sanghoon / pva-faster-rcnn

Demo code for PVANet
https://arxiv.org/abs/1611.08588
Other
650 stars 241 forks source link

generated prototxt and caffemodel ? #90

Open ananddb90 opened 6 years ago

ananddb90 commented 6 years ago

Hi @sanghoon Thank you for your work and share. I tried your script for ResNet50 to combine conv, batchnorm and scale layer into one conv layer. I have doubts regarding generated prototxt and caffemodel

  1. generated prototxt combines only scale and batchnorm into one and keep conv layers as it is before : conv - BN- scale - relu after : conv - BN - scale - relu

  2. generated caffemodel weight should also change as now number of parameters are less to be stored?