Objects/Faces Detection Toolbox

Objects/Faces detection using Local Binary Patterns and Haar features

Description

Objects/Faces detection toolbox v 0.26
--------------------------------------

This toolbox provides some tools for objects/faces detection using Local Binary Patterns (and some variants) and Haar features.
Object/face detection is performed by evaluating trained models over multi-scan windows with boosting models
(such Adaboosting, FastAdaboosting and Gentleboosting) or with linear SVM models.
The main objective of FDT is to bring simple but efficient tools mainly written in C codes with a matlab interface and easy to modify.

BEFORE INSTALLATION, BE SURE TO HAVE A C COMPILER ON YOUR SYSTEM!!!!!

For windows system, recommanded compilers are MSVC/MSVC express (free)/Intel compiler
For Linux system, recommanded compilers are GCC(free)/Intel compiler

PLEASE BE SURE THAT YOU SETUP YOUR COMPILER BEFORE FDT INSTALLATION.

Please open "readme.txt" for full details and references

Acknowledgements

Vcapg2 and Windows Internet Explorer + Google Search = Ur Ls For Jpg Images inspired this file.

• MATLAB release: MATLAB 7.8 (R2009a)
• Other requirements: A C compiler (MSVC/Intel CPP)

[haha]

I was trying to, if i come up with an answer, I will be glad to share it with you

[Sebastien PARIS]

Unfortunatly with two branches ... The table lookup version is much better.. If you want to do it with my toolbox, I will include your work :)

[haha]

Thank you a lot for answering, but did you just use the simple decision stump with two branches as weaklearner other than the lookup table like learner?

[haha]

 

[Sebastien PARIS]

haha,

My choice of decision tree as weaklearner for MB-LBP is not the best one coz LBP values span a non continous space. There is another choice of weak-learner but I didn't implement it

[haha]

Hello, I really appreciate your work a lot as I have stuck by the weak classifier of MB-LBP for a long time. I noticed that you have refered an article Face Detection Based on Multi-Block LBP Representation, did you use the week classifier of multi-branch decision trees in your code?
Thank you a lot for answering my question!

[Sebastien PARIS]

Mina,

I know there are some problems when linking with the inshipped BLAS/LAPACK Lib. Please install the last intel MKL version and link like

mex -v -DOS64 -DOMP fast_haar_adaboost_binary_train_cascade.c "C:\Program Files (x86)\Intel\Composer XE 2013\mkl\lib\intel64\mkl_core.lib" "C:\Program Files (x86)\Intel\Composer XE 2013\mkl\lib\intel64\mkl_intel_lp64.lib" "C:\Program Files (x86)\Intel\Composer XE 2013\mkl\lib\intel64\mkl_intel_thread.lib" -largeArrayDims

[Mina]

Hi,Sebastien.

I try to use the example 1 in train_cascade. However, Matlab crashes on fast_haar_ada_weaklearner mex-file and I get Segmentation error, this happens on BLASCALL.

Here is my system info:
MATLAB Version: 8.0.0.783 (R2012b)
Operating System: Microsoft Windows 7 Version 6.1 (Build 7601: Service Pack 1)
Java Version: Java 1.6.0_17-b04 with Sun Microsystems Inc. Java HotSpot(TM) 64-Bit Server VM mixed mode.

Thanks in advance.

[Tanushri]

Hello,
When I try to run the demo_haar as well as demo_mblbp.m files, it says:

Error using load
Unable to read file model_detector_mblbp_24x24_4: No such file or directory.
similarly is the case for model_detector_haar_24x24.

Kindly help in this regard

[qi]

