libface-general Mailing List for Face Recognition Library (Page 2)
Status: Beta
Brought to you by:
leshiy_uk
You can subscribe to this list here.
2010 |
Jan
|
Feb
|
Mar
(17) |
Apr
(1) |
May
(2) |
Jun
(2) |
Jul
|
Aug
|
Sep
|
Oct
(3) |
Nov
|
Dec
|
---|---|---|---|---|---|---|---|---|---|---|---|---|
2011 |
Jan
|
Feb
|
Mar
|
Apr
|
May
|
Jun
(1) |
Jul
(1) |
Aug
(4) |
Sep
(1) |
Oct
(1) |
Nov
|
Dec
|
2014 |
Jan
|
Feb
|
Mar
|
Apr
(1) |
May
|
Jun
|
Jul
|
Aug
|
Sep
|
Oct
|
Nov
|
Dec
|
From: Aditya B. <adi...@gm...> - 2010-03-18 17:00:26
|
Hi. I'll be committing the code that implements a set of "weighted cascades" soon. Now since face detection is a slow operation, I think that we should use threads to speed up the processing, by running each cascade in a different thread. Does standard C++ have threading? Does pthreads work on windows? Regards, Aditya |
From: Alex J. <ale...@gm...> - 2010-03-18 10:07:56
|
Hi everyone, I just fixed the installation to be into lib64 directory if you have one :) Hopefully this should make things a bit more intuitive. When building probably should specify -DCMAKE_INSTALL_PREFIX=/opt or /opt/local. This will keep all the additional installed stuff separate from the rest of the libraries. This also makes it easier to manage. If you having problems with linker not finding libraries or includes, have a look at the /etc/ld.so.conf to include your install directory and LD_LIBRARY_PATH should also have location of libraries. I don't particularly like idea of having environment variables, but thats the way linker works :( @Aditya I saw you committed a file in examples. I'll have a look at it as soon as I get some free time and give some feedback if needed. I am little bit swamped with work at the moment. Best Regards On 18 Mar 2010, at 06:19, Aditya Bhatt wrote: > Hi, > > @Alex, I don't know about your's, but in my 64-bit archlinux system, all libs go into lib, not lib64, even though lib64 exists. > > Aditya > ------------------------------------------------------------------------------ > Download Intel® Parallel Studio Eval > Try the new software tools for yourself. Speed compiling, find bugs > proactively, and fine-tune applications for parallel performance. > See why Intel Parallel Studio got high marks during beta. > http://p.sf.net/sfu/intel-sw-dev_______________________________________________ > Libface-general mailing list > Lib...@li... > https://lists.sourceforge.net/lists/listinfo/libface-general If we knew what we were doing, it wouldn't be called research, would it? -- Albert Einstein |
From: Aditya B. <adi...@gm...> - 2010-03-18 06:19:41
|
Hi, @Alex, I don't know about your's, but in my 64-bit archlinux system, all libs go into lib, not lib64, even though lib64 exists. Aditya |
From: Alex J. <ale...@gm...> - 2010-03-16 08:50:58
|
Hi, Well, the issue of use of public database for testing. I think we will use a public database at some point for validation. But I want to delay this for as long as possible and I would rather you guys use photos of your own. Use a typical photo from your own photo libraries to work with. The reason for doing it this way, is because people who created the face detection and recognition algorithms would also probably used the same databases for validation of their methods and probably optimised the algorithm for these databases. This means that using it again to test would just skew the results, because we are testing something that was optimised to work on in the first place. Where as using own photos means we are using what are real test cases, as opposed to artificially created in the database. This also gives much greater variety to work with and so on. So for now pick a few pictures of a typical photo resolution, perhaps 7M pixels, with variety of sizes and orientations of faces and work with that. Alex On 15 Mar 2010, at 19:41, Aditya Bhatt wrote: > Hi. > > @Alex : I've committed a class libface::Haarcascades. This is intended to make it easier to deal with multiple cascades and weights for each. It still needs a few more methods to make it more usable, which I'll commit soon enough. > > Also, what face database do you use ? - so that I can use the same and we can have a standard set of test cases. > > Aditya > ------------------------------------------------------------------------------ > Download Intel® Parallel Studio Eval > Try the new software tools for yourself. Speed compiling, find bugs > proactively, and fine-tune applications for parallel performance. > See why Intel Parallel Studio got high marks during beta. > http://p.sf.net/sfu/intel-sw-dev_______________________________________________ > Libface-general mailing list > Lib...@li... > https://lists.sourceforge.net/lists/listinfo/libface-general |
From: Aditya B. <adi...@gm...> - 2010-03-15 19:41:47
|
Hi. @Alex : I've committed a class libface::Haarcascades. This is intended to make it easier to deal with multiple cascades and weights for each. It still needs a few more methods to make it more usable, which I'll commit soon enough. Also, what face database do you use ? - so that I can use the same and we can have a standard set of test cases. Aditya |
From: Alex J. <ale...@gm...> - 2010-03-11 22:44:32
|
Hi all, @Aditya Indeed peopetag was our initial thought, but like you said it was in a little bit of a sorry state, that it didn't compile for me so I left it to be for the time being. But like you said that is something I want to use because then we can fairly easily convert it into Nepomuk module etc. That is something that would be useful for KDE people and I think is also a good idea . If you want to carry on with this and concentrate on salvaging as much of that peopletag as possible and making it work that would be great. I just added you as developer for the project so feel free to rummage around. Sourceforge also has the Mercurial source control, but I am wary about using 2 different subversions and keep the states consistent between them. So when you get it working just send me a tar bal or something and I will add it to the version control. Don't worry about time for now we are in rush at the moment, so just work on it as you become free. Alex |
From: Aditya B. <adi...@gm...> - 2010-03-11 20:25:40
|
