-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.cpp
180 lines (151 loc) · 6.59 KB
/
main.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
/**
* CS61064 - High Perfomance Parallel Programming
* OpenMP/MPI - Assignment 1
* Parallel Image Morphing by Affine Transformations
*
* Author: Utkarsh Patel (18EC35034)
*
* Main driver code
*/
#include <iostream>
#include <iomanip>
#include <vector>
#include <unordered_map>
#include <opencv2/core/core.hpp>
#include <opencv2/imgcodecs.hpp>
#include <opencv2/imgproc/imgproc.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <omp.h>
#include "atx.hpp"
#include "utils.hpp"
int main(int argc, char *argv[])
{
if (argc != 6) {
std::cout << "Use following args:\n" \
"- arg1: path to image\n" \
"- arg2: path to point list 1\n" \
"- arg3: path to point list 2\n" \
"- arg4: path to output image\n" \
"- arg5: dropout probability";
return 0;
}
/* Read image and list of points */
cv::Mat image1, image2;
std::vector<cv::Point2f> points1_all, points2_all;
atx::utils::lookup_table lookup;
atx::utils::read_image(argv[1], argv[2], image1, points1_all);
atx::utils::read_image(argv[1], argv[3], image2, points2_all);
/* Drop some points as per dropout probability */
double dropout = std::stod(argv[5]);
auto retained_indices = atx::utils::random_choice(points1_all.size(), dropout);
std::vector<cv::Point2f> points1(retained_indices.size()), \
points2(retained_indices.size());
for (std::size_t i = 0; i < retained_indices.size(); i++) {
auto idx = retained_indices[i];
points1[i] = points1_all[idx];
points2[i] = points2_all[idx];
}
atx::utils::create_lookup_table(points1, lookup);
/* Perform triangulation */
auto subdiv = atx::utils::get_triangulation(image1, points1);
/* Get triangles */
std::vector<cv::Vec6f> triangles1, triangles2;
subdiv.getTriangleList(triangles1);
std::size_t n_triangles = triangles1.size();
triangles2.resize(n_triangles);
/* For mapping triangles from 1st image to triangles of 2nd image */
std::vector<std::vector<int>> index(n_triangles, std::vector<int>(3));
/* Affine transform matrices for each triangle pair */
std::vector<std::vector<std::vector<double>>> affine_transforms(n_triangles);
std::vector<std::pair<std::pair<int, int>, std::pair<int, int>>> bounds(n_triangles);
#if defined(_OPENMP)
double start_time = omp_get_wtime();
#else
auto start_time = std::chrono::steady_clock::now();
#endif
#pragma omp parallel default(shared)
{
std::size_t i, j;
int y, x;
/* Create index to find corresponding triangle pairs
* No collapse clause used to increase cache hits per thread
*/
#pragma omp for schedule(static)
for (i = 0; i < n_triangles; i++) {
for (j = 0; j < 3; j++) {
index[i][j] = lookup[std::make_pair(triangles1[i][j << 1], triangles1[i][j << 1 | 1])];
}
}
/* Reorder triangles so that triangles1[i] and triangles2[i] belong to
* the same pair
*/
#pragma omp for schedule(static)
for (i = 0; i < index.size(); i++) {
triangles1[i] = cv::Vec6f({points1[index[i][0]].x, points1[index[i][0]].y,
points1[index[i][1]].x, points1[index[i][1]].y,
points1[index[i][2]].x, points1[index[i][2]].y});
triangles2[i] = cv::Vec6f({points2[index[i][0]].x, points2[index[i][0]].y,
points2[index[i][1]].x, points2[index[i][1]].y,
points2[index[i][2]].x, points2[index[i][2]].y});
}
/* Compute affine transformation matrix for each triangle pair */
#pragma omp for schedule(static)
for (i = 0; i < n_triangles; i++) {
affine_transforms[i] = atx::core::get_affine_transform(triangles2[i], triangles1[i]);
bounds[i] = atx::core::get_triangle_bounds(triangles2[i]);
}
/* Here, scheduling is made dynamic as workload in each iteration
* is not same. Chunk size is set to 1.
*/
#pragma omp for schedule(dynamic)
for (i = 0; i < n_triangles; i++) {
/* Get bounds for each triangle */
int ymin = bounds[i].second.first, ymax = bounds[i].second.second;
int xmin = bounds[i].first.first, xmax = bounds[i].first.second;
for (y = ymin; y <= ymax; y++) {
for (x = xmin; x <= xmax; x++) {
if (!atx::core::is_inside_triangle(x, y, triangles2[i])) {
/* Point lies outside the triangle */
continue;
}
if (x < 0 || x >= image2.cols || y < 0 || y >= image2.rows) {
/* Point cannot be accessed in the image */
continue;
}
/* Create a vector V = [Vx, Vy, 1] */
std::vector<std::vector<double>> V(3, std::vector<double>(1, 1));
V[0][0] = static_cast<double>(x);
V[1][0] = static_cast<double>(y);
/* Get transformed vector W = MV = [Wx, Wy, 1] */
auto W = atx::core::matrix_multiply(affine_transforms[i], V);
int tx = static_cast<int>(W[0][0] + 0.5);
int ty = static_cast<int>(W[1][0] + 0.5);
if (!atx::core::is_inside_triangle(tx, ty, triangles1[i])) {
/* Point lies outside the triangle */
continue;
}
if (tx < 0 || tx >= image1.cols || ty < 0 || ty >= image1.rows) {
/* Point cannot be accessed in the image */
continue;
}
image2.at<cv::Vec3b>(y, x) = image1.at<cv::Vec3b>(ty, tx);
}
}
}
}
double elapsed_time;
#if defined(_OPENMP)
double stop_time = omp_get_wtime();
elapsed_time = (stop_time - start_time) * 1e3;
#else
auto stop_time = std::chrono::steady_clock::now();
elapsed_time = std::chrono::duration_cast<std::chrono::duration<double>> \
(stop_time - start_time).count() * 1e3;
#endif
std::cout << std::fixed << std::setprecision(2) << "Finished parallel "\
"region in " << elapsed_time << "ms [#points: " << points1.size() \
<< "] [#triangles: " << n_triangles << "]" << std::endl;
/* Write image to output file*/
cv::imwrite(argv[4], image2);
return 0;
}