hi,Sebastien,when I run the train_cascade.m, I set the specific parameters as follows: options.algoboost = 2,The initial number of negative samples are 2500,The initial number of postives samples are 2500,options.rect_param =[ 19 types of features];
options.maxstage = 9;When the code run the stage=6,the results as follow:generate_data_cascade2 6 in 5132.308 s
Train weaklearner 1 in 2.969 s
stage 6/9, m = 1, alpham = 1.0000
stage 6/9, m = 1, betam = 0.0000
Train weaklearner 2 in 3.074 s
stage 6/9, m = 2, alpham = 1.0000
stage 6/9, m = 2, betam = 0.0000
Train weaklearner 3 in 3.074 s
stage 6/9, m = 3, alpham = 1.0000
stage 6/9, m = 3, betam = 0.0000
Train weaklearner 4 in 3.130 s
stage 6/9, m = 4, alpham = 1.0000
stage 6/9, m = 4, betam = 0.0000
Train weaklearner 5 in 3.088 s
stage 6/9, m = 5, alpham = 0.9010
stage 6/9, m = 5, betam = 0.0020
Train weaklearner 6 in 3.094 s
stage 6/9, m = 6, alpham = 0.7920
stage 6/9, m = 6, betam = 0.0080
Train weaklearner 7 in 3.127 s
stage 6/9, m = 7, alpham = 0.6550
stage 6/9, m = 7, betam = 0.0090
Train weaklearner 8 in 3.116 s
stage 6/9, m = 8, alpham = 0.7340
stage 6/9, m = 8, betam = 0.0090
Train weaklearner 9 in 3.113 s
stage 6/9, m = 9, alpham = 0.6600
stage 6/9, m = 9, betam = 0.0080
Train weaklearner 10 in 3.100 s
stage 6/9, m = 10, alpham = 0.6530
stage 6/9, m = 10, betam = 0.0090
Train weaklearner 11 in 3.140 s
stage 6/9, m = 11, alpham = 0.7430
stage 6/9, m = 11, betam = 0.0090
Train weaklearner 12 in 3.191 s
stage 6/9, m = 12, alpham = 0.7500
stage 6/9, m = 12, betam = 0.0090
Train weaklearner 13 in 3.049 s
stage 6/9, m = 13, alpham = 0.6300
stage 6/9, m = 13, betam = 0.0090
Train weaklearner 14 in 3.194 s
stage 6/9, m = 14, alpham = 0.5940
stage 6/9, m = 14, betam = 0.0090
Train weaklearner 15 in 3.241 s
stage 6/9, m = 15, alpham = 0.5550
stage 6/9, m = 15, betam = 0.0090
Train weaklearner 16 in 3.129 s
stage 6/9, m = 16, alpham = 0.6050
stage 6/9, m = 16, betam = 0.0090
Train weaklearner 17 in 3.128 s
stage 6/9, m = 17, alpham = 0.5730
stage 6/9, m = 17, betam = 0.0090
Train weaklearner 18 in 3.154 s
stage 6/9, m = 18, alpham = 0.5690
stage 6/9, m = 18, betam = 0.0090
Train weaklearner 19 in 3.104 s
stage 6/9, m = 19, alpham = 0.6060
stage 6/9, m = 19, betam = 0.0090
Train weaklearner 20 in 3.251 s
??? Attempted to access
options.betaperstage(6); index out of
bounds because
numel(options.betaperstage)=5.

Error in ==> train_cascade at 547
options.stat(: , nb_stage) =
[stat(:) ;
options.betaperstage(nb_stage) ;
options.alphaperstage(nb_stage)];
Have you met such problem some times ago?How can I improve it? Thanks.

[qi]

 

[Sebastien PARIS]

Neo,

I will have a look to your problem...
Can u send me your exact parameters variable ?

[Neo]

Hi,Sebastien.
I try to use the example 1 in train_cascade. However, It's pop up alert like below:
Attempted to access ufx(3); index out of bounds because numel(ufx)=2.

Error in train_stage_cascade (line 181)
currentthresh = ufx(co);

Error in train_cascade (line 532)[options , Xfa] = train_stage_cascade(Xtrain , ytrain , Xtest , ytest ,
options);

Error in Try_face_dect (line 103)
[options , model]= train_cascade(options);

Please help me since I cannot found the reason.

[qi]

Hi,Sébastien,I find the results are different when I debug the train_cascade.m in liunx(64bits) and in xp(32bits). The code is same . Can you tell me why? Thanks.

[Nagesh]

Hi,Please reply for my previous post...

[Sebastien PARIS]

J : Did you download the last version ?

[J]

i use the example 1 in train_cascade.m
all positive pictures are 24x24
just change the number of training and testing to be smaller
but it always came out the error message as follows
??? Attempted to access options.betaperstage(1); index out of bounds because numel(options.betaperstage)=0.

Error in ==> train_cascade at 527
options.stat(: , nb_stage) = [stat(:) ; options.betaperstage(nb_stage) ; options.alphaperstage(nb_stage)];

can anyone help me i can't figure it out

[Nagesh]

Its very nice Toolbox for face detection,It shows your tremendous effort for that.....Thank You very much for such a good Toolbox
I am referring IEEE paper which uses 20by20 image rather than 24by24 &
5 Haar like (rectangular)features...
Your toolbox is for 24by24 &
2 Haar like features (Whether I am Right?)
So how can I use this toolbox for 20by20 image & 5 Haar like features
([1 2][2 1][1 3][3 1][2 2])
& from where i can download the dataset of 20by20 images...

[Xiaobai]

hi,paris
sorry to bother you again. There's a questions about viola_24x24.mat i want to ask you. Pictures of faces and nonfaces, where do they come from?Did you collect them yourself or you use a popular database like Feret or PIE ?

[Sebastien PARIS]

Re-open the Haar_dico4 with the GUI, you will understand ....

[Xiaobai]

I'm afraid u are right. I don't quite know how to use it. But i figure out what the haar_dico_4.mat does. Are they same?? Can u please give me a demo of how to use gui_features_dictionary.m? Thank you..

[qi]

Dear Sir，
When I debug the omp.c,the result is :omp.c(10) : fatal error C1021: Invalid preprocessor command“#warning” .I try to modify it ,but the result also can not pass,please help me,thanks!

[Sebastien PARIS]

Xiaobai,

Did you run the small utiliy to design Haar patterns in \gui\gui_features_dictionary.m ?

