face detection process...
Locked
 
|
Hi all,
Just few words about face detection process using current svn implementation : - digiKam compiled and installed on a new computer. - Dabatase created from scratch. - duration 35 mn (PIV DC 3.6 Ghz, 2Gb RAM) for these stats : digiKam version 2.0.0-beta1 (rev.: 1211276) AVI: 4 JP2k: 11 JPG: 1204 MOV: 1 MP3: 3 MP4: 2 MPEG: 2 PGF: 1 PNG: 186 PPM: 3 RAW-ARW: 27 RAW-CR2: 17 RAW-CRW: 6 RAW-DCR: 2 RAW-DNG: 62 RAW-HDR: 1 RAW-MRW: 29 RAW-NEF: 16 RAW-ORF: 2 RAW-PEF: 7 RAW-RAF: 9 RAW-RAW: 1 RAW-X3F: 2 TIFF: 50 XCF: 12 Total Items: 1660 : Albums: 106 Tags: 165 : Database backend: QSQLITE - 133 unknown items found. - 5 items are not faces. All others are fine. Bugs : - when scan is complete, progress bar continue to move... - There is an album displayed as Unknown where there is no item. Why ? Look screenshot for details. in fact, it's an icon view problem at scrolling. Moving contents with mouse wheel change icon view. http://www.flickr.com/photos/digikam/5322873069 http://www.flickr.com/photos/digikam/5322873101 http://www.flickr.com/photos/digikam/5323478176 Gilles _______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
|
|
Just to clarify 128 are faces and 5 are not is that right? :)
Alex On 4 Jan 2011, at 10:17, Gilles Caulier wrote: Hi all, _______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
Re: face detection process...
|
|
Yes
Gilles 2011/1/4 Alex Jironkin <[hidden email]>
_______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
|
|
Oh wow, I am genuinely surprised and happy that only 5 out of 133 are false positives. What about the time it took to do the scans? I mean did it even matter?
Alex On 5 Jan 2011, at 07:31, Gilles Caulier wrote: Yes _______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
Re: face detection process...
|
|
As i said previously, it take 37 mns around to scan all my collections...
Gilles 2011/1/5 Alex Jironkin <[hidden email]>
_______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
|
|
How many pictures you have? I just wanted to check because when I run detection its not very good. So 128 faces out of how many possible ones, roughly?
Alex On 5 Jan 2011, at 08:48, Gilles Caulier wrote: As i said previously, it take 37 mns around to scan all my collections... _______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
Re: face detection process...
|
|
AVI: 4
JP2k: 11 JPG: 1204 MOV: 1 MP3: 3 MP4: 2 MPEG: 2 PGF: 1 PNG: 186 PPM: 3 RAW-ARW: 27 RAW-CR2: 17 RAW-CRW: 6 RAW-DCR: 2 RAW-DNG: 62 RAW-HDR: 1 RAW-MRW: 29 RAW-NEF: 16 RAW-ORF: 2 RAW-PEF: 7 RAW-RAF: 9 RAW-RAW: 1 RAW-X3F: 2 TIFF: 50 XCF: 12 Total Items: 1660 Gilles 2011/1/5 Alex Jironkin <[hidden email]>
_______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
|
|
Oh dear. I was excited at first, but wow 128 of 1660 thats terrible. I have another copy in 0.2 branch that I think works a lot better. Will add false positives I think. Also there was a change I had to make to fix a memory leak which will break kface build I think I haven't tested that yet.
Alex On 5 Jan 2011, at 21:56, Gilles Caulier wrote: AVI: 4 _______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
Re: face detection process...
