about ConvAffine layer and momentum

facebookresearch / Detectron

FAIR's research platform for object detection research, implementing popular algorithms like Mask R-CNN and RetinaNet.

Apache License 2.0

26.22k stars 5.45k forks source link

about ConvAffine layer and momentum #927

Open chenhangcal opened 5 years ago

chenhangcal commented 5 years ago

I run the convert_pkl_2_pb.py script with the default faster-rcnn model and get a caffe2 pb file. While in the pb graph I did not see convaffine layer and there's no " _momentum" parameter. Is that right, no momentum when running inference?