I think you can design such patterns with it (not completly sure)

[Xiaobai]

dear Sebastien
I gotta say this toolbox is really of a great help to me. Now i wanna know if your code can compute 45o rotated Haar-like features because I only figued out how to use your code to compute upright figures. Hope u can help me out!! Thx in advance.

[Sebastien PARIS]

xiaobai,

An update will appear with the 2 missing files. Look at demo_haar.html to understand how to compute haar features.

[Xiaobai]

your work is excellent！！
but how come i can't find 'haar_dico_2.mat' negatives.zip and model_detector_haar_24x24 ?
Did you forget or something?
I'v been working on a project lately and i need to compute haar features of 20*20 samples? I don't know which part of your code does this job...sorry i'm not to good at matlab..

[qi]

dear Sebastien
I want to download your new updated Objects/Faces detection toolbox and some references ,but I can not find the address,please send them to my email :
[e-mail],thank you very much.

[Sebastien PARIS]

qi,

With a such rating, you should try another toolbox, or even better do yourself....

[qi]

hi,paris
I saw the Elapsed time of the Sixth example in the demo_haar.m is 612.424368 seconds On the Internet ,but when I try it,the result is (Elapsed time) 2727.809425 seconds. Please help me to find the reasons,thanks.

[amin]

Hi can someone explain to me this error:
Error using detector_haar
sum(cascade(1 , :)) <= T

Error in train_cascade_Example_1 (line 46)
tic,[D , stat] = detector_haar(I , model);,toc

[J]

I'm sorry it doesn't work too.
Could you provide a simple example?
thanks

[Parker]

Did you forget negatives.zip in latest build?

[Sebastien PARIS]

I just uploaded a new version. Hope will fix all the problems. Thank you for reporting.

[J]

train_cascade
alpha0=0.8
beta0=0.01
actually the all parameters are the same as Example 1 in train_cascade.m

[Sebastien PARIS]

J,

What kind of cascade are you using ?
What are the values of alpha0 and beta0 ?

[J]

Hi Sébastien
I have update to version 0.25
but with using the same training code
with Npostrain=Nnegtrain=1000
it runs perfectly in version 0.24 but not in 0.25
follows are the error message

??? Attempted to access options.betaperstage(1); index out of bounds because numel(options.betaperstage)=0.

Error in ==> train_cascade at 527
options.stat(: , nb_stage) = [stat(:) ; options.betaperstage(nb_stage) ;
options.alphaperstage(nb_stage)];

thanks

[payam]

 

[Fa Fa]

Hi Sebastien,
i need to create 511 binary Lookup table, for modified census transform feature and to work with adaboost with training images.
please can you help me?
Thanks

[J]

 

[Sebastien PARIS]

For the last stages of the cascade ... since the FA rate is very low ..., main time is spended to collect FA....

[J]

positive is 24*24 have 40 pictures.
negative is less than 500*500 have 1000 pictures.
most training time is on negative part.

[Sebastien PARIS]

What is the size of your pattern ? (24x24) ?

[J]

Thanks for answering.
But know i have another question.
I'm using "train_cascade" training my own model. And I change Nnegtrain into 1000.
Is it normal that it will take several days for training?
(I mean main loop in train_cascade)

[Sebastien PARIS]

J,

It's a know problem. I have to update a new version well I will find some moments.

As workaround, change the while loop by

while((currentbeta < beta0) && (lufx <= (co+1)))

Sébastien

[J]

HI
i'm trying to train my own model using train_cascade
i use the default value in nargin but change a little bit.
posext into bmp. Npostrain=Nnegtrain=25. Npostest=Nnegtest=15.
With these it cames out the error message 'Attempted to access ufx(3); index out of bounds because numel(ufx)=2.' in train_stage_cascade at 170.
i can't figure out this problem.
thanks.

BTW is the negative picture number must equals to the positive picture number?

[Sebastien PARIS]

Lucas,

You are right, I will change the description in Haar_featlist and dependencies

ny Number of rows of the pattern (default ny = 24)
nx Number of columns of the pattern (default nx = ny)
rect_param Features rectangles parameters (10 x nR), where nR is the total number of rectangles for the patterns.
(default Vertical(2 x 1) [1 ; -1] and Horizontal(1 x 2) [-1 , 1] patterns)
rect_param(: , i) = [ip ; wp ; hp ; nrip ; nr ; xr ; yr ; wr ; hr ; sr], where
if index of the current Haar's feature. ip = [1,...,nF], where nF is the total number of Haar's features
wp width of the current rectangle pattern defining current Haar's feature
hp height of the current rectangle pattern defining current Haar's feature
nrif total number of rectangles for the current Haar's feature if
nr index of the current rectangle in the current Haar's feature, nr=[1,...,nrif]
xr,yr top-left coordinates of the current rectangle of the current pattern
wr,hr width and height of the current rectangle of the current pattern
sr weights of the current rectangle of the current pattern

Please run gui_features_dictionary in the \gui subdir to build such parameters

Outputs
-------

F Features's list (6 x nF) in UINT32 where nF designs the total number of Haar features
F(: , i) = [if ; xf ; yf ; wf ; hf ; ir]
if index of the current feature, if = [1,....,nF] where nF is the total number of Haar's features (see nbfeat_haar function)
xf,yf top-left coordinates of the current Haar's feature
wf,hf width and height of the current feature Haar's feature
ir Linear index of the FIRST rectangle of the current Haar's feature according rect_param definition. ir is used internally in Haar function
(ir/10 + 1) is the matlab index of this first rectangle

[Lucas Chai]

Am I clear? I mean in function F = haar_featlist([ny] , [nx] , [rect_param]), the first parameter of the output F in ur description is iF(1,..,nF), but the result value shows it is actually the ip (index of the pattern in options.rect_param), u can check it.

