-
Notifications
You must be signed in to change notification settings - Fork 7
/
Copy pathz_traverse_express_smile.lua
323 lines (291 loc) · 13.1 KB
/
z_traverse_express_smile.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
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
------------------------------------------------------------
--- Code for
-- Shu et al., Neural Face Editing with Intrinsic Image Disentangling, CVPR 2017.
------------------------------------------------------------
require 'hdf5'
require 'nngraph'
require 'cudnn'
require 'torch'
require 'nn'
require 'cunn'
require 'optim'
require 'image'
require 'pl'
require 'paths'
require 'modules/SHShading'
require 'modules/SACompose'
require 'modules/SHPartialShading'
require 'modules/SHPartialShadingRGB'
require 'modules/SHPartialShadingRGB_bw'
require 'modules/TVLoss'
require 'modules/TVCriterion'
require 'modules/TVSelfCriterion'
require 'modules/TVSelfPartialCriterion'
require 'modules/RangeSelfCriterion'
require 'modules/SmoothSelfCriterion'
require 'modules/SmoothSelfPartialCriterion'
require 'modules/UniLengthCriterion'
require 'modules/PerElementNorm'
require 'modules/BatchWhiteShadingCriterion'
require 'modules/MarginNegMSECriterion'
require 'modules/BatchWhiteShadingCriterion2'
require 'modules/LightCoeffCriterion'
require 'modules/PartialSimNCriterion'
require 'modules/FBMCompose'
require 'modules/MaskedReconCriterion'
ok, disp = pcall(require, 'display')
if not ok then print('display not found. unable to plot') end
formation = require 'dipface_express'
----------------------------------------------------------------------
-- parse command-line optionsls
opt = lapp[[
-s,--save (default 'z_traverse_express_5_smile') subdirectory to save logs
--editsave (default '') subdirectory to save editing
--saveFreq (default 1) save every saveFreq epochs
-n,--network (default 'models/dipface_express_pretrain_CelebA.t7') reload pretrained network
-p,--plot plot while training
-r,--learningRate (default 0.001) learning rate
-b,--batchSize (default 100) batch size
-m,--momentum (default 0) momentum, for SGD only
--coefL1 (default 0) L1 penalty on the weights
--coefL2 (default 0) L2 penalty on the weights
-t,--threads (default 4) number of threads
-g,--gpu (default 2) gpu to run on (default cpu)
-d,--noiseDim (default 512) dimensionality of noise vector
--K (default 1) number of iterations to optimize D for
-w, --window (default 3) windsow id of sample image
--scale (default 64) scale of images to train on
--weight_1 (default 100)
--weight_2 (default 1)
--z_dim (default 128) dimensionality of z code
--zA_dim (default 128) dimensionality of A code
--zN_dim (default 128) dimensionality of N code
--zL_dim (default 10) dimensionality of L code
--zB_dim (default 128) dimensionality of B code
--zM_dim (default 32) dimensionality of M code
--dz_dim (default 128) dimensionality of dz code
--margin (default 20) value of margin
]]
paths.mkdir(opt.save)
nw1 = opt.weight_1/(opt.weight_1 + opt.weight_2)
nw2 = opt.weight_2/(opt.weight_1 + opt.weight_2)
if opt.gpu < 0 or opt.gpu > 3 then opt.gpu = false end
print(opt)
-- fix seed
torch.manualSeed(1)
-- threads
torch.setnumthreads(opt.threads)
print('<torch> set nb of threads to ' .. torch.getnumthreads())
local function sleep(n)
os.execute("sleep " .. tonumber(n))
end
if opt.gpu then
cutorch.setDevice(opt.gpu + 1)
print('<gpu> using device ' .. opt.gpu)
torch.setdefaulttensortype('torch.CudaTensor')
else
torch.setdefaulttensortype('torch.FloatTensor')
end
opt.geometry = {3, opt.scale, opt.scale}
--opt.geometry = {6, opt.scale, opt.scale}
local input_sz = opt.geometry[1] * opt.geometry[2] * opt.geometry[3]
------------------------------------------
-- load pre trained network
------------------------------------------
print('<trainer> reloading previously trained network: ' .. opt.network)
tmp = torch.load(opt.network)
model_D = tmp.D
model_Enc = tmp.Enc
model_Dec = tmp.Dec
if opt.gpu then
print('Copy model to gpu')
model_D:cuda()
model_Enc:cuda()
model_Dec:cuda()
end
----------------------------------------------------------------------
-- Get examples to plot
function getSamples(dataset, N, fixid)
local numperclass = numperclass or 10