|
|
Yes 128 out of 1660 is bad. Is the algorithm used in the program in
0.2 branch different? Just asking as I have not checked it out yet. Amey Dharwadker On 1/6/11, Alex Jironkin <[hidden email]> wrote: > Oh dear. I was excited at first, but wow 128 of 1660 thats terrible. I have > another copy in 0.2 branch that I think works a lot better. Will add false > positives I think. Also there was a change I had to make to fix a memory > leak which will break kface build I think I haven't tested that yet. > > > Alex > > On 5 Jan 2011, at 21:56, Gilles Caulier wrote: > >> AVI: 4 >> JP2k: 11 >> JPG: 1204 >> MOV: 1 >> MP3: 3 >> MP4: 2 >> MPEG: 2 >> PGF: 1 >> PNG: 186 >> PPM: 3 >> RAW-ARW: 27 >> RAW-CR2: 17 >> RAW-CRW: 6 >> RAW-DCR: 2 >> RAW-DNG: 62 >> RAW-HDR: 1 >> RAW-MRW: 29 >> RAW-NEF: 16 >> RAW-ORF: 2 >> RAW-PEF: 7 >> RAW-RAF: 9 >> RAW-RAW: 1 >> RAW-X3F: 2 >> TIFF: 50 >> XCF: 12 >> Total Items: 1660 >> >> Gilles >> >> 2011/1/5 Alex Jironkin <[hidden email]> >> How many pictures you have? I just wanted to check because when I run >> detection its not very good. So 128 faces out of how many possible ones, >> roughly? >> >> Alex >> >> On 5 Jan 2011, at 08:48, Gilles Caulier wrote: >> >>> As i said previously, it take 37 mns around to scan all my collections... >>> >>> Gilles >>> >>> 2011/1/5 Alex Jironkin <[hidden email]> >>> Oh wow, I am genuinely surprised and happy that only 5 out of 133 are >>> false positives. What about the time it took to do the scans? I mean did >>> it even matter? >>> >>> >>> Alex >>> >>> On 5 Jan 2011, at 07:31, Gilles Caulier wrote: >>> >>>> Yes >>>> >>>> Gilles >>>> >>>> 2011/1/4 Alex Jironkin <[hidden email]> >>>> Just to clarify 128 are faces and 5 are not is that right? :) >>>> >>>> >>>> Alex >>>> >>>> On 4 Jan 2011, at 10:17, Gilles Caulier wrote: >>>> >>>>> Hi all, >>>>> >>>>> Just few words about face detection process using current svn >>>>> implementation : >>>>> >>>>> - digiKam compiled and installed on a new computer. >>>>> - Dabatase created from scratch. >>>>> - duration 35 mn (PIV DC 3.6 Ghz, 2Gb RAM) for these stats : >>>>> >>>>> digiKam version 2.0.0-beta1 (rev.: 1211276) >>>>> AVI: 4 >>>>> JP2k: 11 >>>>> JPG: 1204 >>>>> MOV: 1 >>>>> MP3: 3 >>>>> MP4: 2 >>>>> MPEG: 2 >>>>> PGF: 1 >>>>> PNG: 186 >>>>> PPM: 3 >>>>> RAW-ARW: 27 >>>>> RAW-CR2: 17 >>>>> RAW-CRW: 6 >>>>> RAW-DCR: 2 >>>>> RAW-DNG: 62 >>>>> RAW-HDR: 1 >>>>> RAW-MRW: 29 >>>>> RAW-NEF: 16 >>>>> RAW-ORF: 2 >>>>> RAW-PEF: 7 >>>>> RAW-RAF: 9 >>>>> RAW-RAW: 1 >>>>> RAW-X3F: 2 >>>>> TIFF: 50 >>>>> XCF: 12 >>>>> Total Items: 1660 >>>>> : >>>>> Albums: 106 >>>>> Tags: 165 >>>>> : >>>>> Database backend: QSQLITE >>>>> >>>>> - 133 unknown items found. >>>>> - 5 items are not faces. All others are fine. >>>>> >>>>> Bugs : >>>>> - when scan is complete, progress bar continue to move... >>>>> - There is an album displayed as Unknown where there is no item. Why ? >>>>> Look screenshot for details. in fact, it's an icon view problem at >>>>> scrolling. Moving contents with mouse wheel change icon view. >>>>> >>>>> http://www.flickr.com/photos/digikam/5322873069 >>>>> http://www.flickr.com/photos/digikam/5322873101 >>>>> http://www.flickr.com/photos/digikam/5323478176 >>>>> >>>>> Gilles >>>>> _______________________________________________ >>>>> Digikam-devel mailing list >>>>> [hidden email] >>>>> https://mail.kde.org/mailman/listinfo/digikam-devel >>>> >>>> If we knew what we were doing, it wouldn't be called research, would it? >>>> -- Albert Einstein >>>> >>>> >>>> >>>> >>>> _______________________________________________ >>>> Digikam-devel mailing list >>>> [hidden email] >>>> https://mail.kde.org/mailman/listinfo/digikam-devel >>>> >>>> >>>> _______________________________________________ >>>> Digikam-devel mailing list >>>> [hidden email] >>>> https://mail.kde.org/mailman/listinfo/digikam-devel >>> >>> If we knew what we were doing, it wouldn't be called research, would it? >>> -- Albert Einstein >>> >>> >>> >>> >>> _______________________________________________ >>> Digikam-devel mailing list >>> [hidden email] >>> https://mail.kde.org/mailman/listinfo/digikam-devel >>> >>> >>> _______________________________________________ >>> Digikam-devel mailing list >>> [hidden email] >>> https://mail.kde.org/mailman/listinfo/digikam-devel >> >> If we knew what we were doing, it wouldn't be called research, would it? >> -- Albert Einstein >> >> >> >> >> _______________________________________________ >> Digikam-devel mailing list >> [hidden email] >> https://mail.kde.org/mailman/listinfo/digikam-devel >> >> >> _______________________________________________ >> Digikam-devel mailing list >> [hidden email] >> https://mail.kde.org/mailman/listinfo/digikam-devel > > If we knew what we were doing, it wouldn't be called research, would it? > -- Albert Einstein > > > > -- Amey Dharwadker Final Year Undergraduate, Electronics and Communication Engineering, NIT Trichy _______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
Re: face detection process...
