Menu
▾
▴
rtftr-devel — Discussions for rtftr developers
You can subscribe to this list here.
| 2008 |
Jan
|
Feb
|
Mar
|
Apr
|
May
|
Jun
|
Jul
|
Aug
|
Sep
(9) |
Oct
(26) |
Nov
(14) |
Dec
(79) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2009 |
Jan
(6) |
Feb
(7) |
Mar
(55) |
Apr
(13) |
May
|
Jun
|
Jul
|
Aug
(1) |
Sep
|
Oct
|
Nov
|
Dec
|
| 2010 |
Jan
|
Feb
|
Mar
(2) |
Apr
(2) |
May
|
Jun
(1) |
Jul
|
Aug
|
Sep
|
Oct
|
Nov
|
Dec
|
| 2011 |
Jan
|
Feb
|
Mar
|
Apr
|
May
|
Jun
|
Jul
|
Aug
(2) |
Sep
|
Oct
|
Nov
|
Dec
|
| 2012 |
Jan
|
Feb
|
Mar
(7) |
Apr
(2) |
May
|
Jun
|
Jul
|
Aug
|
Sep
|
Oct
|
Nov
|
Dec
|
|
From: Abhishek D. <the...@gm...> - 2009-02-20 14:59:10
|
Hi Bibek, According to our new plan (four modules will appear as a shared library in rtftr project), we don't need a global project wide build system at present. We will need it after integrating everything using OpenMPI. SVN repo has been restructured as shown in this WIKI page (scroll to the bottom of the page) http://apps.sourceforge.net/mediawiki/rtftr/index.php?title=SF_Project_Services_Usage_Guide You can continue kdevelop based development of your module in trunk/modules/rowleynn/cpp and move the shared library to trunk/rtftruser/rtftr/lib/lib*.so (as given in task allocation document distributed in our last meeting). So for now (until OpenMPI based integration), continue your past build system in trunk/modules/rowleynn/cpp Bye On Fri, Feb 20, 2009 at 8:16 PM, Bibek Shrestha <bib...@gm...>wrote: > According to my faint memory, we had initially decided to restructure svn > so that everyone has a consistent build system. > > Has anyone done anything about it? > > -- > Bibek Shrestha > bibekshrestha at gmail dot com > > > ------------------------------------------------------------------------------ > Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, > CA > -OSBC tackles the biggest issue in open source: Open Sourcing the > Enterprise > -Strategies to boost innovation and cut costs with open source > participation > -Receive a $600 discount off the registration fee with the source code: > SFAD > http://p.sf.net/sfu/XcvMzF8H > _______________________________________________ > Rtftr-devel mailing list > Rtf...@li... > https://lists.sourceforge.net/lists/listinfo/rtftr-devel > > -- Abhishek http://adutta.np.googlepages.com |
|
From: Bibek S. <bib...@gm...> - 2009-02-20 14:31:44
|
According to my faint memory, we had initially decided to restructure svn so that everyone has a consistent build system. Has anyone done anything about it? -- Bibek Shrestha bibekshrestha at gmail dot com |
|
From: Abhishek D. <the...@gm...> - 2009-02-20 02:54:38
|
---------- Forwarded message ---------- From: SourceForge.net <no...@so...> Date: Thu, Feb 19, 2009 at 11:42 PM Subject: [ alexandria-Feature Requests-2121794 ] Hosted App for Trac To: no...@so... Feature Requests item #2121794, was opened at 2008-09-21 10:48 Message generated for change (Comment added) made by moorman You can respond by visiting: https://sourceforge.net/tracker/?func=detail&atid=350001&aid=2121794&group_id=1 Please note that this message will contain a full copy of the comment thread, including the initial issue submission, for this request, not just the latest update. Category: None Group: None >Status: Closed Priority: 5 Private: No Submitted By: Abhishek Dutta (thelinuxmaniac) >Assigned to: Service Operations Group (sog) Summary: Hosted App for Trac Initial Comment: Please add Trac (http://trac.edgewall.org/) as a new hosted app. Trac is a web-based software project management and bug/issue tracking system. It provides an interface to Subversion and an integrated wiki. ---------------------------------------------------------------------- >Comment By: Jacob Moorman (moorman) Date: 2009-02-19 12:57 Message: Hello, Trac is now part of our Hosted Apps offering. For more details, see: http://apps.sourceforge.net/trac/sitedocs/wiki/Hosted%20Apps Thank you, Jacob Moorman Director of Operations, SourceForge.net ---------------------------------------------------------------------- Comment By: Francesco Montorsi (frm) Date: 2008-12-12 07:40 Message: I support this request! Trac is really nice for project management and having it in SourceForge as stock app would simply be great... ---------------------------------------------------------------------- You can respond by visiting: https://sourceforge.net/tracker/?func=detail&atid=350001&aid=2121794&group_id=1 -- Abhishek http://adutta.np.googlepages.com |
