mmrotate
mmrotate copied to clipboard
open-mmlab
The performances of Oriented RCNN in this version are not consistent with the official results.
Prerequisite
- [X] I have searched Issues and Discussions but cannot get the expected help.
- [X] I have read the FAQ documentation but cannot get the expected help.
- [X] The bug has not been fixed in the latest version (master) or latest version (1.x).
Task
I'm using the official example scripts/configs for the officially supported tasks/models/datasets.
Branch
1.x branch https://github.com/open-mmlab/mmrotate/tree/1.x
Environment
The test results of Oriented RCNN on DOTA1.0 with the trained model using the default setting are not consistent
with the official results. Please check the code of Oriented RCNN.
Reproduces the problem - code sample
oriented-rcnn-le90_r50_fpn_1x_dota.py train.py
Reproduces the problem - command or script
oriented-rcnn-le90_r50_fpn_1x_dota.py train.py
Reproduces the problem - error message
The test results of Oriented RCNN on DOTA1.0 with the trained model using the default setting are not consistent
with the official results. The differences between them are -0.56 and -1.15. The results of the official and mine are as follows:
Official:
Mine:
Additional information
No response
上图的MultiScaleFlipAug是0.3版本的配置，但是在1x中没有找到，不知道是不是这个原因。
应该不是这个原因。我是在1.x版本下复现的实验结果，使用自己训练的模型，测试结果与官方结果差距有点大，目前还没有找到与官方实现不一致的地方。不知道你用0.3版本复现的结果是多少。
我也是用的1x，使用自己训练的模型
跟官方也有差距
我对了一下数据增广部分的设置，flip和rotate的参数设置不一致，不知道这个影响是不是大
我单尺度结果没问题，你的呢
单尺度差0.56，也不太一致，加WX交流下？
请问你们解决问题了吗？加个wx交流一下吗？
请问你的问题解决了吗，加vx交流一下?