|
|
2011/1/6 Amey Dharwadker <[hidden email]> Yes 128 out of 1660 is bad. No. A lots of my photos do not have face, and there is a lots of duplicates files. It's a test collection on a test computer. My real photo collections is very huge compared and is hosted on a removable media. I will try to process it soon to compare and test. Gilles Caulier. Is the algorithm used in the program in _______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
|
|
In reply to this post by Amey Dharwadker
Yes it is quiet different. The current trunk uses many different cascades then choses "faces" that have an overlap. Where as 0.2 branch I chose the best cascade I could and tweaked parameters. I think I get much better result on the images I have. But that might not be indicative of other people collection. If you checkout 0.2 branch, there is also a GUI u can build and test individual images to see different results. Build with -DBUILD_EXAMPLES_GUI=ON flag in cmake.
Alex On 6 Jan 2011, at 07:58, Amey Dharwadker wrote: > Yes 128 out of 1660 is bad. Is the algorithm used in the program in > 0.2 branch different? Just asking as I have not checked it out yet. > > Amey Dharwadker > > On 1/6/11, Alex Jironkin <[hidden email]> wrote: >> Oh dear. I was excited at first, but wow 128 of 1660 thats terrible. I have >> another copy in 0.2 branch that I think works a lot better. Will add false >> positives I think. Also there was a change I had to make to fix a memory >> leak which will break kface build I think I haven't tested that yet. >> >> >> Alex >> >> On 5 Jan 2011, at 21:56, Gilles Caulier wrote: >> >>> AVI: 4 >>> JP2k: 11 >>> JPG: 1204 >>> MOV: 1 >>> MP3: 3 >>> MP4: 2 >>> MPEG: 2 >>> PGF: 1 >>> PNG: 186 >>> PPM: 3 >>> RAW-ARW: 27 >>> RAW-CR2: 17 >>> RAW-CRW: 6 >>> RAW-DCR: 2 >>> RAW-DNG: 62 >>> RAW-HDR: 1 >>> RAW-MRW: 29 >>> RAW-NEF: 16 >>> RAW-ORF: 2 >>> RAW-PEF: 7 >>> RAW-RAF: 9 >>> RAW-RAW: 1 >>> RAW-X3F: 2 >>> TIFF: 50 >>> XCF: 12 >>> Total Items: 1660 >>> >>> Gilles >>> >>> 2011/1/5 Alex Jironkin <[hidden email]> >>> How many pictures you have? I just wanted to check because when I run >>> detection its not very good. So 128 faces out of how many possible ones, >>> roughly? >>> >>> Alex >>> >>> On 5 Jan 2011, at 08:48, Gilles Caulier wrote: >>> >>>> As i said previously, it take 37 mns around to scan all my collections... >>>> >>>> Gilles >>>> >>>> 2011/1/5 Alex Jironkin <[hidden email]> >>>> Oh wow, I am genuinely surprised and happy that only 5 out of 133 are >>>> false positives. What about the time it took to do the scans? I mean did >>>> it even matter? >>>> >>>> >>>> Alex >>>> >>>> On 5 Jan 2011, at 07:31, Gilles Caulier wrote: >>>> >>>>> Yes >>>>> >>>>> Gilles >>>>> >>>>> 2011/1/4 Alex Jironkin <[hidden email]> >>>>> Just to clarify 128 are faces and 5 are not is that right? :) >>>>> >>>>> >>>>> Alex >>>>> >>>>> On 4 Jan 2011, at 10:17, Gilles Caulier wrote: >>>>> >>>>>> Hi all, >>>>>> >>>>>> Just few words about face detection process using current svn >>>>>> implementation : >>>>>> >>>>>> - digiKam compiled and installed on a new computer. >>>>>> - Dabatase created from scratch. >>>>>> - duration 35 mn (PIV DC 3.6 Ghz, 2Gb RAM) for these stats : >>>>>> >>>>>> digiKam version 2.0.0-beta1 (rev.: 1211276) >>>>>> AVI: 4 >>>>>> JP2k: 11 >>>>>> JPG: 1204 >>>>>> MOV: 1 >>>>>> MP3: 3 >>>>>> MP4: 2 >>>>>> MPEG: 2 >>>>>> PGF: 1 >>>>>> PNG: 186 >>>>>> PPM: 3 >>>>>> RAW-ARW: 27 >>>>>> RAW-CR2: 17 >>>>>> RAW-CRW: 6 >>>>>> RAW-DCR: 2 >>>>>> RAW-DNG: 62 >>>>>> RAW-HDR: 1 >>>>>> RAW-MRW: 29 >>>>>> RAW-NEF: 16 >>>>>> RAW-ORF: 2 >>>>>> RAW-PEF: 7 >>>>>> RAW-RAF: 9 >>>>>> RAW-RAW: 1 >>>>>> RAW-X3F: 2 >>>>>> TIFF: 50 >>>>>> XCF: 12 >>>>>> Total Items: 1660 >>>>>> : >>>>>> Albums: 106 >>>>>> Tags: 165 >>>>>> : >>>>>> Database backend: QSQLITE >>>>>> >>>>>> - 133 unknown items found. >>>>>> - 5 items are not faces. All others are fine. >>>>>> >>>>>> Bugs : >>>>>> - when scan is complete, progress bar continue to move... >>>>>> - There is an album displayed as Unknown where there is no item. Why ? >>>>>> Look