L2_normalization layer instead of sigmoid (in Contrastive loss)

c100rczyk / Team_project_PWR

This is team project implemented as part of the IT specialization in the field of Automation and Robotics, in cooperate with company. Main technologies are: Computer Vision. Language: Python.

MIT License

2 stars 0 forks source link

L2_normalization layer instead of sigmoid (in Contrastive loss) #24

Closed c100rczyk closed 6 months ago

c100rczyk commented 6 months ago

https://stackoverflow.com/questions/59596162/how-to-apply-l2-normalization-to-a-layer-in-keras

c100rczyk commented 6 months ago

With BatchNormalization + Sigmoid , layers the accuracy(top1) was 82% and top5 accuracy was 97.7%

c100rczyk commented 6 months ago

Deleted sigmoid. Added L2 normalization. Model have difficulty to learn. (Earlier , befor change , it was same problem sometimes). But after get Model with Transfer Learning , model learns good , and everything is ok