Pytorch_Retinaface
Pytorch_Retinaface copied to clipboard
请问为什么仅仅在三个特征图上做检测,而不是原论文中的5个?
RT
A good result is obtained on three feature map.It's also to be consistent with the official version of mxnet.
官方源码发布的resnet50权重也仅仅是3个分支