local N = N or 8
local fixid = fixid or false
--local noise_inputs = torch.Tensor(N, opt.noiseDim)
-- Generate samples
local inputs_samples_img = torch.Tensor(N, opt.geometry[1], opt.geometry[2], opt.geometry[3])
local inputs_samples_mask = torch.Tensor(N, opt.geometry[1], opt.geometry[2], opt.geometry[3])
--local inputs_samples_bias = torch.Tensor(N, opt.geometry[1], opt.geometry[2], opt.geometry[3])
math.randomseed( os.time() )
for i = 1,N do
local idx = math.random(dataset:size()[1])
if fixid then
idx = i
end
local sample = dataset[idx]
local sample_img = sample:narrow(1,1,3)
local sample_mask = sample:narrow(1,7,1)
--local sample_bias = sample_img:clone()
sample_mask = torch.repeatTensor(sample_mask,3,1,1)
--sample_bias = sample_bias:fill(1)
inputs_samples_img[i] = sample_img:clone()
inputs_samples_mask[i] = sample_mask:clone()
--inputs_samples_bias[i] = sample_bias:clone()
end
local zA_sample, zN_sample, zL_sample = unpack(model_Enc:forward(inputs_samples_img))
local synth_samples,A_samples,N_samples,Lr_samples,Lg_samples,Lb_samples,S_samples = unpack(model_Dec:forward({zA_sample, zN_sample, zL_sample,inputs_samples_mask}))
synth_samples = nn.HardTanh():forward(synth_samples)
A_samples = nn.HardTanh():forward(A_samples)
N_samples = nn.HardTanh():forward(N_samples)
S_samples = nn.Tanh():forward(S_samples)
N_samples = (N_samples+1)*0.5
local to_plot_0 = {}
local to_plot_1 = {}
local to_plot_2 = {}
local to_plot_3 = {}
local to_plot_4 = {}
for i=1,N do
to_plot_0[#to_plot_0 + 1] = inputs_samples_img[i]:float()
to_plot_1[#to_plot_1 + 1] = synth_samples[i]:float()
to_plot_2[#to_plot_2 + 1] = A_samples[i]:float()
to_plot_3[#to_plot_3 + 1] = N_samples[i]:float()
to_plot_4[#to_plot_4 + 1] = S_samples[i]:float()
end
return to_plot_0, to_plot_1, to_plot_2, to_plot_3, to_plot_4
end
--%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function zBarycenterTraverse(input, srcData, tgtData)
local srcBarycenter = torch.mean(srcData, 1)
local tgtBarycenter = torch.mean(tgtData, 1)
local arrow = tgtBarycenter - srcBarycenter
local arrowArray = torch.repeatTensor(arrow,input:size()[1],1)
local output = input + arrowArray
return output
end
--%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function plotImageMatrix(imagedata, folder, name)
torch.setdefaulttensortype('torch.CudaTensor')
local plot_data = {}
print(imagedata:size())
imagedata = nn.HardTanh():forward(imagedata)
torch.setdefaulttensortype('torch.FloatTensor')
for i =1, imagedata:size()[1] do
plot_data[#plot_data + 1] = imagedata[i]:float()
end
local plot_formatted = image.toDisplayTensor({input = plot_data, nrow=10})
plot_formatted:float()
image.save(folder .. name ..'.png', plot_formatted)
torch.setdefaulttensortype('torch.CudaTensor')
end
--%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
print("Reading positive training data .. ")
local phdf5 = hdf5.open('data/zx_7_d10_inmc_train_smileP.hdf5', 'r')
posData = phdf5:read('zx_7'):all()
phdf5:close()
print("Reading negative training data .. ")
local nhdf5 = hdf5.open('data/zx_7_d10_inmc_train_smileN.hdf5', 'r')
negData = nhdf5:read('zx_7'):all()
nhdf5:close()
print("Reading positive testing data .. ")
local phdf5 = hdf5.open('data/zx_7_d10_inmc_test_smileP.hdf5', 'r')
test_posData = phdf5:read('zx_7'):all()
phdf5:close()
print("Reading negative testing data .. ")
local nhdf5 = hdf5.open('data/zx_7_d10_inmc_test_smileN.hdf5', 'r')
test_negData = nhdf5:read('zx_7'):all()
nhdf5:close()
-------------------------------------------------------------------------------------------
-- compute the Zs of positive data and negative data
local posClusterSize = 2000
local posBatchSize = 100
local negClusterSize = 2000
local negBatchSize = 100
local posInputBatch = torch.Tensor(posBatchSize, opt.geometry[1], opt.geometry[2], opt.geometry[3])
local negInputBatch = torch.Tensor(negBatchSize, opt.geometry[1], opt.geometry[2], opt.geometry[3])
--------------------------------------------------------------------------------------------
--sample positive data zs
local pos_zA = torch.Tensor( posClusterSize, opt.zA_dim)
local pos_zN = torch.Tensor( posClusterSize, opt.zN_dim)
local pos_zL = torch.Tensor( posClusterSize, opt.zL_dim)
