No subject
Mon Dec 15 09:51:36 EST 2008
compares the true transform parameters not to exceed thresholds --<br>
0.025 for scale and 1.0 for translations.</blockquote><div><br>But as Dan m=
entioned its hard for most transforms. The right way to go about it is gene=
rate the target images manually with a known transform. Then after registra=
tion, you'd multiply the obtained transform with the known transform=
9;s inverse and check the L2 norm / bending energy or its equivalent agains=
t a tolerance.<br>
<br>Given the work it requires, its simpler to just compute the RMSE over s=
ource/target images.<br><br></div><blockquote class=3D"gmail_quote" style=
=3D"border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; p=
adding-left: 1ex;">
<br><div><div class=3D"Wj3C7c">
On Wed, Jan 28, 2009 at 11:30 AM, Bill Lorensen <<a href=3D"mailto:bill.=
lorensen at gmail.com">bill.lorensen at gmail.com</a>> wrote:<br>
> Also, their can be multiple baselines for a given test. If you look at=
:<br>
><br>
> <a href=3D"http://public.kitware.com/cgi-bin/viewcvs.cgi/Testing/Data/=
Baseline/Registration/?root=3DInsight" target=3D"_blank">http://public.kitw=
are.com/cgi-bin/viewcvs.cgi/Testing/Data/Baseline/Registration/?root=3DInsi=
ght</a><br>
><br>
> you will see that some tests have multiple baselines because of the<br=
>
> type of variability discussed in this thread.<br>
><br>
> For example, ImageRegistration13Test has 5 baselines.<br>
><br>
> Bill<br>
><br>
> On Wed, Jan 28, 2009 at 11:14 AM, Blezek, Daniel J.<br>
> <<a href=3D"mailto:Blezek.Daniel at mayo.edu">Blezek.Daniel at mayo.edu</=
a>> wrote:<br>
>> Just 2 cents here.<br>
>><br>
>> 1) For a linear registration, you could expect the registration re=
sults<br>
>> to be ~0.1 degrees and ~0.1 mm cross platform. The major sou=
rce of this<br>
>> problem is differences in floating point representations under dif=
ferent<br>
>> compilers/hardware. I shouldn't worry about this, it'=
;d be difficult for<br>
>> a human to see. For something like BSplines, you might have =
a bit more<br>
>> error.<br>
>><br>
>> 2) A very simple way to do this is to do a forward transformation =
using<br>
>> one transform (ground truth), then the inverse transform of the ne=
wly<br>
>> calculated transform. I think this is called Target Registra=
tion Error<br>
>> by Fitzpatrick and Co., but you should look it up. This is w=
here you<br>
>> need to decide on your tolerance. As Karthik mentioned, only=
simple ITK<br>
>> transforms have an inverse, which is really a shame. They ar=
e so<br>
>> useful, even if they are numeric.<br>
>><br>
>> 2a) I suppose you could forward transform the same point using two=
<br>
>> different transforms and see how far apart they are. This se=
ems<br>
>> reasonable, but you'd have to sample a bunch of points to acco=
unt for<br>
>> rotation, and transform centers, etc. And you'd only get=
a distance<br>
>> measure, not a rotational measure.<br>
>><br>
>> For transforms with an inverse, you can do what you are asking, an=
d it<br>
>> would be a valuable contribution to ITK, but it's not general,=
as not<br>
>> all transforms support an inverse. And you could always test=
the<br>
>> transform you care about...<br>
>><br>
>> Incoherent as usual,<br>
>> -dan<br>
>><br>
>> -----Original Message-----<br>
>> From: <a href=3D"mailto:insight-users-bounces at itk.org">insight-use=
rs-bounces at itk.org</a><br>
>> [mailto:<a href=3D"mailto:insight-users-bounces at itk.org">insight-u=
sers-bounces at itk.org</a>] On Behalf Of Andriy Fedorov<br>
>> Sent: Tuesday, January 27, 2009 6:18 PM<br>
>> To: ITK Users<br>
>> Cc: Miller, James V (GE, Research)<br>
>> Subject: [Insight-users] Rigid registration testing and reproducib=
ility<br>
>><br>
>> Hi,<br>
>><br>
>> I would like to do regression tests of rigid registration with rea=
l<br>
>> images, and compare the result with the baseline transform. Here a=
re two<br>
>> related questions.<br>
>><br>
>> 1) ITK registration experts, could you speculate on what is the no=
rmal<br>
>> variability in the rigid registration results run with the same<br=
>
>> parameters, metric initialized with the same seed, when run across=
<br>
>> different platforms? What could be the sources of this variability=
,<br>
>> given the same initialization and same parameters?<br>
>><br>
>> 2) If I understand correctly, the current testing of registration =
that<br>
>> comes with ITK generates a synthetic image with known transform<br=
>
>> parameters, which are compared with the registration-derived trans=
forms.<br>
>><br>
>> The testing framework implemented in itkTestMain.h allows to compa=
re<br>
>> pixelwise two images, but this does not seem to be practical for<b=
r>
>> registration regression testing, because of the variability in the=
<br>
>> registration results I mentioned earlier. Accounting for this<br>
>> variability using tolerance values of just 1 pixel hugely increase=
s the<br>
>> test runtime, but in my experience, the comparison may still fail.=
<br>
>><br>
>> Would it not make sense to augment itkTestMain.h with the capabili=
ty to<br>
>> compare not only images, but transforms? Is this a valid feature<b=
r>
>> request, or am I missing something?<br>
>><br>
>> Thanks<br>
>><br>
>> Andriy Fedorov<br>
>> _______________________________________________<br>
>> Insight-users mailing list<br>
>> <a href=3D"mailto:Insight-users at itk.org">Insight-users at itk.org</a>=
<br>
>> <a href=3D"http://www.itk.org/mailman/listinfo/insight-users" targ=
et=3D"_blank">http://www.itk.org/mailman/listinfo/insight-users</a><br>
>> _______________________________________________<br>
>> Insight-users mailing list<br>
>> <a href=3D"mailto:Insight-users at itk.org">Insight-users at itk.org</a>=
<br>
>> <a href=3D"http://www.itk.org/mailman/listinfo/insight-users" targ=
et=3D"_blank">http://www.itk.org/mailman/listinfo/insight-users</a><br>
>><br>
><br>
_______________________________________________<br>
Insight-users mailing list<br>
<a href=3D"mailto:Insight-users at itk.org">Insight-users at itk.org</a><br>
<a href=3D"http://www.itk.org/mailman/listinfo/insight-users" target=3D"_bl=
ank">http://www.itk.org/mailman/listinfo/insight-users</a><br>
</div></div></blockquote></div><br><br clear=3D"all"><br>-- <br>Karthik Kri=
shnan<br>R&D Engineer,<br>Kitware Inc.<br>Ph: 518 371 3971 x119<br>Fax:=
518 371 3971<br>
--001485f7c1aa1d3eb204618e9038--
More information about the Insight-users
mailing list