Hi everyone,
This is going to be my first question in this mailing list, so bear with me and forgive my inexperience in Caffe since I m fairly new to it.
I have been working on Cifar-10 dataset for a while, implementing different algorithms and see the effect on complex state of the art models.
I'm trying to make VGG-16 model to work on Cifa-10 dataset. I found a similar model here :
https://gist.github.com/jjkjkj/39e87099e9381e6886a5 I took the model without batch normalization for experimenting. But this model uses MSR weight initialization and Caffe master branch does not have
this instead it has MSRA weight initialization. So i replaced msr initialization with msra and modified parameters to make it work. However it seems
it has problems with converging. I m trying different parameters to see if it work.
So my questions are MSR and MSRA initialization different and does anyone have a working VGG model on cifar-10 dataset ?
Regards,
Yusuf