Fiji question

[Robin]

Thanks for your fast response Ignacio, 

I indeed have a question about the WEKA trainer: 

I have now used the tool quite often and it performs well.

The only thing I would like to have done is to re-train the features included in a particular class on other images than the original image I trained initially on. 

It is often the case that for 10% of the images the segmentation is  not done well and I think one could easily  correct this by re-training on the image that shows the problem. 

IS there a way to do this?

Many thanks and best regards, 

Robin

 

On Jun 30, 2014, at 2:15 PM, Ignacio Arganda-Carreras [via ImageJ] <[hidden email]> wrote:

Hello Robin, 

I'm glad to hear you could solve the problem yourself. Please, let me know 
if you need any further help. 

Best, 

ignacio 


On Mon, Jun 30, 2014 at 9:43 AM, Robin <<a href="x-msg://6/user/SendEmail.jtp?type=node&amp;node=5008491&amp;i=0" target="_top" rel="nofollow" link="external">[hidden email]> wrote: 

> Hi, 
> 
> I've reduced the size of my images by factor 4 and could easily complete 
> the 
> task. 
> 
> Here is the workflow I used: 
> 
> 1. Maximum Intensity Projection (MIP) of Z-stack .stk files. 
> 2. Auto level. 
> 3. Resize MIPs. 
> 4. Train classifier and save it as .model file. 
> 5. Apply classifier; all files in complete directory. 
> 6. Save output files into new directory and do not open in FIJI. 
> 
> On my Mac Book Pro; 3Ghz intel Core i7; OSX 10.9.3; 8GB RAM, 
> the job took about 10 mins. 
> 
> R 
> 
> 
> 
> -- 
> View this message in context: 
> http://imagej.1557.x6.nabble.com/Fiji-gets-stuck-at-Trainable-Weka-Segmentation-tp5007076p5008489.html
> Sent from the ImageJ mailing list archive at Nabble.com. 
> 
> -- 
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

-- 
Ignacio Arganda-Carreras, Ph.D. 
Seung's lab, 46-5065 
Department of Brain and Cognitive Sciences 
Massachusetts Institute of Technology 
43 Vassar St. 
Cambridge, MA 02139 
USA 

Phone: [phone] 
Website: http://bioweb.cnb.csic.es/~iarganda/index_EN.html

-- 
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

---

If you reply to this email, your message will be added to the discussion below:

http://imagej.1557.x6.nabble.com/Fiji-gets-stuck-at-Trainable-Weka-Segmentation-tp5007076p5008491.html

To unsubscribe from Fiji gets stuck at Trainable Weka Segmentation, click here.
NAML

[Ignacio Arganda-Carreras]

Hello again, Robin,



> The only thing I would like to have done is to re-train the features
> included in a particular class on other images than the original image I
> trained initially on.
> It is often the case that for 10% of the images the segmentation is  not
> done well and I think one could easily  correct this by re-training on the
> image that shows the problem.
> IS there a way to do this?
>

Yes, of course. You just need to be careful with the type of classifier
you're using. Many of the Weka classifiers start from scratch if you
re-train them (including the default FastRandomForest) so they will lose
any previous information they have. In that case, what you need to do is to
click on "Save Data" after you are satisfied with your first classifier.
This will save the feature information from the samples you traced into a
file. Later, when you need to "re-train", you just click on "Load Data",
select that file, and continue with the interactive learning as before.
This way, the new trained classifier will not only take into account the
information from the new images but also from the old ones.

Does it make sense?

Let me know if you need more help!

ignacio

>
> Many thanks and best regards,
>
>
> Robin
>
>
>
> On Jun 30, 2014, at 2:15 PM, Ignacio Arganda-Carreras [via ImageJ] <
> [hidden email]> wrote:
>
> > Hello Robin,
> >
> > I'm glad to hear you could solve the problem yourself. Please, let me
> know
> > if you need any further help.
> >
> > Best,
> >
> > ignacio
> >
> >
> > On Mon, Jun 30, 2014 at 9:43 AM, Robin <[hidden email]> wrote:
> >
> > > Hi,
> > >
> > > I've reduced the size of my images by factor 4 and could easily
> complete
> > > the
> > > task.
> > >
> > > Here is the workflow I used:
> > >
> > > 1. Maximum Intensity Projection (MIP) of Z-stack .stk files.
> > > 2. Auto level.
> > > 3. Resize MIPs.
> > > 4. Train classifier and save it as .model file.
> > > 5. Apply classifier; all files in complete directory.
> > > 6. Save output files into new directory and do not open in FIJI.
> > >
> > > On my Mac Book Pro; 3Ghz intel Core i7; OSX 10.9.3; 8GB RAM,
> > > the job took about 10 mins.
> > >
> > > R
> > >
> > >
> > >
> > > --
> > > View this message in context:
> > >
> http://imagej.1557.x6.nabble.com/Fiji-gets-stuck-at-Trainable-Weka-Segmentation-tp5007076p5008489.html
> > > Sent from the ImageJ mailing list archive at Nabble.com.
> > >
> > > --
> > > ImageJ mailing list: http://imagej.nih.gov/ij/list.html
> > >
> >
> >
> >
> > --
> > Ignacio Arganda-Carreras, Ph.D.
> > Seung's lab, 46-5065
> > Department of Brain and Cognitive Sciences
> > Massachusetts Institute of Technology
> > 43 Vassar St.
> > Cambridge, MA 02139
> > USA
> >
> > Phone: [phone]
> > Website: http://bioweb.cnb.csic.es/~iarganda/index_EN.html
> >
> > --
> > ImageJ mailing list: http://imagej.nih.gov/ij/list.html
> >
> >
> > If you reply to this email, your message will be added to the discussion
> below:
> >
> http://imagej.1557.x6.nabble.com/Fiji-gets-stuck-at-Trainable-Weka-Segmentation-tp5007076p5008491.html
> > To unsubscribe from Fiji gets stuck at Trainable Weka Segmentation,
> click here.
> > NAML
>
>
>
>
>
> --
> View this message in context:
> http://imagej.1557.x6.nabble.com/Fiji-question-tp5008530.html
> Sent from the ImageJ mailing list archive at Nabble.com.
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--
Ignacio Arganda-Carreras, Ph.D.
Seung's lab, 46-5065
Department of Brain and Cognitive Sciences
Massachusetts Institute of Technology
43 Vassar St.
Cambridge, MA 02139
USA

Phone: [phone]
Website: http://bioweb.cnb.csic.es/~iarganda/index_EN.html

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html