screenshot for details. in fact, it's an icon view problem at >>>>>> scrolling. Moving contents with mouse wheel change icon view. >>>>>> >>>>>> http://www.flickr.com/photos/digikam/5322873069 >>>>>> http://www.flickr.com/photos/digikam/5322873101 >>>>>> http://www.flickr.com/photos/digikam/5323478176 >>>>>> >>>>>> Gilles >>>>>> _______________________________________________ >>>>>> Digikam-devel mailing list >>>>>> [hidden email] >>>>>> https://mail.kde.org/mailman/listinfo/digikam-devel >>>>> >>>>> If we knew what we were doing, it wouldn't be called research, would it? >>>>> -- Albert Einstein >>>>> >>>>> >>>>> >>>>> >>>>> _______________________________________________ >>>>> Digikam-devel mailing list >>>>> [hidden email] >>>>> https://mail.kde.org/mailman/listinfo/digikam-devel >>>>> >>>>> >>>>> _______________________________________________ >>>>> Digikam-devel mailing list >>>>> [hidden email] >>>>> https://mail.kde.org/mailman/listinfo/digikam-devel >>>> >>>> If we knew what we were doing, it wouldn't be called research, would it? >>>> -- Albert Einstein >>>> >>>> >>>> >>>> >>>> _______________________________________________ >>>> Digikam-devel mailing list >>>> [hidden email] >>>> https://mail.kde.org/mailman/listinfo/digikam-devel >>>> >>>> >>>> _______________________________________________ >>>> Digikam-devel mailing list >>>> [hidden email] >>>> https://mail.kde.org/mailman/listinfo/digikam-devel >>> >>> If we knew what we were doing, it wouldn't be called research, would it? >>> -- Albert Einstein >>> >>> >>> >>> >>> _______________________________________________ >>> Digikam-devel mailing list >>> [hidden email] >>> https://mail.kde.org/mailman/listinfo/digikam-devel >>> >>> >>> _______________________________________________ >>> Digikam-devel mailing list >>> [hidden email] >>> https://mail.kde.org/mailman/listinfo/digikam-devel >> >> If we knew what we were doing, it wouldn't be called research, would it? >> -- Albert Einstein >> >> >> >> > > > -- > Amey Dharwadker > Final Year Undergraduate, > Electronics and Communication Engineering, > NIT Trichy > _______________________________________________ > Digikam-devel mailing list > [hidden email] > https://mail.kde.org/mailman/listinfo/digikam-devel If we knew what we were doing, it wouldn't be called research, would it? -- Albert Einstein _______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
|
|
In reply to this post by Gilles Caulier-4
> - 133 unknown items found. > - 5 items are not faces. All others are fine. Judging this result of course depends on the "true positive" number: How many pictures truly have a face? The important part here, for use in digikam, is to run real-world tests: A collection where only 10-20% have faces at all. The initial implementation gave me lots of false positives. And this gave a bad first impression: almost all found entries were not faces. Users will see and judge that it doesnt work at all. We must have most entries correct there. Why the difference? The Positive Predictive Value of a test depends on the prevalence! The current implementation is a two step approach: 1) With a single cascade, faces are scanned 2) With a larger number of cascades, faces are verified. Step 1 determines the upper limit of sensitivity. Step 2 improves specificity and decreases sensitivity (but in my experience, not much) > > Bugs : > - when scan is complete, progress bar continue to move... Does it display a percentage number? > - There is an album displayed as Unknown where there is no item. Why ? Look > screenshot for details. in fact, it's an icon view problem at scrolling. > Moving contents with mouse wheel change icon view. > > http://www.flickr.com/photos/digikam/5322873069 > http://www.flickr.com/photos/digikam/5322873101 > http://www.flickr.com/photos/digikam/5323478176 This is a model-view problem. I have seen it in a few cases, usually a problem with sorting. No idea atm, did never seen that with faces. Marcel _______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
Re: face detection process...