|
From: Abhishek D. <the...@gm...> - 2009-01-10 15:33:04
|
Hi Anjan, The results look good. I think you need not bother about training now. You can perform training during the time when we restart our work after exams (end of Feb. 2009). By that time, the loadshedding will probably decrease. We can also share batteries if required. My home has also started seeing normal loadshedding schedules now (I think that the vip house has been equipped with generators). So I cannot help. Meanwhile, you can experiment on the technique of using both cores for processing. If you have any other plans that will help, we can go for that. bye On Sat, Jan 10, 2009 at 6:51 PM, Anjan <roc...@gm...> wrote: > Hi all, > The adaboost algorithm is improving :) . The image attached is the > detection after training upto 3 stages. The false detections are reduced > when the stages are increased. > > Result: > Stages: 3 > Test Image size: 316x102 > Detection Time: 0.36087 secs (hopefully ok for real-time) > > > Now, i'm sure that the adaboost algorithm training in large scale is ready, > after which the detection rate and false detection will be as expected. But > the problem is with the training, which takes a lot of time (at least a > week). The loadshedding is such a pain in the ass. What do you guys suggest > for that. I'll be looking for how the multi-processing can be done to speed > up the training. I have found that some part of my code can be performed by > multi-processing, but i've not tried it yet. In Open MP we can have multiple > thread, but in my test i found that all those threads run in the same core. > Further research is needed. > What's your view on where can we start the training. > > -- > Anjan Nepal > > GPO Box 11458 > Registered Linux User #395963 > > > > ------------------------------------------------------------------------------ > Check out the new SourceForge.net Marketplace. > It is the best place to buy or sell services for > just about anything Open Source. > http://p.sf.net/sfu/Xq1LFB > _______________________________________________ > Rtftr-devel mailing list > Rtf...@li... > https://lists.sourceforge.net/lists/listinfo/rtftr-devel > > -- Abhishek http://adutta.np.googlepages.com |
|
From: Abhishek D. <the...@gm...> - 2009-01-06 03:49:44
|
Hi all, The final integration plan has been updated in wiki. Check this wiki page for details http://apps.sourceforge.net/mediawiki/rtftr/index.php?title=Final_Integration_Plan bye -- Abhishek http://adutta.np.googlepages.com |
|
From: Abhishek D. <the...@gm...> - 2009-01-05 19:29:59
|
Hi all, The svn repository structure has changed for Final Integration. Check this image for details http://apps.sourceforge.net/mediawiki/rtftr/index.php?title=Image:Rtftr_svn_repo_desc.png bye -- Abhishek http://adutta.np.googlepages.com |
|
From: Abhishek D. <the...@gm...> - 2009-01-02 14:42:01
|
Hi all, The "sendfile" system call in Linux kernel 2.4X seems to be a great method for us to send files (configuration file, images, etc) to different nodes. We can use the unix sockets for sending commands. See this page http://tldp.org/LDP/LGNET/issue91/tranter.html bye -- Abhishek http://adutta.np.googlepages.com |
|
From: Abhishek D. <the...@gm...> - 2009-01-01 13:44:35
|
Hi all, we have a meeting scheduled for tomorrow Time: 02:00PM Date: Jan. 02, 2009 Venue: Lakesh's House bye -- Abhishek http://adutta.np.googlepages.com |
|
From: Abhishek D. <the...@gm...> - 2008-12-30 19:13:26
|
Hi all, I have created a wiki page that contains all the necessary plots that describe the performance of kepenekci C++ implementation. http://apps.sourceforge.net/mediawiki/rtftr/index.php?title=Kepenekci_Performance The results are not amazing but gives some sort of relief. These performance data reflect the status when no optimization has been done. We can certainly achieve better results with some optimizations. bye -- Abhishek http://adutta.np.googlepages.com |
