why I just get mAP = 0.48 using the provided trained model on val dataset of DOTA?

[clw5180]

I use 108000.ckpt, however I only got total mAP of 0.48 on DOTA val dataset. I didn't change anything of cfg parameters. I use the official DOTA_devkit to eval, it seems that some classes doesn't work well, like small vehicle only got 0.2 mAP.

[yangxue0827]

what about test dataset. @clw5180

[clw5180]

what about test dataset. @clw5180

I don't know, because I don't have edu email to register an account.

[yangxue0827]

You can send it to my mail (yangxue0827@126.com) and I'll give you feedback on the results. @clw5180

[clw5180]

You can send it to my mail (yangxue0827@126.com) and I'll give you feedback on the results. @clw5180

感谢。

[bezero]

@yangxue0827 I am having similar issue. Using pre-trained wights provided, when I run the eval.py, results are as follows:

cls : storage-tank|| Recall: 0.5130154639175257 || Precison: 0.015452273009564029|| AP: 0.48442155289713024
cls : harbor|| Recall: 0.6670716889428918 || Precison: 0.0134925901349259|| AP: 0.5672587986511316
cls : basketball-court|| Recall: 0.5392670157068062 || Precison: 0.0007658704340552096|| AP: 0.42400440861403477
cls : large-vehicle|| Recall: 0.520392244261199 || Precison: 0.020670119063426726|| AP: 0.39298557782405563
cls : tennis-court|| Recall: 0.8924731182795699 || Precison: 0.0063206792826409775|| AP: 0.876200854491652
cls : baseball-diamond|| Recall: 0.7804878048780488 || Precison: 0.0016594784743169777|| AP: 0.6900118677005387
cls : ground-track-field|| Recall: 0.7468982630272953 || Precison: 0.001239060454624042|| AP: 0.5572717103613503
cls : swimming-pool|| Recall: 0.5556426332288401 || Precison: 0.002510134357685295|| AP: 0.4978325221870478
cls : helicopter|| Recall: 0.7095238095238096 || Precison: 0.0005679392571811917|| AP: 0.43784103066833346
cls : ship|| Recall: 0.5034660187236475 || Precison: 0.044909797921846115|| AP: 0.4042568479254526
cls : plane|| Recall: 0.8582983822648292 || Precison: 0.022448667771470213|| AP: 0.8423056840439651
cls : bridge|| Recall: 0.3673139158576052 || Precison: 0.0013061440556060623|| AP: 0.24704939710411028
cls : small-vehicle|| Recall: 0.10733287523314028 || Precison: 0.033817993993548165|| AP: 0.0835517742743227
cls : soccer-ball-field|| Recall: 0.6516290726817042 || Precison: 0.0010524270587092387|| AP: 0.4330316144057434
cls : roundabout|| Recall: 0.5352112676056338 || Precison: 0.0010427051796837127|| AP: 0.37644754809753556
mAP is : 0.48763141261642695

especially results for small-vehicle are really low with AP: 0.084