And for the last parameter ir, if (ir/10+1)is given, the meaning is obvious, thanx again

[Sebastien PARIS]

Tsai,

Let me clarify :
ir Linear index of the FIRST rectangle of the current Haar feature according rect_param definition. ir is used internally in Haar function. (ir/10 + 1) is the matlab index of this first rectangle

Hope will help you ...

[Lucas Chai]

Dear Mr Paris,
for the 'haar_featlist' function below, the first and the last parameter of the output are confusd, i read the description of F, the first parameter is ip, but in ur files it is if, and the last parametr is ir index of rectangle definition of the current feature in rect_param, but the value is 0,20,40,70,100, not sure about it. Wud u pls spare some time for more explanations. Thanx

options.F = haar_featlist(Ny , Nx , options.rect_param);

[Sebastien PARIS]

Tsai,

If compilation with -DOMP was succesfull, now you can add a num_threads = number of core in the options structure. Please open *.c header for full instructions.

[Lucas Chai]

Hi, Mr Paris,
Can I ask how to set the OpenMP available?
The compiling result is ok, but not any information on openMP

[Sebastien PARIS]

levgenia,

Yes you have to fill the positives folder. Please read the readme.txt to find datasets of positives samples for face detection.

[Ievgeniia]

Hello,

Thank you very much for the tool!
I have a question regarding positives folder. It seems to be empty. Thus, I receive an error when executing, for instance, example 1 from train_cascade.
What am I doing wrong?
Thanks in advance

[Hussein]

Dear Sir,

Its Working Good Now, Thank You So Much For This An Amazing Tool, I Appreciate Your Effort !!

Warmest Regards,

[Sebastien PARIS]

It should worlks .... How is the light condition? try something with normal light....

[Hussein]

Dear Sir,
Thank you for your instant reply, i located the codes and it compiled perfectly,

Thank You..

I still have one last question, when i run Demo_Detector_haar, it runs fine and have no errors but why i could not see the GREEN square ? do i have to do something with the codes to make that square display ?? ..

Thanks in advance Sir ..

[Sebastien PARIS]

Hussain,

There is an error in mexme_fdt for Linux system (sorry I don't have any Linux/MacOs box). I have to update with others small bugfixes. Please locate

if(ispc)
...
else
libblas = '-lmwblas';
end

should work after ...

Sébastien

[Hussein]

Dear Sir,
I followed your instructions but when i run setup_fdt
it shows

>> setup_fdt
Unzipping negatives picts ...
Compile mex files ...
>>

is it supposed to be like this ?? because i could not see any compiling process !! im using Mac os x 10.6.8 Snow Leo, Matlab 2011a and gcc compiler ..

also when i run mexme_fdt it shows

>> mexme_fdt
??? Undefined function or variable "libblas".

Error in ==> mexme_fdt at 143
str = [str , files1{i} , '.c ' , libblas , strOMP];

please HELP !!

[Sebastien PARIS]

Hello,

As written in the description :
"For Windows system, default LCC compiler included in matlab can't compile all files, you should have some errors during installation.
".
Please choose a modern compiler, MSVC express for example which is free

[kgan]

