This repository contains the source code for the semantic image segmentation method described in the ICCV 2015 paper: Conditional Random Fields as Recurrent Neural Networks. http://crfasrnn.torr.vision/
I was wondering if it is necessary to set decay_mult: 0 in MultiStageMeanfield layer? What I found is that with decay_mult: 1, the learned kernel (i.e. spatial and bilateral filters) weights are almost 0, even when they were initialized with 3.0 and 5.0, respectively. In your prototxt, you are using the default value (i.e., decay_mult: 1).
Would you please share your experience and thoughts on this?
Thanks!
Kun
(p.s. In solver.prototxt, weight_decay: 0.0005 is used.)
Dear @bittnt and @bernard24 ,
I was wondering if it is necessary to set
decay_mult: 0inMultiStageMeanfieldlayer? What I found is that withdecay_mult: 1, the learned kernel (i.e. spatial and bilateral filters) weights are almost 0, even when they were initialized with 3.0 and 5.0, respectively. In your prototxt, you are using the default value (i.e.,decay_mult: 1).Would you please share your experience and thoughts on this?
Thanks! Kun
(p.s. In solver.prototxt,
weight_decay: 0.0005is used.)