reproduction of matchNet using pytorch, since the original implementation is based on caffe and some newer reproductions are based on tensorflow-keras, I reproduce matchNet using pytorch to ease my work
I provide different feature net usiing mobileNetV3' block, in experiments, feature net constructed by mobileNetV3 block has better training efficiency and performace.
Notice that, I use accuracy( correct predictions / total predictions ) instead of FPR@95recall
Thanks the author who provide the mobileNetV3 model implementation at https://github.com/d-li14/mobilenetv3.pytorch