Problem with eiq training ssd_mobilenet_v3 process

キャンセル
次の結果を表示 
表示  限定  | 次の代わりに検索 
もしかして: 

Problem with eiq training ssd_mobilenet_v3 process

3,271件の閲覧回数
hanyanze
Contributor I

When I used eiq for target detection training, I chose the ssd_mobilenet_v3 model under BASE MODEL. Since I trained three categories of boxes, the accuracy rate during the training process has always been the highest at 75%, which will not exceed this value. I don't understand, change It becomes 66.7% with the highest accuracy of the two categories. Do I need to set any background? But I didn't find the relevant settings.

At the same time, the accuracy and loss of the validation dataset are not normal. There is no problem with the dataset, I tried it in my own neural network.

微信图片_20220315165446.png

0 件の賞賛
4 返答(返信)

3,226件の閲覧回数
iris_wen
NXP Employee
NXP Employee

One update from R&D: this issue will be fixed in eIQ protal 2.4.8.

3,208件の閲覧回数
hanyanze
Contributor I

收到!

0 件の賞賛

3,257件の閲覧回数
david_piskula
NXP Employee
NXP Employee

Hello @hanyanze,

could you please provide me with the dataset and a list of the exact settings you use, so that I can attempt to reproduce your issue and report it to the development team in case it's a bug in the tool?

Also, please let me know the version of your eIQ Portal installation.

Best Regards,

David

0 件の賞賛

3,230件の閲覧回数
iris_wen
NXP Employee
NXP Employee

Hi @david_piskula ,

I have reproduced this issue and reported to R&D.