I followed the earlier instructions however i keep on having the same error

Unzipping negatives picts ...
Compile mex files ...
compiling chlbp
compiling chlbp_adaboost_binary_train_cascade
compiling chlbp_adaboost_binary_predict_cascade
compiling chlbp_gentleboost_binary_train_cascade
compiling chlbp_gentleboost_binary_predict_cascade
compiling detector_haar
compiling detector_mblbp
compiling eval_chlbp
compiling eval_haar
compiling eval_haar_subwindow
compiling eval_mblbp
compiling eval_mblbp_subwindows
compiling haar
compiling haar_ada_weaklearner
compiling haar_adaboost_binary_train_cascade
compiling haar_adaboost_binary_predict_cascade
compiling haar_featlist
compiling haar_gentle_weaklearner
compiling haar_gentleboost_binary_train_cascade
compiling haar_gentleboost_binary_predict_cascade
compiling haar_scale
compiling imresize
compiling area
compiling homkermap
compiling homkertable
compiling eval_hmblbp_spyr_subwindow
compiling detector_mlhmslbp_spyr
compiling mblbp
compiling mblbp_ada_weaklearner
compiling mblbp_adaboost_binary_train_cascade
compiling mblbp_adaboost_binary_predict_cascade
compiling mblbp_featlist
compiling mblbp_gentle_weaklearner
compiling mblbp_gentleboost_binary_train_cascade
compiling mblbp_gentleboost_binary_predict_cascade
compiling rgb2gray
compiling fast_rotate
compiling haar_ada_weaklearner_memory
compiling haar_adaboost_binary_predict_cascade_memory
compiling haar_adaboost_binary_train_cascade_memory
compiling haar_gentle_weaklearner_memory
compiling haar_gentleboost_binary_predict_cascade_memory
compiling haar_gentleboost_binary_train_cascade_memory
compiling int8tosparse
compiling fast_haar_ada_weaklearner
compiling fast_haar_adaboost_binary_train_cascade
compiling train_dense
Error linear.cpp: .\tron.h: 4 syntax error; found `function' expecting `;'
Error linear.cpp: .\tron.h: 5 syntax error; found `{' expecting `;'
Error linear.cpp: .\tron.h: 5 skipping `{'
Error linear.cpp: .\tron.h: 6 syntax error; found `:' expecting `;'
Error linear.cpp: .\tron.h: 6 skipping `:'
Error linear.cpp: .\tron.h: 7 syntax error; found `double' expecting `;'
Error linear.cpp: .\tron.h: 7 illegal initialization for `fun'
Error linear.cpp: .\tron.h: 7 syntax error; found `0' expecting `;'
Error linear.cpp: .\tron.h: 7 unrecognized declaration
Warning linear.cpp: .\tron.h: 7 empty declaration
Error linear.cpp: .\tron.h: 8 syntax error; found `void' expecting `;'
Error linear.cpp: .\tron.h: 8 illegal initialization for `grad'
Error linear.cpp: .\tron.h: 8 syntax error; found `0' expecting `;'
Error linear.cpp: .\tron.h: 8 unrecognized declaration
Warning linear.cpp: .\tron.h: 8 empty declaration
Error linear.cpp: .\tron.h: 9 syntax error; found `void' expecting `;'
Error linear.cpp: .\tron.h: 9 illegal initialization for `Hv'
Error linear.cpp: .\tron.h: 9 syntax error; found `0' expecting `;'
Error linear.cpp: .\tron.h: 9 unrecognized declaration
Warning linear.cpp: .\tron.h: 9 empty declaration
Error linear.cpp: .\tron.h: 11 syntax error; found `int' expecting `;'
Error linear.cpp: .\tron.h: 11 illegal initialization for `get_nr_variable'
Error linear.cpp: .\tron.h: 11 syntax error; found `0' expecting `;'
Error linear.cpp: .\tron.h: 11 too many errors

C:\PROGRA~1\MATLAB\R2010B\BIN\MEX.PL: Error: Compile of 'linear.cpp' failed.

Failed to compile mex-files, unzip precompiled mex32

i am using matlab 2010b and LCC compiler for windows

[Claude Rico]

I am a final year student and hope to graduate if i finish my thesis project... can anyone help me how to make a real time face recognition with a displayed information in an extracted face... thanks and God bless!

[guofeng ma]

Dear Paris
By the way, if there is a detailed demon about the ROC drawing. I will be a great help.

[Sebastien PARIS]

Guofeng,

I just uploaded the version v0.22 with more examples especially how to use train_cascade.m. This update will appear in a couple of hours.

[guofeng ma]

i want train a face detector using new data set(haar feature,adaboost and gentleboost,conventional cascade),but do give a demo?
my email [e-mail]

[guofeng ma]

Please help it.
I run the "train_cascade.m Example 1" source code(options.usefa=1 and options.algoboost=0or1,2),but,Xfa is all face images(generate_data_cascade.m). is this right?

[guofeng ma]

Thank you for an answer.
I do not have any problem in myNootBook windowsVista 32. However, I cannot run it in myDeskPC windows7 64.and Compiling it does not have any problem in windows7 64.

[Sebastien PARIS]

ma,

For win32 system, your example is working. I have to check for win64. Thanks for reporting.
N.B. The version 0.3 is upcoming.

[guofeng ma]

 

[guofeng ma]

Please help it.
I become the error when I run the next source code, and Matlab is over.

load viola_24x24
load haar_dico_2

[Ny , Nx , P] = size(X);
II = image_integral_standard(X);
y = int8(y);
options.G = haar_matG(Ny , Nx , rect_param);
options.indexF = int32(0:size(options.G , 2)-1);
options.fine_threshold = 1;
w = (1/length(y))*ones(1 , length(y));

tic,[model , wnew] = fast_haar_ada_weaklearner(II , y , w , options);,toc

I think that it is the issue of fast_haar_ada_weaklearner no.

YmPC information
windows 7 64
Matlab R2011b Compiler:Microsoft Visual C++ 2010 Express in C:Program Files (x86) \Microsoft Visual Studio 10.0

[Mr Smart]

 

[Sebastien PARIS]

Some protections ? what do u mean ?. This code is open, no special protection. which function are u using ?

