataset like...

`reg_decoded_bbox=True,` should be added before `loss_bbox=dict(`

This branch has conflicts that must be resolved @JinYuannn

Can you give me a example?

Thanks for your contribution. Docstring is required in each function, and the relevant format can be found in other scripts.

An example: https://github.com/open-mmlab/mmrotate/blob/d80310aa2c4a67c877c62432cddee246d2a4dbcd/tools/analysis_tools/benchmark.py#L62-L77