local pos_zB = torch.Tensor( posClusterSize, opt.zB_dim)
local pos_zM = torch.Tensor( posClusterSize, opt.zM_dim)
for i = 1, posClusterSize, posBatchSize do
for j = 1,posBatchSize do
local idx = math.random(posData:size()[1])
local sample = posData[idx] -- sample data point
local sample_img = sample:narrow(1,1,3) -- the img in the data point, firt 3 dimenions
posInputBatch[j] = sample_img:clone()
end
zA, zN, zL, zB, zM = unpack(model_Enc:forward(posInputBatch))
pos_zA[{{i, i+posBatchSize-1},{}}] = zA:clone()
pos_zN[{{i, i+posBatchSize-1},{}}] = zN:clone()
pos_zL[{{i, i+posBatchSize-1},{}}] = zL:clone()
pos_zB[{{i, i+posBatchSize-1},{}}] = zB:clone()
pos_zM[{{i, i+posBatchSize-1},{}}] = zM:clone()
end
--sample negative data zs
local neg_zA = torch.Tensor( negClusterSize, opt.zA_dim)
local neg_zN = torch.Tensor( negClusterSize, opt.zN_dim)
local neg_zL = torch.Tensor( negClusterSize, opt.zL_dim)
local neg_zB = torch.Tensor( negClusterSize, opt.zB_dim)
local neg_zM = torch.Tensor( negClusterSize, opt.zM_dim)
for i = 1, negClusterSize, negBatchSize do
for j = 1,negBatchSize do
local idx = math.random(negData:size()[1])
local sample = negData[idx] -- sample data point
local sample_img = sample:narrow(1,1,3) -- the img in the data point, firt 3 dimenions
negInputBatch[j] = sample_img:clone()
end
zA, zN, zL, zB, zM = unpack(model_Enc:forward(negInputBatch))
neg_zA[{{i, i+negBatchSize-1},{}}] = zA:clone()
neg_zN[{{i, i+negBatchSize-1},{}}] = zN:clone()
neg_zL[{{i, i+negBatchSize-1},{}}] = zL:clone()
neg_zB[{{i, i+negBatchSize-1},{}}] = zB:clone()
neg_zM[{{i, i+negBatchSize-1},{}}] = zM:clone()
end
--------------------------------------------------------------------------------------------
-- pret (pre-traversal), the input of the traversal, on the z manifold
local n_pret = 100
local pret_img = torch.Tensor(n_pret, opt.geometry[1], opt.geometry[2], opt.geometry[3])
local pret_mask = torch.Tensor(n_pret, opt.geometry[1], opt.geometry[2], opt.geometry[3])
-- sample images from negative data
for i = 1, n_pret do
--local idx = math.random(test_negData:size()[1])
local idx = i
local sample = test_negData[idx]
local sample_img = sample:narrow(1,1,3)
local sample_mask = sample:narrow(1,7,1)
sample_mask = torch.repeatTensor(sample_mask,3,1,1)
pret_img[i] = sample_img:clone()
pret_mask[i] = sample_mask:clone()
end
pret_zA, pret_zN, pret_zL, pret_zB, pret_zM = unpack(model_Enc:forward(pret_img))
local pret_synth, pret_A, pret_N, pret_Lr, pret_Lg, pret_Lb, pret_S, pret_Nnm, pret_B, pret_M, pret_final = unpack(model_Dec:forward({pret_zA, pret_zN, pret_zL, pret_zB, pret_zM, pret_mask}))
--plot pres
plotImageMatrix(pret_img, opt.save, '/pret_image')
plotImageMatrix(pret_synth, opt.save , '/pret_synth')
plotImageMatrix(pret_A, opt.save , '/pret_A')
plotImageMatrix(pret_N, opt.save , '/pret_N')
plotImageMatrix(pret_S, opt.save , '/pret_S')
plotImageMatrix(pret_B, opt.save , '/pret_B')
plotImageMatrix(pret_M, opt.save , '/pret_M')
plotImageMatrix(pret_final, opt.save , '/pret_final')
--------------------------------------------------------------------------------------------
-- obtain pt (post-traversal), the output of the traversal, on the z manifold
torch.setdefaulttensortype('torch.CudaTensor')
pt_zA = zBarycenterTraverse(pret_zA, neg_zA, pos_zA)
pt_zN = zBarycenterTraverse(pret_zN, neg_zN, pos_zN)
pt_zL = zBarycenterTraverse(pret_zL, neg_zL, pos_zL)
pt_zB = zBarycenterTraverse(pret_zB, neg_zB, pos_zB)
pt_zM = zBarycenterTraverse(pret_zM, neg_zM, pos_zM)
--------------------------------------------------------------------------------------------
-- reconstruct image of postt
-- synthetic image(syn), albedo(A), normal(N), light(Lr, Lg, Lb), shading(S)
local pt_mask = pret_mask:clone()
local pt_synth, pt_A, pt_N, pt_Lr, pt_Lg, pt_Lb, pt_S, pt_Nnm, pt_B, pt_M, pt_final = unpack(model_Dec:forward({pt_zA, pt_zN, pret_zL, pret_zB, pret_zM, pt_mask}))
--plot posts
plotImageMatrix(pt_synth, opt.save , '/pt_synth')
plotImageMatrix(pt_A, opt.save , '/pt_A')
plotImageMatrix(pt_N, opt.save , '/pt_N')
plotImageMatrix(pt_S, opt.save , '/pt_S')
plotImageMatrix(pt_B, opt.save , '/pt_B')
plotImageMatrix(pt_M, opt.save , '/pt_M')
plotImageMatrix(pt_final, opt.save , '/pt_final')