Sebastien

[TSE]

Dear Paris,I was working with your code and try to improve the function of lbp_feature, but I cant make it work it seems that you implemented some protection of your code,if so could you tell me how to remove them?

[Sebastien PARIS]

Tsai ... Unfortunetly no. But I don't think training will talk weeks ... especially with the multi-core support.

[Lucas Chai]

problem solved~ i still donno why. i just put the fold to the root DIR, and it works ok~ And Mr Paris, if i train the face detecter using adaboost for a few weeks, and due to some reasons(like the electricity is cut off), what I can do to continue the boosting progress using ur program

[Lucas Chai]

don't konw why, always appears C:\Users\Tsai\AppData\Local\Temp\mex_wzxP6o\chlbp.obj "D:\Program Files\MATLAB\R2008a\extern\lib\win32\microsoft\libmwblas.lib"
LINK : fatal error LNK1104: cannot open file "chlbp.mexw32"

[Sebastien PARIS]

Never tried with an IP-camera. If you know how to get a frame, there should not have any problem. All USB webcam should work too. For the last question, no limitation is imposed.

[Javier Rivera]

Hi there! I was wondering what kind of webcam I can use with this toolbox. I have a webcam that is accessed from an address (i.e. 111.222.333.444), and I want to know if it is possible to use this code with it, or just USB webcams. Also, how many faces can it detect from a single camera?

[Ashvin Kumar]

 

[Steve]

 

[Sebastien PARIS]

Raida : please install the free msvc express compiler (you should try to find the 2005 express since your are using 2009b). Then type mex -setup in matlab command.
Run setup_fdt after

[Raida Al Alawi]

Dear

My OS is Windows 7 (6 bits)
I am using R2009b version

I run setup_fdt and I got a series of warnings followed by the following error

Compile mex files ...
compiling chlbp
cannot open output file chlbp.mexw32:

C:\PROGRA~2\MATLAB\R2009B\BIN\MEX.PL: Error: Link of 'chlbp.mexw32' failed.

Failed to compile mex-files, unzip precompiled mex32

Then I run mexme_fdt and I got the following:

compiling chlbp
cannot open output file chlbp.mexw32:

C:\PROGRA~2\MATLAB\R2009B\BIN\MEX.PL: Error: Link of 'chlbp.mexw32' failed.

??? Error using ==> mex at 221
Unable to complete successfully.

Error in ==> mexme_fdt at 146
eval(['mex ' str])

Can you please help me in installing and using the toolbox

Regards

[Sebastien PARIS]

Chirath : what is your system ? Matlab version and compiler ?. I think you must use included LCC compiler. Please use the free msvc compiler instead.

[Chirath Dharshana]

Hi Sebastian,
For our project we need to use the face detection tool.
When we tried to compile mex files, following errors occured.

Error linear.cpp: .\tron.h: 4 syntax error; found `function' expecting `;'
Error linear.cpp: .\tron.h: 5 syntax error; found `{' expecting `;'
Error linear.cpp: .\tron.h: 5 skipping `{'
Error linear.cpp: .\tron.h: 6 syntax error; found `:' expecting `;'
Error linear.cpp: .\tron.h: 6 skipping `:'
Error linear.cpp: .\tron.h: 7 syntax error; found `double' expecting `;'
Error linear.cpp: .\tron.h: 7 illegal initialization for `fun'
Error linear.cpp: .\tron.h: 7 syntax error; found `0' expecting `;'
Error linear.cpp: .\tron.h: 7 unrecognized declaration
Warning linear.cpp: .\tron.h: 7 empty declaration
Error linear.cpp: .\tron.h: 8 syntax error; found `void' expecting `;'
Error linear.cpp: .\tron.h: 8 illegal initialization for `grad'
Error linear.cpp: .\tron.h: 8 syntax error; found `0' expecting `;'
Error linear.cpp: .\tron.h: 8 unrecognized declaration
Warning linear.cpp: .\tron.h: 8 empty declaration
Error linear.cpp: .\tron.h: 9 syntax error; found `void' expecting `;'
Error linear.cpp: .\tron.h: 9 illegal initialization for `Hv'
Error linear.cpp: .\tron.h: 9 syntax error; found `0' expecting `;'
Error linear.cpp: .\tron.h: 9 unrecognized declaration
Warning linear.cpp: .\tron.h: 9 empty declaration
Error linear.cpp: .\tron.h: 11 syntax error; found `int' expecting `;'
Error linear.cpp: .\tron.h: 11 illegal initialization for `get_nr_variable'
Error linear.cpp: .\tron.h: 11 syntax error; found `0' expecting `;'
Error linear.cpp: .\tron.h: 11 too many errors

C:\PROGRA~1\MATLAB\R2010A\BIN\MEX.PL: Error: Compile of 'linear.cpp' failed.

Failed to compile mex-files, unzip precompiled mex32

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

Please help us.
Thank you in advance.