|
From: Abhishek D. <the...@gm...> - 2008-12-28 16:44:12
|
Hi all, check this wiki page that describes the Final Integration phase plan https://apps.sourceforge.net/mediawiki/rtftr/index.php?title=Final_Integration Everybody create a detailed specification for the task assigned to you so that everybody knows what to do when integration starts on Jan 3, 2009. Edit according to the instructions on that wiki page. Every member must predefine their requirements for other modules (assigned to other member) in their respective pages. This will make life easier during final integration. also check the Plan and Progress page at https://apps.sourceforge.net/mediawiki/rtftr/index.php?title=Final_Integration and update the details related to your responsibility. Jan 03, 2009 is Sat. and we are starting final integration from that day. We will plan for movie on next saturday (or some other day ) as we cannot waste any time from only 7 days allocated for final integration. bye -- Abhishek http://adutta.np.googlepages.com |
|
From: Abhishek D. <the...@gm...> - 2008-12-27 14:42:42
|
Hi all, You can find a very short HOWTO on using gnuplot for generating nice looking plots at http://apps.sourceforge.net/mediawiki/rtftr/index.php?title=Generating_plots At present we are generating lots of performance related data. So this HOWTO would help in generating simple plots. The R Project for Statistical Computing <http://www.r-project.org/> seems a better choice considering long term requirements. However for now we can use gnuplot. bye -- Abhishek http://adutta.np.googlepages.com |
|
From: Bibek S. <bib...@gm...> - 2008-12-27 05:32:57
|
looking into it finally got to understand how the table in the matlab's adaboost works I'll look into the XML io and reply On Friday 26 December 2008 12:37:08 Anjan wrote: > I think this is a better option. > http://note.sonots.com/OpenCV/XML.html > see example 2 > it is what opencv uses too. > > On Fri, Dec 26, 2008 at 12:30 PM, Anjan <roc...@gm...> wrote: > > Hi Bibek, > > the coding of adaboost is complete, and i think it's selecting features > > wisely. Now we should move on to creating the cascade. We need to discuss > > on how we save the overall cascade in the disk. > > Here is what i have in mind. > > > > * in one file, save the number of stages in the cascade. ( eg. > > stagecount.xml) > > * in other file, the threshold of each stage will be saved ( eg. cvMat* > > stagethreshold = cvCreateMat( no. of stages, 1, CV_32FC1) > > * in other files, the classifiers of one stage will be saved (eg. > > stage_1.xml stage2.xml etc) > > > > So, during detection this is what we will do > > > > * first load stagecount.xml > > * read threshold file > > * create array of the matrices of size no. of stages and load all the > > stages in this array. > > > > I think for continuing the cascade generation also, this will be ok. > > We'll include a flag (continue = 0/1) to cascadegeneration() function and > > if continue=1, we continue the cascade generation process by loading the > > stagecount.xml, thresholds, stages and then creating new nonfaces and > > continuing. > > > > Reply soon. > > -- > > Anjan Nepal > > > > GPO Box 11458 > > Registered Linux User #395963 -- Bibek Shrestha bib...@gm... 9841 825 925 |
|
From: Anjan <roc...@gm...> - 2008-12-26 18:51:36
|
