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
|
/*
* Copyright 2011-2017 Blender Foundation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
CCL_NAMESPACE_BEGIN
ccl_device_inline void kernel_filter_nlm_calc_difference(int x, int y,
int dx, int dy,
const ccl_global float *ccl_restrict weight_image,
const ccl_global float *ccl_restrict variance_image,
ccl_global float *difference_image,
int4 rect, int w,
int channel_offset,
float a, float k_2)
{
float diff = 0.0f;
int numChannels = channel_offset? 3 : 1;
for(int c = 0; c < numChannels; c++) {
float cdiff = weight_image[c*channel_offset + y*w+x] - weight_image[c*channel_offset + (y+dy)*w+(x+dx)];
float pvar = variance_image[c*channel_offset + y*w+x];
float qvar = variance_image[c*channel_offset + (y+dy)*w+(x+dx)];
diff += (cdiff*cdiff - a*(pvar + min(pvar, qvar))) / (1e-8f + k_2*(pvar+qvar));
}
if(numChannels > 1) {
diff *= 1.0f/numChannels;
}
difference_image[y*w+x] = diff;
}
ccl_device_inline void kernel_filter_nlm_blur(int x, int y,
const ccl_global float *ccl_restrict difference_image,
ccl_global float *out_image,
int4 rect, int w, int f)
{
float sum = 0.0f;
const int low = max(rect.y, y-f);
const int high = min(rect.w, y+f+1);
for(int y1 = low; y1 < high; y1++) {
sum += difference_image[y1*w+x];
}
sum *= 1.0f/(high-low);
out_image[y*w+x] = sum;
}
ccl_device_inline void kernel_filter_nlm_calc_weight(int x, int y,
const ccl_global float *ccl_restrict difference_image,
ccl_global float *out_image,
int4 rect, int w, int f)
{
float sum = 0.0f;
const int low = max(rect.x, x-f);
const int high = min(rect.z, x+f+1);
for(int x1 = low; x1 < high; x1++) {
sum += difference_image[y*w+x1];
}
sum *= 1.0f/(high-low);
out_image[y*w+x] = expf(-max(sum, 0.0f));
}
ccl_device_inline void kernel_filter_nlm_update_output(int x, int y,
int dx, int dy,
const ccl_global float *ccl_restrict difference_image,
const ccl_global float *ccl_restrict image,
ccl_global float *out_image,
ccl_global float *accum_image,
int4 rect, int w, int f)
{
float sum = 0.0f;
const int low = max(rect.x, x-f);
const int high = min(rect.z, x+f+1);
for(int x1 = low; x1 < high; x1++) {
sum += difference_image[y*w+x1];
}
sum *= 1.0f/(high-low);
if(out_image) {
accum_image[y*w+x] += sum;
out_image[y*w+x] += sum*image[(y+dy)*w+(x+dx)];
}
else {
accum_image[y*w+x] = sum;
}
}
ccl_device_inline void kernel_filter_nlm_construct_gramian(int fx, int fy,
int dx, int dy,
const ccl_global float *ccl_restrict difference_image,
const ccl_global float *ccl_restrict buffer,
ccl_global float *color_pass,
ccl_global float *variance_pass,
const ccl_global float *ccl_restrict transform,
ccl_global int *rank,
ccl_global float *XtWX,
ccl_global float3 *XtWY,
int4 rect,
int4 filter_rect,
int w, int h, int f,
int pass_stride,
int localIdx)
{
int y = fy + filter_rect.y;
int x = fx + filter_rect.x;
const int low = max(rect.x, x-f);
const int high = min(rect.z, x+f+1);
float sum = 0.0f;
for(int x1 = low; x1 < high; x1++) {
sum += difference_image[y*w+x1];
}
float weight = sum * (1.0f/(high - low));
int storage_ofs = fy*filter_rect.z + fx;
transform += storage_ofs;
rank += storage_ofs;
XtWX += storage_ofs;
XtWY += storage_ofs;
kernel_filter_construct_gramian(x, y,
filter_rect.z*filter_rect.w,
dx, dy, w, h,
pass_stride,
buffer,
color_pass, variance_pass,
transform, rank,
weight, XtWX, XtWY,
localIdx);
}
ccl_device_inline void kernel_filter_nlm_normalize(int x, int y,
ccl_global float *out_image,
const ccl_global float *ccl_restrict accum_image,
int4 rect, int w)
{
out_image[y*w+x] /= accum_image[y*w+x];
}
CCL_NAMESPACE_END
|
