search

ShuangXieIrene / ssds.pytorch

Repository for Single Shot MultiBox Detector and its variants, implemented with pytorch, python3.
MIT License
571 stars 167 forks source link

How to set the parameters of the pre-training model #65

Open Damon2019 opened 4 years ago

Damon2019 commented 4 years ago

i want to use pre-training model

TRAINABLE_SCOPE: 'base,norm,extras,loc,conf' RESUME_SCOPE: 'base,norm,extras,loc,conf' TRAINABLE_SCOPE and RESUME_SCOPE , How should I modify or Which parameters should I remove?
thanks

foreverYoungGitHub commented 4 years ago

if you want to train for a new dataset, you can freeze the base,norm,extras first and then fine-tuning the whole parameter. If you want to train the network from the imageNet pretrain model, only freeze the base part.

Damon2019 commented 4 years ago

i want to train for a new dataset 。Is my parameter setting like this?As follows：

TRAINABLE_SCOPE: 'base,norm,extras,loc,conf' RESUME_SCOPE: 'base,norm,extras' I am a newbie,I hope that you can give me detailed advice.

2017TJM commented 4 years ago

where can i get the mobilenet_v2_rfb_lite_voc_73.4.pth' and other **.pth files

gengpengs commented 4 years ago

在哪里可以获取mobilenet_v2_rfb_lite_voc_73.4.pth'和其他**。pth文件

have you got it ?

foreverYoungGitHub commented 3 years ago

Would you mind attaching the whole config file?

The relative pretrained weights in the master branch are stored in the google drive. You can find the relative links at here https://github.com/ShuangXieIrene/ssds.pytorch#performance.