Re: Deconvolution software (was new posting)

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

Re: Deconvolution software (was new posting)

Karl Bellve-3
I have talked to Arthur and Karl about this.

We plan to open source our deconvolution package and make a plugin for Micro-Manager or maybe a stand alone package that could read Micro-Manager images. The algorithm was shown in our 1996 Science paper (Carrington, et. al).
 
It does use a lot of memory (> 8 x Image size) and it does take time, but we have been told by a small group of outside collaborators that it out performs every other package they have tried.

I can't promise when it will be available. I am currently optimizing a CUDA version before I start porting it to micro-manager.


On Thu, Apr 22, 2010 at 5:33 PM, Arthur D. Edelstein <[hidden email]> wrote:


Has anyone tried any ImageJ plugins with deconvolution? Or other
deconv applications? Which ones do you like?

Best regards,
Arthur



------------------------------------------------------------------------------

_______________________________________________
micro-manager-general mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/micro-manager-general
Reply | Threaded
Open this post in threaded view
|

Re: Deconvolution software (was new posting)

Truant, Ray

We currently use Autodeblur X2, which is seamless with NIS Elements and most meta formats, and we get 1024X1024 stacks to deconvolved in about 60 seconds per channel. It is extremely reliable (X2, not NIS). The 3D iterative algorithm interprets and iteratively corrects the PSF from the dataset.

 

@1024X1024, that would require 8MB of RAM, which means you would have to run a 64-bit system with Windows, or you will hit the 4GB limit with Win32(?).

 

Assuming 4GB of RAM, the maximal image size would be about 700X700, which is fine for most applications. Most people use 512X512 anyway.

 

If you would like us to try this in our imaging, we could compare with other packages.

 

Ideally, a good deconvolution module should be able to read 4D+ data sets and batch process each Z-stack sequentially along t.

 

 As for the time constraint, is the CPU or the graphics card doing the math? Commercials programs process on the GPU for speed.

 

 

Dr. Ray Truant Ph.D

Associate Professor

McMaster University

Department of Biochemistry and Biomedical Sciences

Chair, Huntington Society of Canada Scientific Advisory Board

HSC4N54, 1200 Main Street West

Hamilton, Ontario, Canada L8N3Z5

905-525-9140 ex.22450

905-522-9033 Fax

905-515-5526 blackberry

Skype ray.truant

www.RaytruantLab.ca

www.Macbiophotonics.ca

 

From: Karl Bellve [mailto:[hidden email]]
Sent: Friday, April 23, 2010 9:45 AM
To: Micro-Manager General
Subject: Re: [micro-manager-general] Deconvolution software (was new posting)

 

I have talked to Arthur and Karl about this.

We plan to open source our deconvolution package and make a plugin for Micro-Manager or maybe a stand alone package that could read Micro-Manager images. The algorithm was shown in our 1996 Science paper (Carrington, et. al).
 
It does use a lot of memory (> 8 x Image size) and it does take time, but we have been told by a small group of outside collaborators that it out performs every other package they have tried.

I can't promise when it will be available. I am currently optimizing a CUDA version before I start porting it to micro-manager.

On Thu, Apr 22, 2010 at 5:33 PM, Arthur D. Edelstein <[hidden email]> wrote:



Has anyone tried any ImageJ plugins with deconvolution? Or other
deconv applications? Which ones do you like?

Best regards,
Arthur


------------------------------------------------------------------------------

_______________________________________________
micro-manager-general mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/micro-manager-general
Reply | Threaded
Open this post in threaded view
|

Re: Deconvolution software (was new posting)

Karl Bellve-3

Hi Ray,

With our 3D deconvolution software, you need to acquire an actual PSF to get the best results Currently, our software only works with our own internal image format which is why I am porting it to Micro-manager.

Yes, it does use the GPU, via CUDA, but I still am currently optimizing it. I am still waiting on a GTX480 to arrive since I am running into GPU memory limitations with our 1GB 8800GT graphics card. But, when using small image sets, the GPU version is 2-3x faster compared to the CPU version. I am hoping to get it to 10x faster with optimizations, plus another 4-5x faster when using a faster GPU. Most of the compute time is spent processing FFTs.

Our software is 4D capable but we use a compute cluster to process the restorations.

On Fri, Apr 23, 2010 at 12:50 PM, Truant, Ray <[hidden email]> wrote:

We currently use Autodeblur X2, which is seamless with NIS Elements and most meta formats, and we get 1024X1024 stacks to deconvolved in about 60 seconds per channel. It is extremely reliable (X2, not NIS). The 3D iterative algorithm interprets and iteratively corrects the PSF from the dataset.

 

@1024X1024, that would require 8MB of RAM, which means you would have to run a 64-bit system with Windows, or you will hit the 4GB limit with Win32(?).

 

Assuming 4GB of RAM, the maximal image size would be about 700X700, which is fine for most applications. Most people use 512X512 anyway.

 

If you would like us to try this in our imaging, we could compare with other packages.

 

Ideally, a good deconvolution module should be able to read 4D+ data sets and batch process each Z-stack sequentially along t.

 

 As for the time constraint, is the CPU or the graphics card doing the math? Commercials programs process on the GPU for speed.

 

 

Dr. Ray Truant Ph.D



--
Cheers,

Karl Bellve

------------------------------------------------------------------------------

_______________________________________________
micro-manager-general mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/micro-manager-general