LI Wentong
LI Wentong
Please try the blow suggestions: (1) Change the image size `(1024, 1024)` to `(960, 960)` for test config. This can work while it decreases the performance slightly. (2) Try to...
是的,我在训练DOTA多尺度数据(scale: 0.5, 1.0, 1.5)时, 也出现内存泄漏导致训练中止的情况,具体终端信息显示如下: `lib/python3.7/multiprocessing/semaphore_tracker.py:144: UserWarning: semaphore_tracker: There appear to be 14 leaked semaphores to clean up at shutd len(cache))` . 也知道是cuda代码的问题,但没有找到具体问题在哪。故训练出现中断的情况,一般采用`resume_from`继续训练,最后的结果似乎也没有受到影响。 感谢对此代码的完善以及commits !!!
你好,我是在ubuntu下验证的,win下应该是不行的。 我们即将将代码开源到[mmrotate](https://github.com/open-mmlab/mmrotate)框架下,这个框架应该是可以在win下验证的,你可以先尝试该框架下的其他方法,试试看。
版本要求: ` mmdet >= 2.22.0` | `mmcv-full>=1.5.3,
First, you can refer to this script [parse_pkl_for_evaluation.py](https://github.com/LiWentomng/OrientedRepPoints/blob/main/tools/parse_pkl/parse_pkl_for_evaluation.py) to parse the pkl file to get the standard text format. Then, you can run the [code](https://github.com/LiWentomng/OrientedRepPoints/blob/main/DOTA_devkit/dota_evaluation_task1.py) to get the mAP and...
是的, parse_pkl_for_evaluation.py生成的是results_raw, 对于dota results_raw的合并,请参考该脚本[dota_merge_results.py](https://github.com/LiWentomng/OrientedRepPoints/blob/main/tools/parse_pkl/dota_merge_results.py) 。
您好,该转换函数的代码为cuda代码,具体代码在这个位置[minarearect](https://github.com/LiWentomng/OrientedRepPoints/tree/35f4140347f4a00e382629f22756b0f783446127/mmdet/ops/minarearect),在head的脚本中调用了该函数。
目前的代码版本是mmdet1.1, mmcv的版本可以不局限于mmcv0.3.1, 后面会继续更新到最新的mmdet2.0.
Hello, you may need to check the format of the submitted `.zip` file. [Here](https://captain-whu.github.io/DOTA/submissionformat/example_task1.zip) provides the example with right format.
主要实现 体现在`orientedreppoints_head.py` 脚本中的 `def points_quality_assessment(self, pts_feats, cls_score, pts_pred_init, pts_pred_refine, label, obbox_gt, label_weight, obox_weight, pos_inds): `