amiroor / flannimagecompletion Goto Github PK

Hi, Amir. When using your code, the boundary effects are obvious (with the default parameters in main.cpp, shown in 1). Besides, the functions smoothFn and dataFn seem not to work. Do I miss something?
Could you share about the differences between your implementation and Kaiming's paper?

Hi , I found your code very useful for me to understand this algorithm, but I find an important mistake may lead uncomfortable result.

                    Vec3b v1_a = img_ycb.at<Vec3b>( x1_s_a.y, x1_s_a.x);
                    Vec3b v1_b = img_ycb.at<Vec3b>( x1_s_b.y, x1_s_b.x);
                    Vec3b diff1 = v1_a - v1_b;

                    retMe += diff1[0]*diff1[0] + diff1[1]*diff1[1] + diff1[2]*diff1[2];

In openCV, Vec3b is uchar， it will cause a overflow easly and mess the result.
when I correct this , I get a much better and stable result.