Need help with darknet-mobilenet?
Click the “chat” button below for chat support from the developer who created it, or find similar developers for support.

About the developer

123 Stars 35 Forks 11 Commits 6 Opened issues


mobilenet model in darknet framework , MobilenetYOLO, compress mobilenet

Services available


Need anything else?

Contributors list

# 645,278
11 commits


The mobilenet model of Google's mobileNets in darknet framework. The official paper: MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

I provide a cfg file of mobilenet and a pretrained mobilenet weights on ImageNet.

This model achieved top-1 accuracy 71.1% and top-5 accuracy 90.5% (image size:448x448)

Note: You want to use this model in darknet, you must wirte a group conv layer.

mobilenetyolococo:(approximationly 0.5:0.95mAP 16.2%) 链接: 密码: 4qr8 mobilenetyolovoc:(According to loss value, I guess mAP 0.5mAP 65%-70%) 链接: 密码: qk4q

By visualing the weights of mobilenet, we found that some filters have a weight of zero. We compress the mobilenet with thresholds of 0.0001 and 0.08, respectively. So the model with thresholds 0.0001 achived the same accuracy as original mobilenet (top-1: 0.712540, top-5: 0.903520), and the other model with thresholds 0.08 achived accuracy with top-1 0.682580 and top-5 0.885100. All the compress model without finetune.


Compress mobilenet with threshold of 0.1, then finetune it for 20 epochs, finally this network achived accuracy with top-1 0.697 and top-5 0.897, and the size of this model is 10.1M (18M->10M).

We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.