A caffe implementation of mobilenetv2_ssd convolution network.
1.The mAP for mobilenetv2_ssd on my own dataset is less than the mAP for mobilenetv1_ssd, I don't kown why?
2.My mAP result for mobilenetv1_ssd is a reference to https://github.com/chuanqi305/MobileNet-SSD.
3.If you want to speed up the mobilenet network,you can reference to https://github.com/yonghenglh6/DepthwiseConvolution and change the Group Gonvolution Layer to Depthwise Convolutional Layer.