I think this is what we will be doing for saving the cascade. http://apps.sourceforge.net/mediawiki/rtftr/index.php?title=Saving_of_Cascade_in_OpenCV On Fri, Dec 26, 2008 at 12:37 PM, Anjan <roc...@gm...> wrote: > I think this is a better option. > http://note.sonots.com/OpenCV/XML.html > see example 2 > it is what opencv uses too. > > > On Fri, Dec 26, 2008 at 12:30 PM, Anjan <roc...@gm...> wrote: > >> Hi Bibek, >> the coding of adaboost is complete, and i think it's selecting features >> wisely. Now we should move on to creating the cascade. We need to discuss on >> how we save the overall cascade in the disk. >> Here is what i have in mind. >> >> * in one file, save the number of stages in the cascade. ( eg. >> stagecount.xml) >> * in other file, the threshold of each stage will be saved ( eg. cvMat* >> stagethreshold = cvCreateMat( no. of stages, 1, CV_32FC1) >> * in other files, the classifiers of one stage will be saved (eg. >> stage_1.xml stage2.xml etc) >> >> So, during detection this is what we will do >> >> * first load stagecount.xml >> * read threshold file >> * create array of the matrices of size no. of stages and load all the >> stages in this array. >> >> I think for continuing the cascade generation also, this will be ok. We'll >> include a flag (continue = 0/1) to cascadegeneration() function and if >> continue=1, we continue the cascade generation process by loading the >> stagecount.xml, thresholds, stages and then creating new nonfaces and >> continuing. >> >> Reply soon. >> -- >> Anjan Nepal >> >> GPO Box 11458 >> Registered Linux User #395963 >> >> > > > -- > Anjan Nepal > > GPO Box 11458 > Registered Linux User #395963 > > -- Anjan Nepal GPO Box 11458 Registered Linux User #395963 |
|
From: Abhishek D. <the...@gm...> - 2008-12-26 18:23:57
|
Oh man! You made the 100th svn commit. I was willing to make that milestone commit :) anyway congrats for 100th svn commit to svn team :) bye On Sat, Dec 27, 2008 at 12:00 AM, Anjan <roc...@gm...> wrote: > * Partial cascade generation coded > * Cascade will be saved using CvFileStorage > * The format in which cascade will be saved will be added in wiki > * Detection code from a cascade provided (todo). Cascade generation > code dependent on this. > > -- > Anjan Nepal > > GPO Box 11458 > Registered Linux User #395963 > > > ------------------------------------------------------------------------------ > _______________________________________________ > Rtftr-devel mailing list > Rtf...@li... > https://lists.sourceforge.net/lists/listinfo/rtftr-devel > -- Abhishek http://adutta.np.googlepages.com |
|
From: Anjan <roc...@gm...> - 2008-12-26 18:15:40
|
* Partial cascade generation coded * Cascade will be saved using CvFileStorage * The format in which cascade will be saved will be added in wiki * Detection code from a cascade provided (todo). Cascade generation code dependent on this. -- Anjan Nepal GPO Box 11458 Registered Linux User #395963 |
|
From: Anjan <roc...@gm...> - 2008-12-26 06:52:12
|
I think this is a better option. http://note.sonots.com/OpenCV/XML.html see example 2 it is what opencv uses too. On Fri, Dec 26, 2008 at 12:30 PM, Anjan <roc...@gm...> wrote: > Hi Bibek, > the coding of adaboost is complete, and i think it's selecting features > wisely. Now we should move on to creating the cascade. We need to discuss on > how we save the overall cascade in the disk. > Here is what i have in mind. > > * in one file, save the number of stages in the cascade. ( eg. > stagecount.xml) > * in other file, the threshold of each stage will be saved ( eg. cvMat* > stagethreshold = cvCreateMat( no. of stages, 1, CV_32FC1) > * in other files, the classifiers of one stage will be saved (eg. > stage_1.xml stage2.xml etc) > > So, during detection this is what we will do > > * first load stagecount.xml > * read threshold file > * create array of the matrices of size no. of stages and load all the > stages in this array. > > I think for continuing the cascade generation also, this will be ok. We'll > include a flag (continue = 0/1) to cascadegeneration() function and if > continue=1, we continue the cascade generation process by loading the > stagecount.xml, thresholds, stages and then creating new nonfaces and > continuing. > > Reply soon. > -- > Anjan Nepal > > GPO Box 11458 > Registered Linux User #395963 > > -- Anjan Nepal GPO Box 11458 Registered Linux User #395963 |
|
From: Anjan <roc...@gm...> - 2008-12-26 06:45:10
|
