search

chuanqi305 / MobileNet-SSD

Caffe implementation of Google MobileNet SSD detection network, with pretrained weights on VOC0712 and mAP=0.727.
MIT License
2.02k stars 1.18k forks source link

MobileNetv2 #67

Open fangyue6 opened 6 years ago

fangyue6 commented 6 years ago

Can you write the MobileNetv2-train.prototxt ？

didulle commented 6 years ago

Did anybody have tried to create MobileNetv2-SSD also called SSDLite?

chuanqi305 commented 6 years ago

@fangyue6 @didulle Now I‘m working on it, but currently the caffemodel file is not available. Let's talk about MobileNetv2-SSDLite.

didulle commented 6 years ago

Great!!

Robert-JunWang commented 6 years ago

You can try our Pelee model. It is easier to be trained and has a better precision than MobileNet+SSD series. The link is https://github.com/Robert-JunWang/Pelee.

ujsyehao commented 6 years ago

@Robert-JunWang Hi, I read your paper pelee and view it on GitHub. I have a problem, In README.md, it writes that SSD+mobilenet method gets 72.7 mAP by training on 07+12+coco dataset, but in chuanqi's mobilenet-ssd README.md, it says:

A caffe implementation of MobileNet-SSD detection network, with pretrained weights on VOC0712 and mAP=0.727.

He do not say that trained on 07+12+coco dataset, Is my understand wrong?