search

pyimreg / python-register

Linear and non-linear image registration methods, using scipy and numpy.
Other
62 stars 35 forks source link

No recovery from bad steps. #12

Closed riaanvddool closed 13 years ago

riaanvddool commented 13 years ago

The current Register.register method will never recover from a bad optimizations step.

It needs to be investigated if this is just an implementation error.

The dampening is currently changed on a bad step, but this does not currently affect the warp that is applied before retesting.

nfaggian commented 13 years ago

Did we get anywhere with this? I can spend some time on this tomorrow..

riaanvddool commented 13 years ago

I havnt spent time on it yet.

On Wed, Aug 3, 2011 at 12:28 PM, nfaggian reply@reply.github.com wrote:

Did we get anywhere with this?  I can spend some time on this tomorrow..

Reply to this email directly or view it on GitHub: https://github.com/nfaggian/python-register/issues/12#issuecomment-1716932

nfaggian commented 13 years ago

Thanks to Riaan - I have made a branch "ticket_0012" with a fix to register.py for review. Begs the question how to test this???

riaanvddool commented 13 years ago

I found testing corner cases quite helpful in this case. Ie, try to register two identical images. Also, check the values of p returned by the register function versus what the last good step was.

nfaggian commented 13 years ago

ok

nfaggian commented 13 years ago

I think we have addressed this in issue_26