-
Notifications
You must be signed in to change notification settings - Fork 0
/
fast_guided_filter_generator.cpp
205 lines (163 loc) · 7.45 KB
/
fast_guided_filter_generator.cpp
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
#include <Halide.h>
using namespace Halide;
class FastGuidedFilter : public Halide::Generator<FastGuidedFilter>{
public:
GeneratorParam<int> radius{"radius",8,1,100};
Input<Buffer<uint8_t>> guided{"guided", 3},input{"input", 3};
Input<float> eps{"eps"};
//@TODO set radius as Input param is really slow.
//Input<int> radius{"radius"};
//@TODO still dont know how to write a [up/down]sample func with,arbitrary scale factor.
//Input<int> s{"s"};
Output<Buffer<uint8_t>> result{"result", 3};
void generate() {
Var x("x"),y("y"),c("c");
Func input_c = BoundaryConditions::mirror_image(input);
Func guided_c = BoundaryConditions::mirror_image(guided);
Expr rad = cast<int>(radius/2);
Expr size = 2*rad+1;
Expr area = size*size;
Expr eps_ = eps*area;
Expr area_ = 1.0f/area;
Expr fac = 255.0f/area;
Halide::RDom rb(-rad, size, "rb");
Halide::Func inputF("inputFloat"),guidedF("guidedFloat");
Halide::Func mean_I("mean_I"),mean_p("mean_p"),
II("II"),Ip("IP"),corr_I("corr_I"),corr_Ip("corr_Ip"),
cov_Ip("cov_Ip"),var_I("var_I"),
a("a"),b("b"),mean_a("mean_a"),mean_b("mean_b"),q("q");
Halide::Func mean_I_sx("mean_I_sx"),corr_I_sx("corr_I_sx"),mean_p_sx("mean_p_sx"),
corr_Ip_sx("corr_Ip_sx"),mean_a_sx("mean_a_sx"),mean_b_sx("mean_b_sx");
Func inputSub("inputSub"),guidedSub("guidedSub"),
mean_aUp("mean_aUp"),mean_bUp("mean_bUp");
inputF(x,y,c) = Halide::cast<float>(input_c(x,y,c)/255.0f);
guidedF(x,y,c) = Halide::cast<float>(guided_c(x,y,c));
inputSub(x,y,c) = downsample(inputF)(x,y,c);
guidedSub(x,y,c) = downsample(guidedF)(x,y,c);
mean_I_sx(x,y,c) = sum(guidedSub(x+rb, y, c));
mean_I(x,y,c) = sum(mean_I_sx(x,y+rb,c));
II(x,y,c) = guidedSub(x,y,c)*guidedSub(x,y,c);
Ip(x,y,c) = inputSub(x,y,c)*guidedSub(x,y,c);
corr_I_sx(x,y,c) = sum(II(x+rb,y,c));
corr_I(x,y,c) = sum(corr_I_sx(x,y+rb,c));
mean_p_sx(x,y,c) = sum(inputSub(x+rb,y,c));
mean_p(x,y,c) = sum(mean_p_sx(x,y+rb,c));
corr_Ip_sx(x,y,c) = sum(Ip(x+rb,y,c));
corr_Ip(x,y,c) = sum(corr_Ip_sx(x,y+rb,c));
cov_Ip(x,y,c) = corr_Ip(x,y,c)-mean_I(x,y,c)*mean_p(x,y,c)*area_;
var_I(x,y,c) = corr_I(x,y,c)-mean_I(x,y,c)*mean_I(x,y,c)*area_;
a(x,y,c) = cov_Ip(x,y,c)/(var_I(x,y,c)+eps_);
b(x,y,c) = (mean_p(x,y,c)-a(x,y,c)*mean_I(x,y,c))*area_;
mean_a_sx(x,y,c) = sum(a(x+rb,y,c));
mean_a(x,y,c) = sum(mean_a_sx(x,y+rb,c));
mean_b_sx(x,y,c) = sum(b(x+rb,y,c));
mean_b(x,y,c) = sum(mean_b_sx(x,y+rb,c));
mean_aUp(x,y,c) = upsample(mean_a)(x,y,c);
mean_bUp(x,y,c) = upsample(mean_b)(x,y,c);
q(x,y,c) = mean_aUp(x,y,c)*guidedF(x,y,c) + mean_bUp(x,y,c);