Hi Bibek, the coding of adaboost is complete, and i think it's selecting features wisely. Now we should move on to creating the cascade. We need to discuss on how we save the overall cascade in the disk. Here is what i have in mind. * in one file, save the number of stages in the cascade. ( eg. stagecount.xml) * in other file, the threshold of each stage will be saved ( eg. cvMat* stagethreshold = cvCreateMat( no. of stages, 1, CV_32FC1) * in other files, the classifiers of one stage will be saved (eg. stage_1.xml stage2.xml etc) So, during detection this is what we will do * first load stagecount.xml * read threshold file * create array of the matrices of size no. of stages and load all the stages in this array. I think for continuing the cascade generation also, this will be ok. We'll include a flag (continue = 0/1) to cascadegeneration() function and if continue=1, we continue the cascade generation process by loading the stagecount.xml, thresholds, stages and then creating new nonfaces and continuing. Reply soon. -- Anjan Nepal GPO Box 11458 Registered Linux User #395963 |
|
From: Abhishek D. <the...@gm...> - 2008-12-25 07:35:48
|
Hi all, Please enable $Id$ tag support in svn so that it is easy to see who last changed the contents of a file. This will be very useful when we move on to integration of all module. A short HOWTO on enabling $Id$ tag is given at http://apps.sourceforge.net/mediawiki/rtftr/index.php?title=SF_Project_Services_Usage_Guide Reminder: We have a meeting scheduled for Dec. 28, 2008 (Sunday). The venue will be finalized on Dec. 27, 2008 (Saturday) bye -- Abhishek http://adutta.np.googlepages.com |
|
From: Abhishek D. <the...@gm...> - 2008-12-24 05:40:46
|
Hi all, I mistakenly wrote kepenekci_part1: * kepenekci_part1 tasks completed * @todo integration with kepenekci_part2 will not proceed However, I meant "@todo integration with kepenekci_part2 will now proceed" bye -- Abhishek http://adutta.np.googlepages.com |
|
From: lakesh <lak...@gm...> - 2008-12-22 15:58:15
|
Hi abhishek, I was trying to find the high energy points from the Gabor filtered images. But now again I have some doubts. As we discussed earlier we move a window of size WxW through the Gabor filtered outputs. For each point x0,y0 the center of the window W0 we try to verify the condition as given in the paper. But now I think even if for 1 Gabor filter the condition is satisfied then it is regarded as feature. I could hardly match the conditions for 3 Gabor filter coefficients. So there is no way of matching the condition for all the 40 filters. However this time the features are much more localized at the center instead of the edges near the eyes,mouth and nose. I get lots of features if I pass the points as feature for just one match. Again as before if I pass the points as feature for two matches, the results are better with less number of features. I will do some more experiments and let you know the results. -- Have a nice day |
|
From: Anjan <roc...@gm...> - 2008-12-22 15:31:44
|
forgot to add...the problem with this i see is that each time for updating the matrix in the sequence, it has to be popped and again pushed back. On Mon, Dec 22, 2008 at 9:14 PM, Anjan <roc...@gm...> wrote: > Hi, > Abhishek, your idea for the sorting in matrix was good, but it didn't > exactly do what i wanted. i.e only that column was sorted and not the > corresponding records. > Bibek, do you think we can use CvSeq of OpenCV for creating the table that > we need in adaboost. See the sample code in the attachment. I think it is > ok. > > -- > Anjan Nepal > > GPO Box 11458 > Registered Linux User #395963 > > -- Anjan Nepal GPO Box 11458 Registered Linux User #395963 |
|
From: Anjan <roc...@gm...> - 2008-12-22 15:29:38
|
Hi, Abhishek, your idea for the sorting in matrix was good, but it didn't exactly do what i wanted. i.e only that column was sorted and not the corresponding records. Bibek, do you think we can use CvSeq of OpenCV for creating the table that we need in adaboost. See the sample code in the attachment. I think it is ok. -- Anjan Nepal GPO Box 11458 Registered Linux User #395963 |
|
From: Abhishek D. <the...@gm...> - 2008-12-21 01:51:31
|
Hi Anjan,
Vigra has option to get different types of views of a matrix (column view,
row view, etc). After obtaining a particular view, you can use the sort
method. I don't have the details.