[Sebastien PARIS]

Van: Yes for me these 2 files merged are my positives (there are some negatives also inside but I don't use them really)

I didn't know this paper. The idea is good to use "meta" haar features, but if I understand correctly you'll have much more features. Actually you have to modify mode code to create such features. I don't think it's really hard but should be very memory requiering

[Van]

 

[Van]

Hi Sebastien,
Thank you for your quick response. As I understand, these 2 files, viola_24x24, and jesen_24x24, are face databse in gray scale. Matrix x is face image, and y is idicator whether or not it's a posotive or negative. Am I right?
For increasing efficient of detection, I'd read some papers, one of them is:
Joint Haar-like Features for Face Detection
URL: dtpapers.googlecode.com/files/01544911.pdf
The paper above is introduce some extend Haar-like feature to improve efficience of detection. So, I wonder where you define Haar-like feature, and how you define them.

Thank you,

Van Quach

[Sebastien PARIS]

Van: You will need a such database for training your positives examples... I am sure they are a lot of such databases in the web. When you'll have, you maybe need to downscale each images to something close to 24x24 pixels

Sebastien

[Van]

Hi Paris,
Your code is very great. I learn a lot from your code. Regard to training haar detector, as I understood, before running train_cascade, we have to have 2 matrix: jensen_24x24 and viola_24x24 for detecting face. However, in my project, I intend to detect user's eyes, not the face. Could you please give me a hint how to create the matrix for training Haar detector? For more detail, I don't know how to create Xpos matrix for detecting eye:
----------------------
load viola_24x24
Xpos = X(: , : , find(y == 1));
load jensen_24x24
Xpos = cat(3 , Xpos , X(: , : , find(y == 1)));

...

[options , model] = train_cascade(Xpos , options);

-------
Thank you for your hard working :)

[Sebastien PARIS]

Jingsong,

Thx for reporting, fixed in the next update.

[Jingsong Xu]

excellent work.

I meet one problem.
When I use chlbp.c, here is the samples in chlbp.c

clear, close all
load viola_24x24
Ny = 24;
Nx = 24;
options.N = [8 , 4 , 12];
options.R = [1 , 1 , 2];
options.map = zeros(2^max(options.N) , length(options.N));

mapping = getmapping(options.N(1),'u2');
options.map(1:2^options.N(1) , 1) = mapping.table';
options.map(1:2^options.N(2) , 2) = (0:2^options.N(2)-1)';
mapping = getmapping(options.N(3),'u2');
options.map(1:2^options.N(3) , 3) = mapping.table';
options.shiftbox = cat(3 , [Ny , Nx ; 1 , 1] , [16 , 16 ; 4 , 4] , [Ny , Nx ; 1 , 1]);

options.T = 50;

H = chlbp(X , options);

after this I type

H2 = chlbp(X(:,:,2) , options);

and i find H2 is different H(:,:,2)?why?

[zhu ?]

nice work!!

[Sebastien PARIS]

Daniel,

Whithin few days. I am finishing all the test and debugging.

[Daniel]

Thank you very much for your quick answer Sebastien. Do you know aproximately when are you going to upload the new version?

Thank you, I'm very pleased to use your software.

[Sebastien PARIS]

Daniel,

A new major update is coming with better cascade training algorithms and plenty of other stuff.
However the actual parameter of demo_detector_haar should work for 1280x960 also

[Daniel]

Hi to everybody. Could anybody tell me which are the best input arguments for 'detector_haar' function to detect faces from a video source like this: 'vid = videoinput('winvideo', 1, 'RGB24_1280x960');'

I'm using demo_detector_haar and I don't know what's the best 'haar.model', 'cascade', 'min_detect', etc...

How can I know it? Have I do my own haar.model? how?

thank you very much for your work.

[shri]

hi am newbie please explain what feature_list func does? thanking you

[Sebastien PARIS]

If I can remember well, there was some numerical discrepencies without if a long serie of the same class occurs....

[Lucas Chai]

in function haar_adaboost_binary_model_cascade(x,y),why do u do permutation? what is the difference without it?

[zhu ?]

Hi, I confused on coventional cascade structure and multi-exit cascade structure. I need your kind to help me, I think they different like these:

Suppose that there are 10 Weaklearns: ① ② ③ ④ ⑤ ⑥ ⑦ ⑧ ⑨ ⑩

coventional cascade structure maybe like this:
layer-1 ①
layer-2 ② ③
layer-3 ④ ⑤ ⑥
layer-4 ⑦ ⑧ ⑨ ⑩

multi-exit cascade structure maybe like this:
layer-1 ①
layer-2 ① ② ③
layer-3 ① ② ③ ④ ⑤ ⑥
layer-4 ① ② ③ ④ ⑤ ⑥ ⑦ ⑧ ⑨ ⑩

This is my idea，can you agree with me ?

[zhu ?]

Hi Sebastien,
Which face databse you have used in your code? I found you use Viola_24×24.mat and Jensen_24×24.mat,but can you tell me they are come from which face database.

[Lucas Chai]

Sebastien Paris , wud pls upload the realadaboost code for training ? In this way, ur code will be more pupular

[Sebastien PARIS]

Please compile mex files for your system (and more generaly, read full instruction in readme.txt)

run "mexme_fdt.m" after be sure that "mex -setup" had been done before.

