[SFM] Installation and examples documentation suggestions

[adrien-jacquot]

System information (version)

• OpenCV => 4.0.1
• Operating System / Platform => Ubuntu 18.04
• Compiler => g++

Detailed description

I had troubles following the installation guidelines of SFM. In my opinion here what's missing:

• VTK has to be taken into account during OpenCV compilation. Otherwise (at least in my case), the SFM examples are not compiled.
  So the OpenCV compilation looks like this:

cmake -D OPENCV_EXTRA_MODULES_PATH=<opencv_contrib>/modules   -D BUILD_EXAMPLES=ON –D WITH_VTK=ON <opencv_source_directory>

Additionally, in my case, the way to install VTK was simply this (and not the whole process mentioned in the VTK wiki):

sudo apt-get isntall libvtk7-dev

Reference: http://answers.opencv.org/question/32502/opencv-249-viz-module-not-there/?answer=32847#post-id-32847

• The input parameters of example_sfm_scene_reconstruction are a bit tedious to understand.

The focal length is expressed in pixels, and for non-experts like me it requires a bit of research to understand how to convert a mm focal length to a pixel one.
I used this link to do the conversion: http://answers.opencv.org/question/17076/conversion-focal-distance-from-mm-to-pixels/

Same goes for the camera intrinsic, I'm currently using the center of the images (width/2, height/2), but it is a bit unclear to me if this is the right thing to do

I think it'd be nice to elaborate on these 2 points in the documentation :)

[alalek]

how to convert a mm focal length

Camera intrinsics should be calibrated instead of hardcoding them (usually we don't have ideal camera/optics). Take a look on interactive calibration tool.
Non-accurate values may harm final result of algorithms.

[adrien-jacquot]

Thanks!
I think that the SFM module re-evaluates the intrinsics at runtime, but the final estimation (printed in the console) is influenced by the value passed by the user (initialization?).

[LaurentBerger]

@adrien-jacquot it is levenberg marquardt solver