Hi, Should we really make an interface from scratch? There's the nepomuk-peopletag thingy that you probably know about ( from the long digikam bugzilla thread ). Apart from the main things that it does, tt also allows you to drag and move the tag square around the picture. There are some compile errors, but those are related only to the nepomuk part, and the code with errors seems to be a deprecated nepomuk component. So I grabbed some of his code and extracted the relevant (non-nepomuk) parts into another project. The code should work OK when finished, but a main.cpp is yet to be added due to lack of time on my side. Feel free to finish it, very little is to be added to get it up and running. I've pushed it here : http://bitbucket.org/aditya_bhatt/tagthing/ I feel that we should use this, because it is sort of "nepomuk-pluggable". So once we get it working OK, nepomuk integration should be possible. Oh, and I use bitbucket (with mercurial) because git and svn are blocked in my college campus. Is there an option to use hg in sourceforge? That would be great and convenient for me. Else I'll have to talk with my college's sysadmin to unblock the ports. One more thing: Clicking on "Send a request to join this project" in libface's page sends me to a page which says "contact the following project admins to join" followed by your and the previous owner's names. It isn't entirely clear how I should join, so if you can add me, my username is adityabhatt. Regards, Aditya |
From: Alex J. <ale...@gm...> - 2010-03-11 10:16:10
|
Hi everyone, Now that there seem to be some interest in the project I think it will be worth coordinating our efforts together. First of all let me start by welcoming everyone to the project. The reason why I got involved in this is because I got fed up with so many commercially available products that allow you to have face detection and recognition and where as similar things in open community are few and far apart. Also given that I have access to the current research, this is a very good place for trying the theory on real world examples. So this is the ethos of the project, now on to the project itself. Here is the structure of the library. The whole library will be accessed through the Libface class (LibFace.h). The end users of the library will have no knowledge and or access directly to any of the OpenCV structures and classes. The reason for doing this is because they only have to include this library and compile their projects with only one flag and so it makes things a lot easier from their point of view. As you know there are 2 aspects of the library: face detection and recognition. I went with something I knew would work so that the library can be created and used, then further development can be done to improve the backend of it, and this should be almost invisible to the end users. For this reason if you writing a face detection class it needs to inherit from LibFaceDetectCore class(LibfaceCore.h) and if you writing a face recognition it needs to inherit from LibFaceRecognitionCore class (LibfaceCore.h) The 2 classes only have virtual methods. This allows for the uniform access to the backend independent of the implementation. So what I want before we get on with implementing is to finish off Eigenfaces and face detection based on Haar features. Last night I got the face detection to work but a lot of false negatives (Type II errors). I think it is important to get the face detection to work properly first, even if it means false negatives are not zero, but fairly low. Because without having coherent set of faces to work with there is no meaningful way of testing the face recognition part. This involves checking against different types of cascades that OpenCV 2.0 comes with and seeing which ones are better. Maybe even creating a "vote" system that allows for things with highest number of votes be passed as face. Then we can get on with recognition. Also I have been working on a Qt interface for this. Something very simple that allows rendering images and superposing rectangles on top for faces. I got QGraphicalScene to work, it is still a bit messy. But this is something I want to explore further. So will need someone to help out with that. Lastly, if you indeed interested in working on this library further, and I think there is a lot of future for it then log in to Sourceforge and request to be added to the project and I shall add you promptly to the set of developers so you can commit code and changes. :) Sorry about long first email, just a lot had to be said :) Alex |
From: Alex J. <ale...@gm...> - 2010-03-11 09:57:03
|
Hi everyone, Now that there seem to be some interest in the project I think it will be worth coordinating our efforts together. First of all let me start by welcoming everyone to the project. The reason why I got involved in this is because I got fed up with so many commercially available products that allow you to have face detection and recognition and where as similar things in open community are few and far apart. Also given that I have access to the current research, this is a very good place for trying the theory on real world examples. So this is the ethos of the project, now on to the project itself. Here is the structure of the library. The whole library will be accessed through the Libface class (LibFace.h). The end users of the library will have no knowledge and or access directly to any of the OpenCV structures and classes. The reason for doing this is because they only have to include this library and compile their projects with only one flag and so it makes things a lot easier from their point of view. As you know there are 2 aspects of the library: face detection and recognition. I went with something I knew would work so that the library can be created and used, then further development can be done to improve the backend of it, and this should be almost invisible to the end users. For this reason if you writing a face detection class it needs to inherit from LibFaceDetectCore class(LibfaceCore.h) and if you writing a face recognition it needs to inherit from LibFaceRecognitionCore class (LibfaceCore.h) The 2 classes only have virtual methods. This allows for the uniform access to the backend independent of the implementation. So what I want before we get on with implementing is to finish off Eigenfaces and face detection based on Haar features. Last night I got the face detection to work but a lot of false negatives (Type II errors). I think it is important to get the face detection to work properly first, even if it means false negatives are not zero, but fairly low. Because without having coherent set of faces to work with there is no meaningful way of testing the face recognition part. This involves checking against different types of cascades that OpenCV 2.0 comes with and seeing which ones are better. Maybe even creating a "vote" system that allows for things with highest number of votes be passed as face. Then we can get on with recognition. Also I have been working on a Qt interface for this. Something very simple that allows rendering images and superposing rectangles on top for faces. I got QGraphicalScene to work, it is still a bit messy. But this is something I want to explore further. So will need someone to help out with that. Lastly, if you indeed interested in working on this library further, and I think there is a lot of future for it then log in to Sourceforge and request to be added to the project and I shall add you promptly to the set of developers so you can commit code and changes. :) Sorry about long first email, just a lot had to be said :) Alex |