suzhenghang / mobilenetv2

MobileNetv2

Run the script benchmark_mobilenetv2.sh and you can get the result: top1/top5: 0.7123/0.9018

You can use this model to do a lot of things such as training a smaller mobilenetv2 (By moving params or knowledge distillation)

Small MobileNetV1/V2 is friendly used on a mobile device (U-net for semanic segmentation and refinedet for object detection)

Training details for ImageNet2012 : type: "SGD" lr_policy: "poly" base_lr: 0.045 power: 1 momentum: 0.9 weight_decay: 0.00004

The pytorch version: https://github.com/suzhenghang/MobileNetV2_Pytorch

The prototxt of SmallMobileUnet: https://github.com/suzhenghang/MobileNetv2/blob/master/MobileUnet1by2.prototxt

The prototxt of SmallRefineDet_mobilev2_ssdlite: https://github.com/suzhenghang/MobileNetv2/blob/master/SmallRefineDet_mobilev2_ssdlite.prototxt

mobilenetv2's Issues

About：SmallRefineDet_mobilev2_ssdlite.prototxt

您好，在您的refinedet_mobilenetv2中,最后的P6、P5网络层中为什么num_output=6class_num？而在标准的refinedet中num_output=3class_num.

How long for forwarding?(224x224 image)

Hi, can you share your time for forwarding of a 224x224 image? And is it as fast as the speed reported in the mobilev2 paper.
I have train this model with caffe, but the speed is lower than mobilenetv1, also the paper.

There is a typo in your prototxt file

MobileNetV2_deploy.prototxt
line63 : name: "relu1_new"
line3196: name: "relu1_new"

Two layers are using the same name.

Unknown layer type: DepthwiseConvolution

Hello, which version of your caffe version is used, the original version of refinedet does not support depthwise.

What is your accuracy and ask for caffemodel

Hi
Can you share your accuracy or even caffemodel?
When I run shufflenet, the utilization rate of the computer's hard disk often reaches 100%. Have you ever met this situation?
Thanks!

hi, when i train your seg .there is a ..

Could you please share your training configuration file?

Thanks for your sharing your model. I also trained mobilenetv2 by myself. But unfortunately I can only get 70.9% top-1 on imagenet 2012. I will be very appreciated if you could share your training configuration to help me improve my training. Thanks :)

How can we use it?

Could you provide the instructions? I am a new one in deep learning domain.I donnot know how can I run it at VS