search

wangzheallen / transfer_ijcv

This is the release code and model for submission of Transferring Object-Scene Convolutional Neural Networks for Event Recognition in Still Images
9 stars 3 forks source link

About the version of your caffe #1

Open CuriousCat-7 opened 6 years ago

CuriousCat-7 commented 6 years ago

Hi, I am trying to convert your caffe model to torch and pytorch. But the well-used caffe2torch converter(loadcaffe) does not recognize the BN layer. And other converters do not recognize the parameters(frozen, slop filter, etc) in batch normalization layers.

So I am wondering does this project use official caffe code? Actually the BN layer of the .prototext files cannot be recognized by http://ethereon.github.io/netscope/#/editor, a well-known caffe visualization online tool.

So, could you please updata the .prototxt and .caffemodel in official caffe? Or, are there any pre-trained model written in other frameworks?

Thank you Best, Neo

wanglimin commented 6 years ago

These models are trained using our modified caffe toolbox:

https://github.com/yjxiong/caffe