search

ysh329 / tiny-deep-learning-models

fast, tiny deep learning models for mobile and embedded devices
GNU General Public License v3.0
2 stars 0 forks source link

squeezenet-ssd, mobilenet-ssd, mobilenetv2-ssd #1

Open ysh329 opened 6 years ago

ysh329 commented 6 years ago

chuanqi305/MobileNet-SSD: Caffe implementation of Google MobileNet SSD detection network, with pretrained weights on VOC0712 and mAP=0.727. https://github.com/chuanqi305/MobileNet-SSD

chuanqi305/MobileNetv2-SSDLite: Caffe implementation of SSD detection on MobileNetv2, converted from tensorflow. https://github.com/chuanqi305/MobileNetv2-SSDLite

chuanqi305/SqueezeNet-SSD: SSD detection network of SqueezeNet, with pretrained weights on VOC0712 and mAP=0.643 https://github.com/chuanqi305/SqueezeNet-SSD

ujsyehao commented 5 years ago

@ysh329 I train squeezenet-ssd without pretrained weights, and get mAP 64.4

ysh329 commented 5 years ago

@ujsyehao eh, great! 👍 What's your optimization method? Is over-fitting?

ujsyehao commented 5 years ago

just SGD, no over-fitting

ysh329 commented 5 years ago

@ujsyehao I think if pretrained model used, higher mAP and faster convergence are possible.

ujsyehao commented 5 years ago

@ysh329 yeah.