The best option I see is that you store your matrix in a vector< vector<int>
> STL container and use the STL sort algorithm to sort each column.
eg:
// NOTE: you have to create 6 rows of 15000 length records as you want to
sort each column
vector < vector <int> > mydata(6, vector<int>(15000) )
vector < vector <int> >::iterator v_itr;
for( v_itr = mydata.begin(); v_itr != mydata.end(); v_itr++)
{
sort( *v_itr.begin(), *v_itr.end() );
}
// now everything in mydata is sorted column wise
hope this helps
bye
On Sat, Dec 20, 2008 at 9:59 PM, Anjan <roc...@gm...> wrote:
> Hi,
> In adaboost implementation in C++, i need to process a very large matrix,
> around size of 15000x6. I think OpenCV's cvMat can be used to store it. But,
> I need to sort the matrix according to one of the columns. Do you guys have
> any idea if opencv, vigra, octave etc has this function? I'm trying to avoid
> doing this myself.
> On Sat, Dec 20, 2008 at 9:29 PM, lakesh <lak...@gm...> wrote:
>
>> Hi Abhishek,
>> I think you deleted the matlab and kepenekci_part2 directories during your
>> previous commits. I am adding them again ok.
>>
>>
>> On Sat, Dec 20, 2008 at 9:19 PM, lakesh <lak...@gm...>wrote:
>>
>>> Also about sigma most of the papers they have taken 2pi like us. But this
>>> is totally arbitrary as said by one of the people in the groups
>>>
>>>
>>> On Sat, Dec 20, 2008 at 9:15 PM, lakesh <lak...@gm...>wrote:
>>>
>>>> Hi Abhishek,
>>>>
>>>> I was tuning the parameters of the Gabor filters and I think I got some
>>>> improvements. I changed the window size and the threshold and some other
>>>> parameters. There is slight improvement; however we can tune the parameters
>>>> to any thing we want that is we can overfit it to a specific image. Also
>>>> other thing is that we don't need to get the high energy points to the
>>>> specific locations like eyes,nose mouth etc. I got some papers where they
>>>> have used adaboost for feature selection and the features are located at
>>>> random locations like us. Also in most of the papers they have taken
>>>> frequency as sqrt(2). The ebgm paper has also taken the same frequency. Also
>>>> I think while comparing the responses of the images to get the feature point
>>>> we should use directly the magnitude. I mean when we shift through the image
>>>> using the window of size WxW to compare the results, we should use the
>>>> magnitude of the response for the comparison and not directly the complex
>>>> numbers. I tried to compare the complex number and it considers only the
>>>> real part while doing the comparison. So I think it is not the correct way
>>>> to compare. Though in Kepenekci it seems like they have directly compared
>>>> the complex numbers, I think we should use the magnitude values directly.
>>>>
>>>> Lets sit and discuss the parameter selection when you finish the C++
>>>> implementation.
>>>> Currently I am laying my hands with everything I can find and checking
>>>> the results.
>>>>
>>>>
>>>>
>>>> On Sat, Dec 20, 2008 at 8:46 PM, Abhishek Dutta <
>>>> the...@gm...> wrote:
>>>>
>>>>> Hi lakesh,
>>>>> I decided to use "vigra<http://kogs-www.informatik.uni-hamburg.de/%7Ekoethe/vigra/>"
>>>>> library instead of GSL. GSL was not sufficient and did not fulfill all our
>>>>> requirements.
