search

kevinlin311tw / caffe-cvprw15

:heart::coffee: Deep Learning of Binary Hash Codes for Fast Image Retrieval (CVPRW15)
Other
526 stars 218 forks source link

why lr in fc8_kevin is not as large as fc8_pascal ? #19

Closed ToTheBeginning closed 6 years ago

ToTheBeginning commented 7 years ago

They are both new layers, why the lrs are different?

kevinlin311tw commented 6 years ago

It mainly depends on the dataset which you would like to fine-tune. In this repo, we empirically set the learning rate and found it works.