Fully connected to convolution layer

conv_to_linear.lua

require 'nn'
 
-- you just need to provide the linear module you want to convert,
-- and the dimensions of the field of view of the linear layer

function convertLinear2Conv1x1(linmodule,in_size)
--[[
    Convert Linear modules to convolution modules. 
    
    Arguments
        linmodule - pointer of the module to be scanned for convertion
        in_size - 2x1 table containing the convolution stride (stride_x, stride_y) for the convolution module. 
            
    Return values
        convmodule - output convolution module (nn.SpatialConvolution())      
    
    Example:
            input = torch.rand(3,6,6)
            
            m = nn.Linear(3*6*6,10)
             
            mm = convertLinear2Conv1x1(m,{6,6})
             
            output_lin = m:forward(input:view(3*6*6))
            output_conv = mm:forward(input)
    
--]]
   local s_in = linmodule.weight:size(2)/(in_size[1]*in_size[2])
   local s_out = linmodule.weight:size(1)
   local convmodule = nn.SpatialConvolution(s_in,s_out,in_size[1],in_size[2],1,1)
   convmodule.weight:copy(linmodule.weight)
   convmodule.bias:copy(linmodule.bias)
   return convmodule
end

function convertLinear2Conv1x1_v2(linmodule,in_size, modulepointer)
--[[
    Convert Linear modules to convolution modules. Additionally, the convolutional module can be specified.
    
    Arguments
        linmodule - pointer of the module to be scanned for convertion.
        in_size - 2x1 table containing the convolution stride (stride_x, stride_y) for the convolution module. 
        modulepointer - Pointer to the convolutional module to be used for convolution (eg: nn.SpatialConvolution, cudnn.SpatialConvolition)
            
    Return values
        convmodule - output convolution module     
    
    Example:
            input = torch.rand(3,6,6)
            
            m = nn.Linear(3*6*6,10)
             
            mm = convertLinear2Conv1x1_v2(m,{6,6}, nn.SpatialConvolution)
             
            output_lin = m:forward(input:view(3*6*6))
            output_conv = mm:forward(input)
--]]
   local s_in = linmodule.weight:size(2)/(in_size[1]*in_size[2])
   local s_out = linmodule.weight:size(1)
   local moduletype = modulepointer or nn.SpatialConvolution
   local convmodule = moduletype(s_in,s_out,in_size[1],in_size[2],1,1)
   convmodule.weight:copy(linmodule.weight)
   convmodule.bias:copy(linmodule.bias)
   return convmodule
end