>>>>>
>>>>> "vigra" allows to generate gabor filter family by just supplying the
>>>>> no. of orientations, scale and center frequency. According to the "vigra"
>>>>> documentation, center frequency should not exceed 0.375. However, I was
>>>>> using sqrt(2) = 1.414 as frequency in matlab implementation. Moreover, it
>>>>> selects the best gaussian function (the sigma that we needed to supply)
>>>>> based on a mathematical formula (that I didn't understand). The output (at
>>>>> image size 128x128) looks good. But this does not assure better final
>>>>> results. I will update you with final results after I complete the code to
>>>>> compute feature vectors.
>>>>>
>>>>> bye
>>>>>
>>>>> --
>>>>> Abhishek
>>>>> http://adutta.np.googlepages.com
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Have a nice day
>>>>
>>>
>>>
>>>
>>> --
>>> Have a nice day
>>>
>>
>>
>>
>> --
>> Have a nice day
>>
>>
>> ------------------------------------------------------------------------------
>>
>> _______________________________________________
>> Rtftr-devel mailing list
>> Rtf...@li...
>> https://lists.sourceforge.net/lists/listinfo/rtftr-devel
>>
>>
>
>
> --
> Anjan Nepal
>
> GPO Box 11458
> Registered Linux User #395963
>
>
--
Abhishek
http://adutta.np.googlepages.com
|
|
From: Anjan <roc...@gm...> - 2008-12-20 16:14:10
|
Hi, In adaboost implementation in C++, i need to process a very large matrix, around size of 15000x6. I think OpenCV's cvMat can be used to store it. But, I need to sort the matrix according to one of the columns. Do you guys have any idea if opencv, vigra, octave etc has this function? I'm trying to avoid doing this myself. On Sat, Dec 20, 2008 at 9:29 PM, lakesh <lak...@gm...> wrote: > Hi Abhishek, > I think you deleted the matlab and kepenekci_part2 directories during your > previous commits. I am adding them again ok. > > > On Sat, Dec 20, 2008 at 9:19 PM, lakesh <lak...@gm...> wrote: > >> Also about sigma most of the papers they have taken 2pi like us. But this >> is totally arbitrary as said by one of the people in the groups >> >> >> On Sat, Dec 20, 2008 at 9:15 PM, lakesh <lak...@gm...>wrote: >> >>> Hi Abhishek, >>> >>> I was tuning the parameters of the Gabor filters and I think I got some >>> improvements. I changed the window size and the threshold and some other >>> parameters. There is slight improvement; however we can tune the parameters >>> to any thing we want that is we can overfit it to a specific image. Also >>> other thing is that we don't need to get the high energy points to the >>> specific locations like eyes,nose mouth etc. I got some papers where they >>> have used adaboost for feature selection and the features are located at >>> random locations like us. Also in most of the papers they have taken >>> frequency as sqrt(2). The ebgm paper has also taken the same frequency. Also >>> I think while comparing the responses of the images to get the feature point >>> we should use directly the magnitude. I mean when we shift through the image >>> using the window of size WxW to compare the results, we should use the >>> magnitude of the response for the comparison and not directly the complex >>> numbers. I tried to compare the complex number and it considers only the >>> real part while doing the comparison. So I think it is not the correct way >>> to compare. Though in Kepenekci it seems like they have directly compared >>> the complex numbers, I think we should use the magnitude values directly. >>> >>> Lets sit and discuss the parameter selection when you finish the C++ >>> implementation. >>> Currently I am laying my hands with everything I can find and checking >>> the results. >>> >>> >>> >>> On Sat, Dec 20, 2008 at 8:46 PM, Abhishek Dutta < >>> the...@gm...