TDA2XX: SFM_POSE.bin format

No, It is 4 byte float. I can only read reasonanble values with 4 byte float, where the first 9 data agree with camera's inter parameter. If I read is as 8 byte double, all data messed up.

Therefore, it must be 4 byte. It is why I raise this question, that (9+12N)*4 can not equal the total file size. Do I have a corrupted SFM_POSE.bin file. If that is the case, can you give me a good one?

Regards,

Xinlu

Hi,

I suspect some corruption in this file. I believe you must have parsed this file as per the data format given by me earlier. Can you please share the parsed information in float 4 byte format ?

I can check the validity of the data and will let you know my findings.

Regards

Deepak Poddar

the first 45 data in my SFM_POSE.bin is

(array([ 6.66660e+02,  0.00000e+00,  6.40000e+02,  0.00000e+00,
        6.66660e+02,  3.60000e+02,  0.00000e+00,  0.00000e+00,
        1.00000e+00,  1.00000e+00,  0.00000e+00,  0.00000e+00,
        0.00000e+00,  0.00000e+00,  1.00000e+00,  0.00000e+00,
        0.00000e+00,  0.00000e+00,  0.00000e+00,  1.00000e+00,
        0.00000e+00,  1.00000e+00,  0.00000e+00,  0.00000e+00,
        0.00000e+00,  0.00000e+00,  1.00000e+00,  0.00000e+00,
        0.00000e+00,  0.00000e+00,  0.00000e+00,  1.00000e+00,
        0.00000e+00,  9.99994e-01,  3.21913e-03, -1.14692e-03,
        1.25308e-01, -3.21884e-03,  9.99995e-01,  2.50728e-04,
       -5.26145e-02,  1.14772e-03, -2.47035e-04,  9.99999e-01,
       -9.90721e-01], dtype=float32),

the last 24 data is:

array([ 9.96781e-01,  2.79001e-02,  7.51552e-02,  1.44736e+00,
       -3.40575e-02,  9.96056e-01,  8.19354e-02, -1.38838e+01,
       -7.25727e-02, -8.42313e-02,  9.93800e-01, -1.04500e+02,
        9.96781e-01,  2.79001e-02,  7.51552e-02,  1.44736e+00,
       -3.40575e-02,  9.96056e-01,  8.19354e-02, -1.38838e+01,
       -7.25727e-02, -8.42313e-02,  9.93800e-01, -1.04500e+02],
      dtype=float32))

I do not know how to attach the original file in this post.

Thanks,

Xinlu

Hi Xinlu,

to attach file when you click "Reply" button there is "Insert Code, Attach Files and more..." to the bottom right which you can use.

Regards,
Yordan

I tried it. It won't allow me to post after attach the file. Not sure why. Might because that it is a binary file. the file size if not too big.

Xinlu

Hi,

I did some check on data, and I could not find any problem. However question still remains why this size is not matching formula of (9+12*N)*4.

first 9 elements [666.66   0.0   6.40000e+02 0.0         6.66660e+02   3.60000e+02   0.0   0.0         1.0] fits for K = [fx, 0 , cx, 0, fy, cy, 0, 0, 1].

Next 12 elements

[  1.0   0.0   0.0         0.0
  0.0   1.0   0.0         0.0
  0.0   0.0   1.0         0.0]

is first P0 matrix of size 3x4.

Last 12 element

Pn = [        9.96781e-01   2.79001e-02   7.51552e-02   1.44736e+00
       -3.40575e-02   9.96056e-01   8.19354e-02 -1.39E+01
       -7.25727e-02 -8.42E-02   9.93800e-01 -1.05E+02]

is also another valid pose matrix as its top-left 3x3 matrix is rotation matrix

R = [        9.96781e-01   2.79001e-02   7.51552e-02
       -3.40575e-02   9.96056e-01   8.19354e-02
       -7.25727e-02 -8.42E-02   9.93800e-01] 

This R is valid matrix, as det(R) is 1.0.

Another R matrix corresponding to P2 is

R= [  .999994   .00321913 -0.00114692
-0.00321884   .999995   .000250728
  .00114772 -0.000247035   .999999]

and determinant of that also is 1.0.

You can check this for all R matrices. det(R) should be 1, for all pose matrices.

I am checking internally why the size of this file is not matching the formula, but meanwhile you can check R matrix consistency on all 12 element pose matrices.

Also I noticed that last pose matrix is same as second last pose matrix. Also first pose matrix is same as second one. Is this two time replication happening for all the pose matrices.

Regards

Deepak Poddar