-
Notifications
You must be signed in to change notification settings - Fork 23
/
Copy pathCompactBilinearPooling.lua
125 lines (110 loc) · 4.07 KB
/
CompactBilinearPooling.lua
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
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
local CompactBilinearPooling, parent = torch.class('nn.CompactBilinearPooling', 'nn.Module')
require 'spectralnet'
-- Reference:
-- Multimodal Compact Bilinear Pooling for Visual Question Answering and Visual Grounding
-- Fukui et al. (2016) http://arxiv.org/abs/1606.01847
function CompactBilinearPooling:__init(outputSize, homogeneous)
assert(outputSize and outputSize >= 1, 'missing outputSize!')
self.outputSize = outputSize
self.homogeneous = homogeneous
self:reset()
self.debug = false
end
function CompactBilinearPooling:reset()
self.h1 = torch.Tensor()
self.h2 = torch.Tensor()
self.s1 = torch.Tensor()
self.s2 = torch.Tensor()
self.y = torch.Tensor()
self.gradInput = {}
self.tmp = torch.Tensor()
end
function CompactBilinearPooling:sample()
self.h1:uniform(0,self.outputSize):ceil()
self.h2:uniform(0,self.outputSize):ceil()
self.s1:uniform(0,2):floor():mul(2):add(-1)
self.s2:uniform(0,2):floor():mul(2):add(-1)
end
function CompactBilinearPooling:psi()
self.y:zero()
local batchSize = self.input[1]:size(1)
for i=1,2 do
if self.homogeneous then -- using the same samples
self.y[i]:indexAdd(2,self.h1,torch.cmul(self.s1:repeatTensor(batchSize,1),self.input[i]))
else
if i==1 then
self.y[i]:indexAdd(2,self.h1,torch.cmul(self.s1:repeatTensor(batchSize,1),self.input[i]))
else
self.y[i]:indexAdd(2,self.h2,torch.cmul(self.s2:repeatTensor(batchSize,1),self.input[i]))
end
end
end
end
function CompactBilinearPooling:conv(x,y)
local batchSize = x:size(1)
local dim = x:size(2)
local function makeComplex(x,y)
self.x_ = self.x_ or torch.CudaTensor()
self.x_:resize(x:size(1),1,1,x:size(2),2):zero()
self.x_[{{},{1},{1},{},{1}}]:copy(x)
self.y_ = self.y_ or torch.CudaTensor()
self.y_:resize(y:size(1),1,1,y:size(2),2):zero()
self.y_[{{},{1},{1},{},{1}}]:copy(y)
end
makeComplex(x,y)
self.fft_x = self.fft_x or torch.CudaTensor(batchSize,1,1,dim,2)
self.fft_y = self.fft_y or torch.CudaTensor(batchSize,1,1,dim,2)
self._conv = self._conv or torch.CudaTensor()
self._conv:resize(batchSize,1,1,dim*2)
cufft.fft1d(self.x_:view(x:size(1),1,1,-1), self.fft_x)
cufft.fft1d(self.y_:view(y:size(1),1,1,-1), self.fft_y)
cufft.ifft1d(self.fft_x:cmul(self.fft_y), self._conv)
return self._conv:resize(batchSize,1,1,dim,2):select(2,1):select(2,1):select(3,1)
end
function CompactBilinearPooling:updateOutput(input)
if self.debug then sys.tic(1) end
self.input = input
local inputSizes1 = input[1]:size()
local inputSizes2 = input[2]:size()
if 0==#self.h1:size() then
self.h1:resize(inputSizes1[#inputSizes1])
self.h2:resize(inputSizes2[#inputSizes2])
self.s1:resize(inputSizes1[#inputSizes1])
self.s2:resize(inputSizes2[#inputSizes2])
self:sample() -- samples are fixed
end
if 2 > #inputSizes1 then -- no batch
self.y:resize(2, self.outputSize)
elseif 2 == #inputSizes1 then -- batch
local batchSize = inputSizes1[1]
self.y:resize(2, batchSize, self.outputSize)
else
assert(false, '# of dimensions > 2')
end
if self.debug then print('pre:', sys.toc(1)) end
self:psi()
if self.debug then print('psi:', sys.toc(1)) end
self.output = self:conv(self.y[1], self.y[2])
if self.debug then print('conv:', sys.toc(1)) end
return self.output
end
function CompactBilinearPooling:updateGradInput(input, gradOutput)
local dim = input[1]:size(2)
local batchSize = input[1]:size(1)
self.gradInput = self.gradInput or {}
for k=1,2 do
self.gradInput[k] = self.gradInput[k] or input[k].new()
self.gradInput[k]:resizeAs(input[k]):zero()
self.tmp = self.tmp or gradOutput.new()
self.tmp:resizeAs(gradOutput)
self.tmp = self:conv(gradOutput, self.y[k%2+1])
if k==1 then
self.gradInput[k]:index(self.tmp, 2, self.h1)
self.gradInput[k]:cmul(self.s1:repeatTensor(batchSize,1))
else
self.gradInput[k]:index(self.tmp, 2, self.h2)
self.gradInput[k]:cmul(self.s2:repeatTensor(batchSize,1))
end
end
return self.gradInput
end