Skip to first unread message
Stephen Pegoraro
May 11, 2015, 6:28:08 AM5/11/15
to caffe...@googlegroups.com
Hi all,
I'm writing an fpga implementation of the feedforward components of the network and intend to use weights that have already been trained.
I've managed to get the weight and bias tensors out of each layer just fine and now plan on using them, however for consistency I'd like to know whether the indices of a symmetric convolution kernel are the standard (-2, -1, 0, 1 [for 4x4 kernel]) in caffe or whether they're something different?
Cheers,
Steve
I'm writing an fpga implementation of the feedforward components of the network and intend to use weights that have already been trained.
I've managed to get the weight and bias tensors out of each layer just fine and now plan on using them, however for consistency I'd like to know whether the indices of a symmetric convolution kernel are the standard (-2, -1, 0, 1 [for 4x4 kernel]) in caffe or whether they're something different?
Cheers,
Steve
Reply all
Reply to author
Forward
0 new messages