Jian Ding

https://drive.google.com/drive/folders/1IsVLm7Yrwo18jcx0XjnCzFQQaf1WQEv8?usp=sharing for Google Drive or https://pan.baidu.com/s/1aPeoPaQ0BJTuCsGt_DrdmQ#list/path=%2F for Baidu Drive But currently, we only released the RoITransformer model.

In this pytorch repo, we only implement the rotated position sensitive roi align in https://github.com/dingjiansw101/AerialDetection/tree/master/mmdet/ops/psroi_align_rotated For the full light-head R-CNN OBB with rotated position sensitive roi align, you can refer...

You should transfer the xml files to txt files in DOTA format.

The --validate use the coco API for evaluation which is different from the DOTA evaluation API. Currently, we do not support the --validate.

Please see the demo https://github.com/dingjiansw101/AerialDetection/blob/master/demo_large_image.py

This code could be used for HRSC 2016. You need to write new configs.

It is transformed to coco format.

For evaluation on test set, you should refer to the evaluation server https://captain-whu.github.io/DOTA/evaluation.html. Also we provide the evaluation code in https://github.com/CAPTAIN-WHU/DOTA_devkit. You can use the code to evaluation in val...

Yes, the official model is trained on train+val set. The results are from the evaluation server.

论文中的数据都是用训练+验证集训练，然后在测试集上测试的。 你应该提交你的结果到dota测评服务器上测试。