Selecting training dataset

[ksmdnl]

Did anyone happen to learn the model only on VOC2012 dataset? If yes, how is the model performance learned with the default parameters?

As for me, it scores only 65% mAP for 55k iterations, which I think relatively poor. Looking forward to discussing this!

[Rook1e-li]

I have trained on VOC2007 dataset, and it scores 67%mAP for 15k iterations, the same with Fast R-CNN