Jiaming Han

Re backbone can be viewed as a single module, while RiRoI Align must works with Re backbone.

Note train.txt and test.txt in HRSC2016 are .txt files recording image names without extension. For example: ``` P00001 P00002 ... ```

Please check your `.txt` result file first.

please manually remove the `.bmp` extension in your `.txt` result file. We will update related code later.

win下编译，主要与vs和cuda有关，建议针对报错信息逐一搜索，如参考：https://www.jianshu.com/p/0c9542cef58a

猜测未完成安装。安装前需要运行：`bash compile.sh` https://github.com/csuhan/ReDet/blob/26167a451d690abdc1d1ba235f48a423e1d73b46/compile.sh#L75-L77

Maybe the mismatch between your `pytorch` and `cudatoolkit`. Please check it.

Maybe we use empty images for training. I'm not sure. Try to change the `difficult` key to `difficult=-1`. https://github.com/csuhan/ReDet/blob/f4a8e7dadec990f8a92437024416f0fb07a868db/DOTA_devkit/prepare_dota1_v2.py#L139 The multiscale result is not very stable since we use **random**...

Our code does not support online validation: ` --validate`. You can convert your data to DOTA's `txt` format, then you can train and test your model as DOTA.

ReDet supports cpu inference. Try to set `device='cpu'` in https://github.com/csuhan/ReDet/blob/0b9addf3c2734659fd6ffc7824f2e659fde4419c/demo_large_image.py#L91