Create Convolution layer followed by Relu activation
Create Convolution layer followed by Relu activation
previous node for convolution layer node
convolution size info. Should be (input plane number, output plane number, kernel size, stride size, pad size). We'are assuming kernel width = kernel height, stride width = stride height, pad width = pad height
layer name
prefix for the layer name
kernel group number
whether to propagate gradient back
Relu node
Load model weights and bias from source model to target model
Load model weights and bias from source model to target model
source model
target model
whether to match all layers' weights and bias, if not, only load existing source weights and bias
select results (confs || locs || priorboxes), use JoinTable to concat them into one tensor
select results (confs || locs || priorboxes), use JoinTable to concat them into one tensor
start index of the result
dimension to join
specify the number of dimensions for the input
result layer name
share the storage of SpatialConvolution fInput note that this sharing only works for Inference only
share the storage of SpatialConvolution fInput note that this sharing only works for Inference only
model to share
Stop the input gradient of layers whose name ended with priorbox in a model, their input gradient are not computed.
Stop the input gradient of layers whose name ended with priorbox in a model, their input gradient are not computed.
the graph model