search

tanakataiki / ssd_kerasV2

BetterNetWork4SSD
MIT License
144 stars 56 forks source link

Training #1

Open tanakataiki opened 6 years ago

tanakataiki commented 6 years ago

As I customed ssd for relatively log time. I found it pretty hard to train by means of hyper parameter setting and sensitivity of architecture. If your have environment to train then Upload your weights is very welcome let's make it better.

algila commented 5 years ago

Hi, may I recomment the users uploading their weights in folder "YourFineTunedWeights" to communicate also the dataset used doing training ? It should be beneficial for the other users doing Transfer learning with those.

tanakataiki commented 5 years ago

It's good idea.

E12dward commented 5 years ago

hi,i just train my model SSD300shufflnetV2 by following your strategy in SSD_train.ipynb ,but it seems hard to converge to a low loss (loss=4) . is there something wrong? thanks for your help