> wrote: >>> >>>> Hi lakesh, >>>> I decided to use "vigra<http://kogs-www.informatik.uni-hamburg.de/%7Ekoethe/vigra/>" >>>> library instead of GSL. GSL was not sufficient and did not fulfill all our >>>> requirements. >>>> >>>> "vigra" allows to generate gabor filter family by just supplying the no. >>>> of orientations, scale and center frequency. According to the "vigra" >>>> documentation, center frequency should not exceed 0.375. However, I was >>>> using sqrt(2) = 1.414 as frequency in matlab implementation. Moreover, it >>>> selects the best gaussian function (the sigma that we needed to supply) >>>> based on a mathematical formula (that I didn't understand). The output (at >>>> image size 128x128) looks good. But this does not assure better final >>>> results. I will update you with final results after I complete the code to >>>> compute feature vectors. >>>> >>>> bye >>>> >>>> -- >>>> Abhishek >>>> http://adutta.np.googlepages.com >>>> >>> >>> >>> >>> -- >>> Have a nice day >>> >> >> >> >> -- >> Have a nice day >> > > > > -- > Have a nice day > > > ------------------------------------------------------------------------------ > > _______________________________________________ > Rtftr-devel mailing list > Rtf...@li... > https://lists.sourceforge.net/lists/listinfo/rtftr-devel > > -- Anjan Nepal GPO Box 11458 Registered Linux User #395963 |
|
From: lakesh <lak...@gm...> - 2008-12-20 15:44:50
|
Hi Abhishek, I think you deleted the matlab and kepenekci_part2 directories during your previous commits. I am adding them again ok. On Sat, Dec 20, 2008 at 9:19 PM, lakesh <lak...@gm...> wrote: > Also about sigma most of the papers they have taken 2pi like us. But this > is totally arbitrary as said by one of the people in the groups > > > On Sat, Dec 20, 2008 at 9:15 PM, lakesh <lak...@gm...> wrote: > >> Hi Abhishek, >> >> I was tuning the parameters of the Gabor filters and I think I got some >> improvements. I changed the window size and the threshold and some other >> parameters. There is slight improvement; however we can tune the parameters >> to any thing we want that is we can overfit it to a specific image. Also >> other thing is that we don't need to get the high energy points to the >> specific locations like eyes,nose mouth etc. I got some papers where they >> have used adaboost for feature selection and the features are located at >> random locations like us. Also in most of the papers they have taken >> frequency as sqrt(2). The ebgm paper has also taken the same frequency. Also >> I think while comparing the responses of the images to get the feature point >> we should use directly the magnitude. I mean when we shift through the image >> using the window of size WxW to compare the results, we should use the >> magnitude of the response for the comparison and not directly the complex >> numbers. I tried to compare the complex number and it considers only the >> real part while doing the comparison. So I think it is not the correct way >> to compare. Though in Kepenekci it seems like they have directly compared >> the complex numbers, I think we should use the magnitude values directly. >> >> Lets sit and discuss the parameter selection when you finish the C++ >> implementation. >> Currently I am laying my hands with everything I can find and checking >> the results. >> >> >> >> On Sat, Dec 20, 2008 at 8:46 PM, Abhishek Dutta <the...@gm... >> > wrote: >> >>> Hi lakesh, >>> I decided to use "vigra<http://kogs-www.informatik.uni-hamburg.de/%7Ekoethe/vigra/>" >>> library instead of GSL. GSL was not sufficient and did not fulfill all our >>> requirements. >>> >>> "vigra" allows to generate gabor filter family by just supplying the no. >>> of orientations, scale and center frequency. According to the "vigra" >>> documentation, center frequency should not exceed 0.375. However, I was >>> using sqrt(2) = 1.414 as frequency in matlab implementation. Moreover, it >>> selects the best gaussian function (the sigma that we needed to supply) >>> based on a mathematical formula (that I didn't understand). The output (at >>> image size 128x128) looks good. But this does not assure better final >>> results. I will update you with final results after I complete the code to >>> compute feature vectors. >>> >>> bye >>> >>> -- >>> Abhishek >>> http://adutta.np.googlepages.com >>> >> >> >> >> -- >> Have a nice day >> > > > > -- > Have a nice day > -- Have a nice day |
23 messages has been excluded from this view by a project administrator.
