question about r-fcn caffe

[msukoz]

Thanks for sharing.

Now I am training faster rcnn+resnet50 using matlab interface. The caffe in faster rcnn is from r-fcn, but there is something wrong with the experiment. Faster rcnn their own caffe dont have bn/scale layers, so I use caffe from r-fcn, but the result is strange, like error_fg = 1, error_bg=0 from very beginning. So I tried faster rcnn(caffe from r-fcn)+vgg16, the same question. While, it is normal in faster rcnn(its own caffe) +vgg16.

Can I get some suggestions?
Thanks