forked from NVIDIA/cuda-samples
-
Notifications
You must be signed in to change notification settings - Fork 0
/
imageDenoising_nlm_kernel.cuh
153 lines (130 loc) · 6.75 KB
/
imageDenoising_nlm_kernel.cuh
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
/* Copyright (c) 2022, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
////////////////////////////////////////////////////////////////////////////////
// NLM kernel
////////////////////////////////////////////////////////////////////////////////
__global__ void NLM(TColor *dst, int imageW, int imageH, float Noise,
float lerpC, cudaTextureObject_t texImage) {
const int ix = blockDim.x * blockIdx.x + threadIdx.x;
const int iy = blockDim.y * blockIdx.y + threadIdx.y;
// Add half of a texel to always address exact texel centers
const float x = (float)ix + 0.5f;
const float y = (float)iy + 0.5f;
if (ix < imageW && iy < imageH) {
// Normalized counter for the NLM weight threshold
float fCount = 0;
// Total sum of pixel weights
float sumWeights = 0;
// Result accumulator
float3 clr = {0, 0, 0};
// Cycle through NLM window, surrounding (x, y) texel
for (float i = -NLM_WINDOW_RADIUS; i <= NLM_WINDOW_RADIUS; i++)
for (float j = -NLM_WINDOW_RADIUS; j <= NLM_WINDOW_RADIUS; j++) {
// Find color distance from (x, y) to (x + j, y + i)
float weightIJ = 0;
for (float n = -NLM_BLOCK_RADIUS; n <= NLM_BLOCK_RADIUS; n++)
for (float m = -NLM_BLOCK_RADIUS; m <= NLM_BLOCK_RADIUS; m++)
weightIJ += vecLen(tex2D<float4>(texImage, x + j + m, y + i + n),
tex2D<float4>(texImage, x + m, y + n));
// Derive final weight from color and geometric distance
weightIJ =
__expf(-(weightIJ * Noise + (i * i + j * j) * INV_NLM_WINDOW_AREA));
// Accumulate (x + j, y + i) texel color with computed weight
float4 clrIJ = tex2D<float4>(texImage, x + j, y + i);
clr.x += clrIJ.x * weightIJ;
clr.y += clrIJ.y * weightIJ;
clr.z += clrIJ.z * weightIJ;
// Sum of weights for color normalization to [0..1] range
sumWeights += weightIJ;
// Update weight counter, if NLM weight for current window texel
// exceeds the weight threshold
fCount += (weightIJ > NLM_WEIGHT_THRESHOLD) ? INV_NLM_WINDOW_AREA : 0;
}
// Normalize result color by sum of weights
sumWeights = 1.0f / sumWeights;
clr.x *= sumWeights;
clr.y *= sumWeights;
clr.z *= sumWeights;
// Choose LERP quotient basing on how many texels
// within the NLM window exceeded the weight threshold
float lerpQ = (fCount > NLM_LERP_THRESHOLD) ? lerpC : 1.0f - lerpC;
// Write final result to global memory
float4 clr00 = tex2D<float4>(texImage, x, y);
clr.x = lerpf(clr.x, clr00.x, lerpQ);
clr.y = lerpf(clr.y, clr00.y, lerpQ);
clr.z = lerpf(clr.z, clr00.z, lerpQ);
dst[imageW * iy + ix] = make_color(clr.x, clr.y, clr.z, 0);
}
}
extern "C" void cuda_NLM(TColor *d_dst, int imageW, int imageH, float Noise,
float lerpC, cudaTextureObject_t texImage) {
dim3 threads(BLOCKDIM_X, BLOCKDIM_Y);
dim3 grid(iDivUp(imageW, BLOCKDIM_X), iDivUp(imageH, BLOCKDIM_Y));
NLM<<<grid, threads>>>(d_dst, imageW, imageH, Noise, lerpC, texImage);
}
////////////////////////////////////////////////////////////////////////////////
// Stripped NLM kernel, only highlighting areas with different LERP directions
////////////////////////////////////////////////////////////////////////////////
__global__ void NLMdiag(TColor *dst, unsigned int imageW, unsigned int imageH,
float Noise, float lerpC,
cudaTextureObject_t texImage) {
const int ix = blockDim.x * blockIdx.x + threadIdx.x;
const int iy = blockDim.y * blockIdx.y + threadIdx.y;
// Add half of a texel to always address exact texel centers
const float x = (float)ix + 0.5f;
const float y = (float)iy + 0.5f;
if (ix < imageW && iy < imageH) {
// Normalized counter for the weight threshold
float fCount = 0;
// Cycle through NLM window, surrounding (x, y) texel
for (float i = -NLM_WINDOW_RADIUS; i <= NLM_WINDOW_RADIUS; i++)
for (float j = -NLM_WINDOW_RADIUS; j <= NLM_WINDOW_RADIUS; j++) {
// Find color distance between (x, y) and (x + j, y + i)
float weightIJ = 0;
for (float n = -NLM_BLOCK_RADIUS; n <= NLM_BLOCK_RADIUS; n++)
for (float m = -NLM_BLOCK_RADIUS; m <= NLM_BLOCK_RADIUS; m++)
weightIJ += vecLen(tex2D<float4>(texImage, x + j + m, y + i + n),
tex2D<float4>(texImage, x + m, y + n));
// Derive final weight from color and geometric distance
weightIJ =
__expf(-(weightIJ * Noise + (i * i + j * j) * INV_NLM_WINDOW_AREA));
// Increase the weight threshold counter
fCount += (weightIJ > NLM_WEIGHT_THRESHOLD) ? INV_NLM_WINDOW_AREA : 0;
}
// Choose LERP quotient basing on how many texels
// within the NLM window exceeded the LERP threshold
float lerpQ = (fCount > NLM_LERP_THRESHOLD) ? 1.0f : 0;
// Write final result to global memory
dst[imageW * iy + ix] = make_color(lerpQ, 0, (1.0f - lerpQ), 0);
};
}
extern "C" void cuda_NLMdiag(TColor *d_dst, int imageW, int imageH, float Noise,
float lerpC, cudaTextureObject_t texImage) {
dim3 threads(BLOCKDIM_X, BLOCKDIM_Y);
dim3 grid(iDivUp(imageW, BLOCKDIM_X), iDivUp(imageH, BLOCKDIM_Y));
NLMdiag<<<grid, threads>>>(d_dst, imageW, imageH, Noise, lerpC, texImage);
}