|
|
2011/1/6 Marcel Wiesweg <[hidden email]>
Not too many. there is a lots of nature photo. some from my son and familly events. Judging visually, this sound correct for me. i can test with my huge collection at home where a lots of picture have face. I can select alum where i know that all photo have at least a face and see the result. The important part here, for use in digikam, is to run real-world tests: A No percentage. In fact progress bar use moving mode indication (when minimum and maximum values are set to 0) Gilles
_______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
|
|
> Not too many. there is a lots of nature photo. some from my son and familly > events. Judging visually, this sound correct for me. Then the result may be pretty good :-) > > i can test with my huge collection at home where a lots of picture have > face. I can select alum where i know that all photo have at least a face > and see the result. We'll get user feedback. So far at least no open complaints about detection performance ;-) > > > - when scan is complete, progress bar continue to move... > > > > Does it display a percentage number? > > No percentage. In fact progress bar use moving mode indication (when > minimum and maximum values are set to 0) Seems everyone but me has this bug! In BatchFaceDetector::startAlbumListing() (utilities/batch/batchfacedetector.cpp:179), the numbers are calculated. Why are they 0? Marcel _______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
Re: face detection process...
|
|
2011/1/7 Marcel Wiesweg <[hidden email]>
are they 0? The code sound logic for me. It's like a bug in DB interface... Gilles _______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
|
|
In reply to this post by alexjironkin
Alex Jironkin píše v Čt 06. 01. 2011 v 09:40 +0000:
> Yes it is quiet different. The current trunk uses many different > cascades then choses "faces" that have an overlap. Where as 0.2 > branch Is this trunk or 0.2 branch anonymously available somewhere? I can only find this with some files about face detection: http://websvn.kde.org/branches/extragear/graphics/digikam/ Nothing in trunk: http://websvn.kde.org/trunk/extragear/graphics/digikam/ Sorry if it is a stupid question, I do not know much about svn. regards, Milan -- http://www.milan-knizek.net/ _______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
Re: face detection process...
|
|
2011/1/10 Milan Knížek <[hidden email]>
Alex Jironkin píše v Čt 06. 01. 2011 v 09:40 +0000: Yes, all is there... Gilles Caulier _______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
Re: face detection process...
|
|
The libface library itself is hosted on sourceforge -- http://libface.sourceforge.net/file/Home.html
Marty 2011/1/10 Gilles Caulier <[hidden email]>
_______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
|
|
In reply to this post by alexjironkin
Alex Jironkin píše v Čt 06. 01. 2011 v 09:40 +0000:
> Yes it is quiet different. The current trunk uses many different > cascades then choses "faces" that have an overlap. Where as 0.2 branch > I chose the best cascade I could and tweaked parameters. I think I get > much better result on the images I have. But that might not be > indicative of other people collection. If you checkout 0.2 branch, > there is also a GUI u can build and test individual images to see > different results. Build with -DBUILD_EXAMPLES_GUI=ON flag in cmake. The 0.2 branch of libface really does a better job at face detection than current digiKam's svn/trunk. libfaceGUI is usually able to detect more faces, while not adding false positives. P.S. For testing purposes, it would be easier (faster) if libfaceGUI accepted a CLI parameter being a filename of the image. Regards, Milan -- http://www.milan-knizek.net/ _______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
Re: face detection process...
|
|
Right, the libface 0.2 branch is doing lot better than the current
Digikam version. -- Amey Dharwadker Final Year Undergraduate, Electronics and Communication Engineering, NIT Trichy _______________________________________________ Digikam-devel mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-devel |
«
Return to digikam-devel
|
1 view|%1 views
| Free forum by Nabble | Edit this page |