最近ブログを更新していませんでした。
画像処理やAI系で個人で数時間でできそうな面白いネタがあまりないんだよねぇ。
暇なので、相棒のテレビをみていたら、動画の顔を別の人の顔に変えるDeepfakeというソフトがあるらしい。
なんかいろいろ悪いことに使われそう。
FaceSwapという技術で動画から静止画を切り出して一枚一枚変換しているのね。
ソースを見てみたら、おじさんの大嫌いなPythonで書かれているじゃないか!
C++でできないかと思っていたところC++のFaceSwapのソースを見つけました。
https://learnopencv.com/face-swap-using-opencv-c-python/
これならば数時間で改造して遊ぶことができる。
改造して、以前作った最小のOpenCVで動くようにしてみました。
みんな、トランプ大統領の顔を変えて実験するのね。
おーなんかできた。ソースコードも意外と短い。
----------------------------
#include "opencv2_core.hpp"
#include "opencv2_imgproc.hpp"
#include "opencv2_imgproc_imgproc_c.h"
#include "opencv2_imgcodecs.hpp"
#include "opencv2_photo.hpp"
#include <iostream>
#include <fstream>
#include <string>
using namespace cv;
using namespace std;
//Read points from text file
vector<Point2f> readPoints(string pointsFileName){
vector<Point2f> points;
ifstream ifs (pointsFileName.c_str());
float x, y;
int count = 0;
while(ifs >> x >> y)
{
points.push_back(Point2f(x,y));
}
return points;
}
// Apply affine transform calculated using srcTri and dstTri to src
void applyAffineTransform(Mat &warpImage, Mat &src, vector<Point2f> &srcTri, vector<Point2f> &dstTri)
{
// Given a pair of triangles, find the affine transform.
Mat warpMat = getAffineTransform( srcTri, dstTri );
// Apply the Affine Transform just found to the src image
warpAffine( src, warpImage, warpMat, warpImage.size(), INTER_LINEAR, BORDER_REFLECT_101);
}
// Calculate Delaunay triangles for set of points
// Returns the vector of indices of 3 points for each triangle
static void calculateDelaunayTriangles(Rect rect, vector<Point2f> &points, vector< vector<int> > &delaunayTri){
// Create an instance of Subdiv2D
Subdiv2D subdiv(rect);
// Insert points into subdiv
for( vector<Point2f>::iterator it = points.begin(); it != points.end(); it++)
subdiv.insert(*it);
vector<Vec6f> triangleList;
subdiv.getTriangleList(triangleList);
vector<Point2f> pt(3);
vector<int> ind(3);
for( size_t i = 0; i < triangleList.size(); i++ )
{
Vec6f t = triangleList[i];
pt[0] = Point2f(t[0], t[1]);
pt[1] = Point2f(t[2], t[3]);
pt[2] = Point2f(t[4], t[5 ]);
if ( rect.contains(pt[0]) && rect.contains(pt[1]) && rect.contains(pt[2])){
for(int j = 0; j < 3; j++)
for(size_t k = 0; k < points.size(); k++)
if(abs(pt[j].x - points[k].x) < 1.0 && abs(pt[j].y - points[k].y) < 1)
ind[j] = k;
delaunayTri.push_back(ind);
}
}
}
// Warps and alpha blends triangular regions from img1 and img2 to img
void warpTriangle(Mat &img1, Mat &img2, vector<Point2f> &t1, vector<Point2f> &t2)
{
Rect r1 = boundingRect(t1);
Rect r2 = boundingRect(t2);
// Offset points by left top corner of the respective rectangles
vector<Point2f> t1Rect, t2Rect;
vector<Point> t2RectInt;
for(int i = 0; i < 3; i++)
{
t1Rect.push_back( Point2f( t1[i].x - r1.x, t1[i].y - r1.y) );
t2Rect.push_back( Point2f( t2[i].x - r2.x, t2[i].y - r2.y) );
t2RectInt.push_back( Point(t2[i].x - r2.x, t2[i].y - r2.y) ); // for fillConvexPoly
}
// Get mask by filling triangle
Mat mask = Mat::zeros(r2.height, r2.width, CV_32FC3);
fillConvexPoly(mask, t2RectInt, Scalar(1.0, 1.0, 1.0), 16, 0);
// Apply warpImage to small rectangular patches
Mat img1Rect;
img1(r1).copyTo(img1Rect);
Mat img2Rect = Mat::zeros(r2.height, r2.width, img1Rect.type());
applyAffineTransform(img2Rect, img1Rect, t1Rect, t2Rect);
multiply(img2Rect,mask, img2Rect);
multiply(img2(r2), Scalar(1.0,1.0,1.0) - mask, img2(r2));
img2(r2) = img2(r2) + img2Rect;
}
int main( int argc, char** argv)
{
//Read input images
string filename1 = "ted_cruz.jpg";
string filename2 = "donald_trump.jpg";
Mat img1 = imread(filename1);
Mat img2 = imread(filename2);
Mat img1Warped = img2.clone();
//Read points
vector<Point2f> points1, points2;
points1 = readPoints(filename1 + ".txt");
points2 = readPoints(filename2 + ".txt");
//convert Mat to float data type
img1.convertTo(img1, CV_32F);
img1Warped.convertTo(img1Warped, CV_32F);
// Find convex hull
vector<Point2f> hull1;
vector<Point2f> hull2;
vector<int> hullIndex;
convexHull(points2, hullIndex, false, false);
for(int i = 0; i < hullIndex.size(); i++)
{
hull1.push_back(points1[hullIndex[i]]);
hull2.push_back(points2[hullIndex[i]]);
}
// Find delaunay triangulation for points on the convex hull
vector< vector<int> > dt;
Rect rect(0, 0, img1Warped.cols, img1Warped.rows);
calculateDelaunayTriangles(rect, hull2, dt);
// Apply affine transformation to Delaunay triangles
for(size_t i = 0; i < dt.size(); i++)
{
vector<Point2f> t1, t2;
// Get points for img1, img2 corresponding to the triangles
for(size_t j = 0; j < 3; j++)
{
t1.push_back(hull1[dt[i][j]]);
t2.push_back(hull2[dt[i][j]]);
}
warpTriangle(img1, img1Warped, t1, t2);
}
// Calculate mask
vector<Point> hull8U;
for(int i = 0; i < hull2.size(); i++)
{
Point pt(hull2[i].x, hull2[i].y);
hull8U.push_back(pt);
}
Mat mask = Mat::zeros(img2.rows, img2.cols, img2.depth());
fillConvexPoly(mask,&hull8U[0], hull8U.size(), Scalar(255,255,255));
// Clone seamlessly.
Rect r = boundingRect(hull2);
Point center = (r.tl() + r.br()) / 2;
Mat output;
img1Warped.convertTo(img1Warped, CV_8UC3);
seamlessClone(img1Warped,img2, mask, center, output, NORMAL_CLONE);
imwrite("output.jpg", output);
//imshow("Face Swapped", output);
//waitKey(0);
//destroyAllWindows();
return 1;
}
----------------------------
0 件のコメント:
コメントを投稿