MobileNet-Caffe from https://github.com/shicai/MobileNet-Caffe scale: 0.017 crop_size: 224 mean_value: [103.94,116.78,123.68] test image "cat.jpg" index in synset.txt is 282 For mobilenet v1, "mobilenet_deploy_depthconv.prototxt" uses DepthwiseConv instead of group Convolution. For mobilenet v2, "mobilenet_v2_deploy_inplace_depthconv.prototxt" uses DepthwiseConv instead of group Convolution and adds layer inplace; re-train caffemodel as "mobilenet_v2_iter_7600.caffemodel".