1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
|
-----------------------------------------------------------------------------------------------
-- Dumps the output for each layer in the network. Only works with sequential networks
--
-- Usage: th dump_sequential_output.lua
--
-- Peter Abeles
-----------------------------------------------------------------------------------------------
require 'lfs'
require 'torch'
require 'nn'
local spatialInput = torch.randn(2,3,32,32):float()
local function processSequence( file, sequence , startIndex, inputTensor )
-- print("=============== ENTER processSequence() "..sequence.__typename)
local output = inputTensor
local num = startIndex
for i=1,sequence:size() do
local m = sequence.modules[i]
print( "name = "..m.__typename )
if m.__typename == "nn.Sequential" then
output,num = processSequence(file,m,num,output)
else
output = m:forward(output)
-- print(" ... saving "..num)
torch.save(paths.concat(file,string.format('layer%03d.t7',num)), output)
end
num = num + 1
end
return output,num
end
for file in lfs.dir("./") do
if lfs.attributes(file,"mode") == "directory" then
local modelPath = paths.concat(file,"model.net")
if lfs.attributes(modelPath,"mode") == "file" then
print("Found network "..file)
torch.save(paths.concat(file,'layers_input.t7'), spatialInput)
local network = torch.load(paths.concat(file,"model_float.net")):float()
network:evaluate()
local output,num = processSequence(file,network,1,spatialInput)
-- print(output)
-- output = network:forward(spatialInput)
-- print(output)
end
end
end
|
