training on voc2007+2012，result is very poor

[H-YunHui]

How did you solve it？

[yyx1107]

How many epochs did you train? And what's the mAP?

[H-YunHui]

epochs=100，map=0.351 I think the epochs chose too few and models did not converge. And you?

[abhiksark]

This is happening with me as well. And also i did not train for all the classes i only trained for Person Class

[H-YunHui]

@abhiksark If you only trained for Person Class,what's you mAP and epochs?

[abhiksark]

10 Epochs mAP was 0.26

[onefish51]

权重	数据集	mAP
weights/yolov3.weights	coco	1.6121330694239914e-07
checkpoints_coco_person/yolov3_ckpt_72.pth	coco_person	0.6085335843614279
checkpoints_coco_person2/yolov3_ckpt_66.pth	coco_person	0.5898045606894886
checkpoints/yolov3_ckpt_27.pth	caltech	0.21640950928418473
checkpoints_caltech/yolov3_ckpt_29.pth	caltech	0.22963548969096598

I also train on coco for person class and caltech pedestrain detection ,after serveral epochs ,I test the weights, the result is very poor! I did not know the reason!

[zhongxing9006]

@onefish51 Do you solve this problem? I met the same problem.

[heragong]

same problem here. train on coco, 100 epochs, 0.2 mAP...

[ardeal]

@abhiksark @zhongxing9006 @onefish51 @H-YunHui @yyx1107

have you solved the problem? Please see my issue #504 I am experiencing the same problem. I just clone the code and prepare COCO. nothing is changed in the code. I have trained more than 150 epochs, but the mAP is around 0.28.