PyTorch-YOLOv3 icon indicating copy to clipboard operation
PyTorch-YOLOv3 copied to clipboard

Published 20 hours ago • verified publisher icon eriklindernoren

reproduce the mAP in readme

Open yanxurui opened this issue 5 years ago • 11 comments

How specifically can I reproduce the mAP results mentioned in readme?

I have run test.py using the weights from the original darnet but only got mAP of 0.5145. How can I get mAP around 55? What's the mAP using the weights trained by this implementation?

Thanks.

yanxurui avatar Jun 16 '19 11:06 yanxurui

Same, waiting for answer

CodePothunter avatar Jun 17 '19 12:06 CodePothunter

waiting for answer

ShuaiZ1037 avatar Jul 06 '19 11:07 ShuaiZ1037

I meet the same problem, waitting the answer. The #164 has some disscusion.

SWHL avatar Jul 18 '19 09:07 SWHL

@SWHL I changed the batchsize intest.py and change the cfg.file(batch=1,subdivisions=1),but it didn't work.I also get 51.45.

yyx1107 avatar Jul 24 '19 06:07 yyx1107

Can you help us?@eriklindernoren

yyx1107 avatar Jul 24 '19 06:07 yyx1107

Any pointers on this @eriklindernoren ?

mvcaro avatar Aug 08 '19 12:08 mvcaro

Same mAP: 0.5145

talenz avatar Aug 12 '19 07:08 talenz

Same mAP: 0.5145

betterhalfwzm avatar Aug 14 '19 07:08 betterhalfwzm

same mAP: 0.5145

cjnjuwhy avatar Oct 17 '19 14:10 cjnjuwhy

same mAP: 0.5145

BaiqiangGit avatar Oct 25 '19 11:10 BaiqiangGit

There is a bug in reading Detection label and larger batch makes the problem worse. See https://github.com/eriklindernoren/PyTorch-YOLOv3/issues/243

dhkwon1122 avatar Jan 01 '20 23:01 dhkwon1122