-
Notifications
You must be signed in to change notification settings - Fork 3
/
demo_ours_FN2cssftsd.m
175 lines (148 loc) · 5.67 KB
/
demo_ours_FN2cssftsd.m
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
addpath tools
addpath(fullfile(db_matlab_root_dir,'db_util'));
addpath(fullfile(db_matlab_root_dir,'measures'));
clear all
para_sample_rate=10;
number_sample=4;
%iter_h=50;
threshold_RANSAC=5;
para_a=0.5;
para_b=1.8;
ival =2;
Ftype = 'FN2';
% Ftype = 'FN2-css-ft-sd';
threshold_co=0.05;
%map =[1,1;2,1;3,1;4,1;1,2;2,2;3,2;4,2;1,3;2,3;3,3;4,3;1,4;2,4;3,4;4,4];
w=480;
h=854;
x=1:w;
y=1:h;
xx=repmat(x',1,h);
yy=repmat(y,w,1);
point=[reshape(xx,1,w*h);reshape(yy,1,w*h)];
point_extend = [point.^2;point(1,:).*point(2,:);point;ones(1,w*h)];
%[w_d,h_d,~]=size(flow_d);
w_d=47;
h_d=85;
img_temp=ones(w_d,h_d);
% [point_d(:,1),point_d(:,2)]=find(img_temp);
[point_d1,point_d2]=find(img_temp);
point_d = [point_d1,point_d2]';
point_d=point_d*para_sample_rate;
point_d_extend=[point_d.^2;point_d(1,:).*point_d(2,:);point_d;ones(1,w_d*h_d)];
% Get the ids of all sequences
seq_ids = db_seqs();
%%
%parameter
%for ival =2
for para_a=2.85
for para_b = 0.33
for number_sample=16
pieces_x = 4;%ceil(sqrt(number_sample*4*w/h));
pieces_y = 8;%ceil(sqrt(number_sample*4*h/w));
num_pieces = pieces_x*pieces_y;
map = constructmap(pieces_x,pieces_y);
for iter_h = 5
select_idx_record=zeros(iter_h,number_sample);
num_record=zeros(iter_h,1);
for times =1
for speed =25
% Name of a result technique
%%
%times
% result_id = ['Ours_ai_pa028_pb020_t' sprintf('%03d',int32(times))];
%result_id = ['Ours_FN2cssftsd' sprintf('%03d',int32(para_a*100)) sprintf('_pb%03d',int32(para_b*100))];
% result_id = 'Ours_FN2cssftsd_test';
result_id = 'Ours_FN2_test-iterh5';
% result_id = 'Ours_ai_test';
result_dir = fullfile(db_root_dir,'Results','Segmentations',db_imsize, result_id);
mkdir(result_dir);
%%
for s_id = 1:length(seq_ids)
seq_id = seq_ids{s_id};
mkdir(fullfile(result_dir,seq_id));
frame_ids = db_frame_ids(seq_ids{s_id});
fprintf('%s contains %d images: \n',seq_ids{s_id},length(frame_ids));
scene_speed_tmp =0;
for f_id = 1:length(frame_ids)-1
flopath = fullfile(db_root_dir,'FLO',Ftype,db_imsize,sprintf('ival%01d',ival),seq_id,sprintf('%05d.flo',f_id));
%flopath = 'G:\data\flow_final\13\flow_1\0432.flo';
flow=readFlowFile(flopath);
% imgcl=flowToColor(flow);
% figure(2)
% imshow(imgcl);
%下采样
flow_d=img_desample(flow,para_sample_rate);
% [w_d,h_d,~]=size(flow_d);
flow_d=reshape(flow_d,w_d*h_d,2);
num_record_max=0;
flo_diff_max =10^10;
point_selet_idx_max = 0;
for k=1:iter_h
%with CRA
area_idx=randperm(num_pieces,number_sample);
area_idx_x=map(area_idx,1);
area_idx_y=map(area_idx,2);
area_idx_x=floor((area_idx_x'-1+rand(1,number_sample))*w_d/pieces_x)+1;
area_idx_y=floor((area_idx_y'-1+rand(1,number_sample))*h_d/pieces_y)+1;
point_selet_idx=((area_idx_y-1)*w_d+area_idx_x)';%尝试改这里
select_idx_record(k,:)=point_selet_idx;
point_bt=point_d(:,point_selet_idx);
point_bt_extend = [point_bt.^2;point_bt(1,:).*point_bt(2,:);point_bt;ones(1,number_sample)];
point_ft=[point_bt;ones(1,number_sample)]+[flow_d(point_selet_idx,:)';zeros(1,number_sample)];
H = point_ft*point_bt_extend'*(point_bt_extend*point_bt_extend')^(-1);
flow_d_temp=H*point_d_extend-[point_d;ones(1,w_d*h_d)];
flow_diff=sum(abs(flow_d_temp(1:2,:)'-flow_d),2)*10;
flow_diff = threshold_RANSAC-flow_diff;
idx=find(uint8(flow_diff));
sum_diff =sum(flow_diff(idx));
num_record(k,1)=size(idx,1);
if(size(idx,1)>num_record_max)
point_selet_idx_max=k;
num_record_max = size(idx,1);
flo_diff_max=sum_diff;
elseif(size(idx,1)==num_record_max)
if(flo_diff_max>sum_diff)
point_selet_idx_max=k;
num_record_max = size(idx,1);
flo_diff_max=sum_diff;
end
end
end
%[~,ord]=sort(num_record,1,'ascend');
point_selet_idx=select_idx_record(point_selet_idx_max,:);
point_bt=point_d(:,point_selet_idx);
point_bt_extend = [point_bt.^2;point_bt(1,:).*point_bt(2,:);point_bt;ones(1,number_sample)];
point_ft=[point_bt;ones(1,number_sample)]+[flow_d(point_selet_idx,:)';zeros(1,number_sample)];
H = point_ft*point_bt_extend'*(point_bt_extend*point_bt_extend')^(-1);
%% 动态更新阈值
%threshold_RANSAC=sqrt(H(1,3)^2+H(2,3)^2)*para_b+para_a;
scene_speed =mean(sqrt(sum(flow_d(point_selet_idx,:).^2,2)));
threshold_RANSAC=scene_speed*para_b+para_a;
ival = floor(speed/scene_speed*ival)+1;
if ival>5
ival =5;
end
%% 利用H和光流图求前景mask_ori
flow_temp=H*point_extend-[point;ones(1,w*h)];
flow_temp=reshape(flow_temp(1:2,:)',w,h,2);
% imgcl=flowToColor(flow_temp);
% figure(3)
% imshow(imgcl);
flow_diff=sum(abs(flow_temp-flow),3)*10;
mask_ori=uint8(flow_diff-threshold_RANSAC*10)*255;
% figure(1)
% imshow(mask_ori);
wpath = fullfile(result_dir, seq_id, [frame_ids{f_id} '.png']);
imwrite(mask_ori,wpath);
end
end
[eval, raw_eval] = eval_result(result_id,{'J'},'all');
result = mean(eval.J.mean);
end
end
end
end
end
end
%end