-
Notifications
You must be signed in to change notification settings - Fork 0
/
c3d.lua
97 lines (83 loc) · 3.11 KB
/
c3d.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
local nninit = require 'nninit'
local function c3d(batchSize)
---[[
-- Create table describing C3D configuration
local cfg = {64 , 'M1' , 128 , 'M' , 256 , 'M' , 256 , 'M' , 256 ,'M'}
local features = nn.Sequential()
do
local iChannels = 3;
for k,v in ipairs(cfg) do
if v == 'M' then
features:add(nn.VolumetricMaxPooling(2,2,2,2,2,2):ceil())
elseif v == 'M1' then
features:add(nn.VolumetricMaxPooling(1,2,2,1,2,2):ceil())
else
local oChannels = v;
features:add(nn.VolumetricConvolution(iChannels,oChannels,3,3,3,1,1,1,1,1,1):init('weight',nninit.kaiming))
features:add(nn.ReLU(true))
iChannels = oChannels;
end
end
end
--features:get(1).gradInput = nil
local classifier = nn.Sequential()
classifier:add(nn.View(256*1*4*4))
classifier:add(nn.Linear(256*1*4*4, 2048):init('weight',nninit.kaiming))
classifier:add(nn.ReLU(true))
classifier:add(nn.Dropout(0.7))
classifier:add(nn.Linear(2048, 2048):init('weight',nninit.kaiming))
classifier:add(nn.ReLU(true))
classifier:add(nn.Dropout(0.7))
classifier:add(nn.Linear(2048, 101):init('weight',nninit.kaiming)) -- UCF-101
--classifier:add(nn.LogSoftMax())
local model = nn.Sequential()
model:add(features):add(classifier)
return model, {batchSize,3,16,112,112}
end
--]]
--[[
local cfg = {64, 'M1', 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'}
local features = nn.Sequential()
do
local iChannels = 3;
for k,v in ipairs(cfg) do
if v == 'M' then
features:add(nn.VolumetricMaxPooling(2,2,2,2,2,2):ceil())
elseif v == 'M1' then
features:add(nn.VolumetricMaxPooling(1,2,2,1,2,2):ceil())
else
local oChannels = v;
features:add(nn.VolumetricConvolution(iChannels,oChannels,3,3,3,1,1,1,1,1,1))
features:add(nn.ReLU(true))
iChannels = oChannels;
end
end
end
--features:get(1).gradInput = nil
local classifier = nn.Sequential()
classifier:add(nn.View(512*1*4*4))
classifier:add(nn.Linear(512*1*4*4, 4096))
classifier:add(nn.ReLU(true))
classifier:add(nn.Dropout(0.5))
classifier:add(nn.Linear(4096, 4096))
classifier:add(nn.ReLU(true))
classifier:add(nn.Dropout(0.5))
classifier:add(nn.Linear(4096, 101):init('weight',nninit.kaiming)) -- UCF-101
classifier:add(nn.LogSoftMax())
local model = nn.Sequential()
model:add(features):add(classifier)
net = torch.load("/mnt/data1/Nets/DownloadedNets/conv3d_deepnetA_sport1m_iter_1900000.t7")--("/mnt/data1/c3d_experiment_overfit_control/c3d.t7")
convs ={1,4,7,9,12,14,17,19}
for i,v in ipairs(convs) do
model.modules[1].modules[v].weight = net.modules[v].weight:clone()
model.modules[1].modules[v].bias = net.modules[v].bias:clone()
end
fc = {2,5}
for i,v in ipairs(fc) do
model.modules[2].modules[v].weight = net.modules[v+21].weight
model.modules[2].modules[v].bias = net.modules[v+21].bias
end
return model, {batchSize,3,16,112,112}
end
--]]
return c3d