-
Notifications
You must be signed in to change notification settings - Fork 0
/
preprocess.lua
209 lines (195 loc) · 6.9 KB
/
preprocess.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
require 'nn'
require 'optim'
require 'torch'
require 'nn'
require 'math'
require 'cunn'
require 'cutorch'
require 'loadcaffe'
require 'image'
require 'hdf5'
cjson=require('cjson')
require 'xlua'
dofile('loadImage.lua')
-------------------------------------------------------------------------------
-- Input arguments and options
-------------------------------------------------------------------------------
cmd = torch.CmdLine()
cmd:text()
cmd:text('Options')
cmd:option('-img_dir','data/tvsum50/frames_all/','path to the images directory')
cmd:option('-labels','data/tvsum50/frame_file_all.txt','path to the labels file')
cmd:option('-cnn_proto', 'models/VGG_ILSVRC_19_layers_deploy.prototxt', 'path to the cnn prototxt')
cmd:option('-cnn_model', 'models/VGG_ILSVRC_19_layers.caffemodel', 'path to the cnn model')
cmd:option('-batch_size', 10, 'batch_size')
cmd:option('-output_size', 1, 'Size of LSTM unit output')
cmd:option('-num_batches', 200, 'number of batches')
cmd:option('-seq_len', 128, 'seq_len')
cmd:option('-sampling', 10, 'sampling rate. Default 1 in 10')
cmd:option('-out_train_data', 'data/tvsum50/train_data.t7', 'training data file')
cmd:option('-out_train_labels', 'data/tvsum50/train_data_labels.t7', 'training data labels')
cmd:option('-out_test_data', 'data/tvsum50/test_data.t7', 'test data file')
cmd:option('-out_test_labels', 'data/tvsum50/test_data_labels.t7', 'test data labels')
cmd:option('-gpuid', 1, 'which gpu to use. -1 = use CPU')
cmd:option('-backend', 'cudnn', 'nn|cudnn')
opt = cmd:parse(arg)
print(opt)
cutorch.setDevice(opt.gpuid)
----------------------------------------------------------------------
-- 1. Load images and labels
print('Train Labels File: ' .. opt.labels)
local labels = io.open(opt.labels, 'r');
train_list = {}
test_list = {}
for line in labels:lines() do
if not line then
break
end
local img_name, score = unpack(line:split(" "));
local vid = unpack(line:split("_"));
vid = tonumber(vid)
print(img_name, score, vid)
if (vid % 5 == 0) then
if test_list[vid] == nil then
print('Adding empty table for test vid id:' .. vid)
test_list[vid] = {}
end
-- Add frame details to this video's table
table.insert(test_list[vid], {name = img_name, label = tonumber(score)})
-- print(#test_list)
else
if train_list[vid] == nil then
print('Adding empty table for train vid id:' .. vid)
train_list[vid] = {}
end
-- Add frame details to this video's table
table.insert(train_list[vid], {name = img_name, label = tonumber(score)})
-- print(#train_list)
end
end
labels:close()
local num_train_videos = 0
local num_test_videos = 0
for vid,frames in pairs(train_list) do
num_train_videos = num_train_videos + 1
print('Frames for: ' .. vid)
for fid,frame_info in ipairs(frames) do
print(fid, frame_info)
end
end
for vid,frames in pairs(test_list) do
num_test_videos = num_test_videos + 1
--print('Frames for: ' .. vid)
--for fid,frame_info in ipairs(frames) do
-- print(fid, frame_info)
--end
end
print('Number of train videos: ' .. num_train_videos);
print('Number of test videos: ' .. num_test_videos);
---------------------------------------------------------------------
-- 2. Load Model
net=loadcaffe.load(opt.cnn_proto, opt.cnn_model, opt.backend);
net:evaluate()
net=net:cuda()
---------------------------------------------------------------------
-- 3. Extract Features
local batch_size = opt.batch_size
local output_size = opt.output_size
local num_batches = opt.num_batches
local seq_len = opt.seq_len
local sampling = opt.sampling
local function printBatch(batch)
for i,eg in pairs(batch) do
print('Eg: ' .. i)
print('Vid:' .. eg.video_id)
print('Fid:' .. eg.frame_id)
end
end
-- Training Data
local function sampleBatch()
local sample = {}
for i = 1,batch_size do
local vid = 0
local fid = 0
-- Randomly select a training video.
while(true) do
vid = torch.ceil(torch.uniform() * 40)
if (vid % 5 ~= 0) then
break
end
end
-- Randomly select a starting frame such there are sufficient
-- frames for seqLen at specified sampling rate.
local total_frames = #(train_list[vid])
local last_possible_fid = total_frames - ((seq_len - 1) * sampling)
-- print('TotalFrames, LastPossibleFid : ' .. total_frames, last_possible_fid)
assert(last_possible_fid > 0)
fid = torch.ceil(torch.uniform() * last_possible_fid)
-- Insert vid,fid pair in sample.
table.insert(sample, {video_id = vid, frame_id = fid})
end
-- printBatch(sample)
return sample
end
-- Training Data
local train_inputs, train_targets = {}, {}
for i = 1,num_batches do
xlua.progress(i, num_batches)
train_inputs[i], train_targets[i] = {}, {}
local batch = sampleBatch()
print('Processing Batch: ' .. i)
for j = 1,seq_len do
-- Process j-th frames for all egs in the batch.
-- print('Processing Seq: ' .. j)
local frames = torch.CudaTensor(batch_size, 3, 224, 224)
local labels = torch.CudaTensor(batch_size, output_size)
for k = 1,batch_size do
-- Read j-th frame from kth eg in the batch.
local cur_vid = batch[k].video_id
local cur_fid = batch[k].frame_id + (j - 1) * sampling
-- print(cur_vid, cur_fid)
-- print('Loading Frame: ' .. train_list[cur_vid][cur_fid].name)
frames[k] = loadim(opt.img_dir .. train_list[cur_vid][cur_fid].name):cuda()
labels[k] = torch.CudaTensor({train_list[cur_vid][cur_fid].label})
end
-- Add features and labels for j-th frames for all egs in the batch.
net:forward(frames)
table.insert(train_inputs[i], net.modules[43].output:clone())
table.insert(train_targets[i], labels:clone())
collectgarbage()
end
-- Save partial dataset
if (i % 50 == 0) then
torch.save(opt.out_train_data .. i, train_inputs)
torch.save(opt.out_train_labels .. i, train_targets)
end
end
-- Test Data
local test_inputs, test_targets = {}, {}
-- Need to keep batch size 1 since all test videos can have different lengths.
batch_size = 1
for i = 1,num_test_videos do
xlua.progress(i, num_test_videos)
test_inputs[i], test_targets[i] = {}, {}
print('Processing Test Video: ' .. i)
local vid = i * 5
local fid = torch.ceil(torch.uniform() * sampling)
while fid < #(test_list[vid]) do
-- Process j-th frame for test video.
local frames = torch.CudaTensor(batch_size, 3, 224, 224)
local labels = torch.CudaTensor(batch_size, output_size)
for k = 1,batch_size do
frames[k] = loadim(opt.img_dir .. test_list[vid][fid].name):cuda()
labels[k] = torch.CudaTensor({test_list[vid][fid].label})
end
-- Add features and labels for j-th frames for all egs in the batch.
net:forward(frames)
table.insert(test_inputs[i], net.modules[43].output:clone())
table.insert(test_targets[i], labels:clone())
collectgarbage()
-- Increment fid
fid = fid + sampling
end
end
torch.save(opt.out_test_data, test_inputs)
torch.save(opt.out_test_labels, test_targets)