result(x,y,c) = Halide::cast<uint8_t>(Halide::clamp(q(x,y,c)*fac, 0.0f, 255.0f));
// Provide estimates on the inputs
guided.set_estimates({{0, 512}, {0, 512}, {0, 3}});
input.set_estimates({{0, 512}, {0, 512}, {0, 3}});
// Provide estimates on the parameters
eps.set_estimate(0.0004);
// Provide estimates on the pipeline output
result.set_estimates({{0, 512}, {0, 512}, {0, 3}});
/*/================
Image Size:512x512x3
CPU:I5-3210m
--------------------
Opencv: 32ms
Halide manually-tuned: 41ms
Halide auto-schedule: 6ms
================/*/
if(auto_schedule){
//nothing
}
else{
int vec=8;
Var x_outer, y_outer, x_inner, y_inner, tile_index;
result
.tile(x,y,x_inner,y_inner, 64,64)
.parallel(y)
;
mean_b.compute_root()
.tile(x,y,x_inner,y_inner,64,64)
.parallel(y);
mean_b_sx.compute_root()
.tile(x,y,x_inner,y_inner,128,64)
.reorder(y_inner,x_inner)
.parallel(y)
;
mean_a.compute_root()
.tile(x,y,x_inner,y_inner,64,64)
.parallel(y);
mean_a_sx.compute_root()
.tile(x,y,x_inner,y_inner,128,64)
.reorder(y_inner,x_inner)
.parallel(y)
;
a.compute_root();
b.compute_at(mean_b_sx, x)
.vectorize(x,vec);
corr_Ip.compute_root()
.tile(x,y,x_inner,y_inner,64,64)
.parallel(y);
corr_Ip_sx.compute_root()
.tile(x,y,x_inner,y_inner,128,64)
.reorder(y_inner,x_inner)
.parallel(y)
;
Ip.compute_at(corr_Ip_sx,x)
.vectorize(x,vec);
corr_I.compute_root()
.tile(x,y,x_inner,y_inner,64,64)
.parallel(y);
corr_I_sx.compute_root()
.tile(x,y,x_inner,y_inner,128,64)
.reorder(y_inner,x_inner)
.parallel(y)
;
II.compute_at(corr_I_sx,x)
.vectorize(x,vec);
mean_p.compute_root()
.tile(x,y,x_inner,y_inner,64,64)
.parallel(y);
mean_p_sx.compute_root()
.tile(x,y,x_inner,y_inner,128,64)
.reorder(y_inner,x_inner)
.parallel(y)
;
mean_I.compute_root()
.tile(x,y,x_inner,y_inner,64,64)
.parallel(y);
mean_I_sx.compute_root()
.tile(x,y,x_inner,y_inner,128,64)
.reorder(y_inner,x_inner)
.parallel(y)
;
inputF.compute_root();
guidedF.compute_root();
}
}
private:
//@TODO Copy from local_laplacian_generator.cpp
Var x, y, c, k;
// Downsample with a 1 3 3 1 filter
Func downsample(Func f) {
using Halide::_;
Func downx, downy;
downx(x, y, _) = (f(2 * x - 1, y, _) + 3.0f * (f(2 * x, y, _) + f(2 * x + 1, y, _)) + f(2 * x + 2, y, _)) / 8.0f;
downy(x, y, _) = (downx(x, 2 * y - 1, _) + 3.0f * (downx(x, 2 * y, _) + downx(x, 2 * y + 1, _)) + downx(x, 2 * y + 2, _)) / 8.0f;
return downy;
}
// Upsample using bilinear interpolation
Func upsample(Func f) {
using Halide::_;
Func upx, upy;
upx(x, y, _) = lerp(f(x / 2, y, _), f((x + 1) / 2, y, _), ((x % 2) * 2 + 1) / 4.0f);
upy(x, y, _) = lerp(upx(x, y / 2, _), upx(x, (y + 1) / 2, _), ((y % 2) * 2 + 1) / 4.0f);
return upy;
}
};
HALIDE_REGISTER_GENERATOR(FastGuidedFilter, fast_guided_filter)