[vtkusers] Image "swimming" with vtkImagePlaneWidget?

[Cory Quammen]

I posted a patch on gerrit here:

http://review.source.kitware.com/#/c/6425/

On Thu, Jul 12, 2012 at 4:46 PM, Cory Quammen <cquammen at cs.unc.edu> wrote:
> [bringing in the VTK developers list]
>
> I found some suspect code in vtkImagePlaneWidget.cxx:
>
> 1661   double planeOrigin[4];
> 1662   this->PlaneSource->GetOrigin(planeOrigin);
> 1663
> 1664   planeOrigin[3] = 1.0;
> 1665   double originXYZW[4];
> 1666   this->ResliceAxes->MultiplyPoint(planeOrigin, originXYZW);
> 1667
> 1668   this->ResliceAxes->Transpose();
> 1669   double neworiginXYZW[4];
> 1670   this->ResliceAxes->MultiplyPoint(originXYZW, neworiginXYZW);
> 1671
> 1672   this->ResliceAxes->SetElement(0,3,neworiginXYZW[0]);
> 1673   this->ResliceAxes->SetElement(1,3,neworiginXYZW[1]);
> 1674   this->ResliceAxes->SetElement(2,3,neworiginXYZW[2]);
> 1675
> 1676   this->Reslice->SetResliceAxes(this->ResliceAxes);
>
> For reasons I don't understand, the origin of the plane is transformed
> by the ResliceAxes transformation matrix, then the result is then
> multiplied back through the transpose of the ResliceAxes matrix,
> which, if I'm not mistaken, should be the inverse of the matrix if it
> is orthonormal (which it appears to be). That means the neworiginXYZW
> should be equivalent to planeOrigin and we can skip all the calls to
> MultiplyPoint(). The swimming, I guess, results from numerical error
> in the forward and inverse matrix transforms of the plane origin.
>
> I suspect the code should be
>
> double planeOrigin[4];
> this->PlaneSource->GetOrigin(planeOrigin);
> planeOrigin[3] = 1.0;
>
> this->ResliceAxes->Transpose();
>
> this->ResliceAxes->SetElement(0,3,planeOrigin[0]);
> this->ResliceAxes->SetElement(1,3,planeOrigin[1]);
> this->ResliceAxes->SetElement(2,3,planeOrigin[2]);
>
> this->Reslice->SetResliceAxes(this->ResliceAxes);
>
> If I change the code to the above, the swimming no longer occurs.
>
> I'll submit a gerrit patch if my proposed fix looks reasonable.
>
> Thanks,
> Cory
>
> On Thu, Jul 12, 2012 at 3:03 PM, Cory Quammen <cquammen at cs.unc.edu> wrote:
>> By the way, I'm building VTK from git commit
>>
>> 2d86fe9782e36e6a5efad240bc5acef746f058ab
>>
>> which was created June 29, 2012.
>>
>> Thanks,
>> Cory
>>
>> On Thu, Jul 12, 2012 at 2:55 PM, Cory Quammen <cquammen at cs.unc.edu> wrote:
>>> Hi all,
>>>
>>> I'm using vtkImagePlaneWidget in an application and I noticed a
>>> problem when rotating the widget. The image appears to "swim" relative
>>> to the widget plane when rotating the plane, jumping around in
>>> seemingly random directions rather than varying smoothly as the plane
>>> is rotated. In-plane and out-of-plane translation seems fine, as does
>>> resizing the plane. I see this problem on both a Mac with an AMD
>>> graphics card and a Windows laptop with an NVIDIA GPU.
>>>
>>> Has anyone else seen this behavior?
>>>
>>> Thanks,
>>> Cory
>>>
>>> --
>>> Cory Quammen
>>> Research Associate
>>> Department of Computer Science
>>> The University of North Carolina at Chapel Hill
>>
>>
>>
>> --
>> Cory Quammen
>> Research Associate
>> Department of Computer Science
>> The University of North Carolina at Chapel Hill
>
>
>
> --
> Cory Quammen
> Research Associate
> Department of Computer Science
> The University of North Carolina at Chapel Hill



-- 
Cory Quammen
Research Associate
Department of Computer Science
The University of North Carolina at Chapel Hill