The performance of rotated_reppoints_r50_fpn_1x_dota_oc model.

[najingligong1111]

In the paper "Oriented RepPoints for Aerial Object Detection Results on DOTA" , we report full experimental results of single scale to make a fair comparison with the previous methods on DOTA dataset. With ResNet-50-FPN and ResNet-101-FPN as the backbones, our method obtains 75.97% and 76.52% mAP, respectively. The result of "rotated_reppoints_r50_fpn_1x_dota_oc " model is 59.44 , why there is such a big gap?