Doubt about GetK function in Camera module

[satyajitghana]

What exactly is the difference between the GetK implementation in current OpenMVS and the old OpenMVS

Old OpenMVS

template <typename TYPE>
TMatrix<TYPE, 3, 3> GetK(MVS::Camera camera, uint32_t width, uint32_t height) {
  ASSERT(width > 0 && height > 0);
  const float fScale(camera.GetNormalizationScale(width, height));
  if (camera.K(0, 2) == 0 && camera.K(1, 2) == 0)
    return camera.ComposeK(TYPE(camera.K(0, 0) * fScale),
                           TYPE(camera.K(1, 1) * fScale), width, height);
  TMatrix<TYPE, 3, 3> fullK(TMatrix<TYPE, 3, 3>::IDENTITY);
  fullK(0, 0) = camera.K(0, 0) * fScale;
  fullK(1, 1) = camera.K(1, 1) * fScale;
  fullK(0, 2) = camera.K(0, 2) * fScale;
  fullK(1, 2) = camera.K(1, 2) * fScale;
  return fullK;
}

New OpenMVS

	template<typename TYPE>
	inline TMatrix<TYPE,3,3> GetK(uint32_t width, uint32_t height) const {
		ASSERT(width>0 && height>0);
		const float scale(GetNormalizationScale(width, height));
		if (K(0,2) != 0 || K(1,2) != 0)
			return GetScaledK(scale);
		ASSERT(ISZERO(K(0,1)));
		return ComposeK(
			TYPE(K(0,0)*scale), TYPE(K(1,1)*scale),
			width, height );
	}

I had some issues recently trying to create a mesh from point cloud, and it turns out there is some problem in the GetK method in the new openmvs. I switched to the old implementation and it started to work.

What was changed about GetK that it behaves so differently?

[cdcseacave]

it was a very old change, but as far as I remember, the new logic scales differently the principal point, and if you are using an old format the principal point was stored in a different way