Matrix Multiplication wrong in opencv?

Question

I know that there are 3 ways to multiply matrices in opencv, one is:

cvGEMM(M1, M1T, 1, NULL, 0, Mult, 0);

the other:

cvMulTransposed(M1, Mult, 0);

and third:

cvMul(M1, M1T, Mult);

I am basically multiplying a matrix by its transpose. After multiplication the matrices should be symmetrical.. but each of the methods above is giving me an unsymmetrical and completely wrong output.. here is the original (M1) matrix:

rows: 5
   cols: 50
   dt: f
   data: [ 60., 89., 86., 102., 58., 51., 143., 187., 140., 64., 80.,
       169., 184., 172., 67., 90., 174., 191., 175., 41., 97., 86., 171.,
       104., 87., 145., 164., 176., 157., 119., 176., 177., 180., 179.,
       150., 44., 182., 148., 182., 38., 48., 197., 171., 171., 39., 48.,
       169., 163., 141., 39., 93., 149., 169., 184., 56., 120., 166.,
       182., 168., 144., 132., 172., 187., 173., 157., 164., 157., 152.,
       166., 172., 175., 157., 133., 170., 140., 180., 164., 173., 175.,
       152., 178., 176., 166., 143., 158., 177., 186., 172., 138., 141.,
       182., 191., 169., 159., 34., 185., 154., 155., 171., 32., 81.,
       154., 163., 112., 52., 126., 163., 183., 165., 53., 136., 176.,
       184., 174., 51., 148., 173., 178., 139., 160., 158., 147., 163.,
       154., 135., 146., 157., 181., 161., 79., 39., 172., 187., 174.,
       45., 44., 187., 153., 173., 39., 46., 187., 145., 160., 36., 38.,
       177., 155., 150., 37., 81., 154., 163., 112., 52., 126., 163.,
       183., 165., 53., 136., 176., 184., 174., 51., 148., 173., 178.,
       139., 160., 158., 147., 163., 154., 135., 146., 157., 181., 161.,
       79., 39., 172., 187., 174., 45., 44., 187., 153., 173., 39., 46.,
       187., 145., 160., 36., 38., 177., 155., 150., 37., 76., 165.,
       195., 110., 100., 131., 180., 188., 184., 73., 146., 179., 193.,
       183., 96., 117., 148., 146., 185., 97., 104., 147., 157., 174.,
       172., 129., 171., 177., 181., 173., 123., 126., 186., 194., 171.,
       91., 74., 109., 194., 145., 33., 117., 166., 180., 176., 35., 77.,
       155., 152., 177. ]

but multiplying this by its transpose gives this:

rows: 5
   cols: 5
   dt: f
   data: [ 927321., 1014163., 923303., 923303., 947641., 1014163.,
       1260101., 1062130., 1062130., 1102823., 923303., 1062130.,
       1004488., 1004488., 990651., 923303., 1062130., 1004488.,
       1004488., 990651., 947641., 1102823., 990651., 990651., 1116004. ]

the values are unsymmetrical, also how they got so large is what I dont understand?

cvMultransposed and cvGEMM are giving the same result as copied above.. cvMul is giving some runtime error and returning garbage values. any suggestions?

Most of the time that I find a "wrong" result, it's because I didn't understand the API or what the correct answer would look like. A bug in publicly available software is less likely. A large user base almost guarantees that someone else would have spotted the bug before I did and gotten it fixed. I'm almost certain that the problem lies with you, your understanding, and your code. — duffymo
im not saying that its a bug, and im sure its a mistake by me, but ive been working on it all day and couldnt fix it, that is why I asked for help and suggestions here. — StuckInPhDNoMore
@Faraz Khan: You should post some code that generates this unexpected behaviour so we can understand why. — Tugrul Ates

etarion etarion · Accepted Answer · 2011-07-25T13:48:29

In addition to the expected large output values (that japreiss already told you about), the matrix that you get IS symmetrical.

I've aligned it for better readability:

reshape(a,5,5)

ans =

      927321     1014163      923303      923303      947641
     1014163     1260101     1062130     1062130     1102823
      923303     1062130     1004488     1004488      990651
      923303     1062130     1004488     1004488      990651
      947641     1102823      990651      990651     1116004

So, there is nothing wrong here, and you should double-check that X is indeed the case if you go and ask why X occurs in future ...

Matrix Multiplication wrong in opencv?

2 Answers