search

open-mmlab / mmrotate

OpenMMLab Rotated Object Detection Toolbox and Benchmark
https://mmrotate.readthedocs.io/en/latest/
Apache License 2.0
1.88k stars 558 forks source link

When training the r3det on SSDD, the precision is too low. #984

Open xiaxiangyu opened 10 months ago

xiaxiangyu commented 10 months ago

Branch

master branch https://mmrotate.readthedocs.io/en/latest/

📚 The doc issue

When I was using r3det, I only modified the num_class and img_scale of the config file. The data set used was SSDD, and only one class of ship was used. The recall of the test result was 0.944 and 0.933, but the precision was only 0.485. What is the reason? Do I get the wrong precision? I modified the eval_map.py accoring to issue 534. Hope to get the precison value! code

Suggest a potential alternative/fix

No response

Royalbx commented 9 months ago

I train r3det with dataset SSDD, the recall of the test result was 0.88, but the precision was only 0.28.why? I modified the image scale is (640,640), 50epoch, milestones=[32, 44]

xiaxiangyu commented 8 months ago

I train r3det with dataset SSDD, the recall of the test result was 0.88, but the precision was only 0.28.why? I modified the image scale is (640,640), 50epoch, milestones=[32, 44]

How is this problem. Can you tell me how to address this issue.

exesit commented 6 months ago

我也存在这种问题，复现与论文差很多