[Mariam Z]

while tryin the demo_mblbp, i found the error' Undefined function or method 'mblbp_featlist' for input arguments of type 'double'...can u please help me wid it?

[Sebastien PARIS]

Feel free to modify the code, it's an open and simple for that. However, I will say for you last remark, that the weaklearners selected by the fast algorithm and the classic gentleboost are almost the same (see HTML documentation)

[zhu ?]

I confused on the way to select min error, i think it may be selected like this:
alpha = var1c - var2c;

beta = (m1c*var2c - m2c*var1c);

gamma = m2c*m2c*var1c - m1c*m1c*var2c + 2*var1c*var2c*log(ctep*(var2c/var1c));

std1c = 1.0/sqrt(var1c);
if(alpha==0)
{
x1=-gamma/(2*beta);
x2=x1;
std2c = std1c;
}
else
{
std2c = 1.0/sqrt(var2c);
delta = beta*beta - alpha*gamma;
if(delta==0)
{
x1=(m1c+m2c)/2;
x2=x1;
}
else
{
delta=abs(delta);
sqrtdelta = sqrt(delta);
x1 = (-beta + sqrtdelta)/(alpha);
x2 = (-beta - sqrtdelta)/(alpha);
}
}


if(m1c>m2c)
{
if(x1=x2)
{
err11= 0.5*(p2*(1.0 - erf((x1 - m2c)*std2c)) + p1*(1.0 + erf((x1 - m1c)*std1c)));
err10=err11;
Err1=err11;
Err2=Err1;
opt1=1;
opt2=opt1;
temp_th=x1;
opt=opt1;
tempmin=Err1;

}
else
{
err11= 0.5*(p2*(1.0 - erf((x1 - m2c)*std2c)) + p1*(1.0 + erf((x1 - m1c)*std1c)));
err10=err11;
Err1=err11;
opt1=1;
err21= 0.5*(p2*(1.0 - erf((x2 - m2c)*std2c)) + p1*(1.0 + erf((x2 - m1c)*std1c)));
err20= 0.5*(p2*(1.0 + erf((x2 - m2c)*std2c)) + p1*(1.0 - erf((x2 - m1c)*std1c)));
if(err21<err20)
{
Err2=err21;
opt2=1;
}
else
{
Err2=err20;
opt2=-1;
}
if(Err1<Err2)
{
tempmin=Err1;
opt=opt1;
temp_th=x1;
}
else
{
tempmin=Err2;
opt=opt2;
temp_th=x2;
}
}
}
else
{
if(x1=x2)
{
err20= 0.5*(p2*(1.0 + erf((x2 - m2c)*std2c)) + p1*(1.0 - erf((x2 - m1c)*std1c)));
err21=err20;
Err2=err20;
Err1=Err2;
opt2=-1;
opt1=opt2;
temp_th=x2;
opt=opt2;
tempmin=Err2;

}
else
{
err20= 0.5*(p2*(1.0 + erf((x2 - m2c)*std2c)) + p1*(1.0 - erf((x2 - m1c)*std1c)));
err21=err20;
Err2=err20;
opt2=-1;
err10= 0.5*(p2*(1.0 + erf((x1 - m2c)*std2c)) + p1*(1.0 - erf((x1 - m1c)*std1c)));
err11= 0.5*(p2*(1.0 - erf((x1 - m2c)*std2c)) + p1*(1.0 + erf((x1 - m1c)*std1c)));
if(err11<err10)
{
Err1=err11;
opt1=1;
}
else
{
Err1=err10;
opt1=-1;
}
if(Err1<Err2)
{
tempmin=Err1;
opt=opt1;
temp_th=x1;
}
else
{
tempmin=Err2;
opt=opt2;
temp_th=x2;
}
}
}


if(tempmin<Errormin)
{
Errormin=tempmin;
featuresIdx_opt = f;
th_opt = temp_th;
a_opt = opt;
}

[zhu ?]

i still can't understand you method for calculating erf,can you tell me how to resolve ?

[Sebastien PARIS]

Hi,

It's right, is a bug... However, in pratice the case where alpha = 0, never occurs.... Thanks you for reporting.

[zhu ?]

hi
i can 't understand the way to calculate erf,can you tell me you method ? 3q

[zhu ?]

hi
i think x1 = -gamma/beta is right,but in you code you write
x1 = -beta/gamma,so i think it may be a bug.

[Sebastien PARIS]

Hi,

In "Digital Image Processing", Rafael C. Gonzalez, P 622-623 see equations. Actually my implementation seems right.

p1 = 0.5;
p2 = 0.5;

m1 = -3 ;
m2 = 1;

s1 = (1.2)^2;
s2 = (1.4)^2;
cte = p1/p2;
alpha = s1 - s2;
beta = 2*(m1.*s2 - m2.*s1);
%beta = (m1.*s2 - m2.*s1);

gamma = (m2*m2).*s1 - (m1*m1).*s2 + 2*s1*2*s2.*log(cte.*sqrt(s2/s1));% 2*s1*2*s2.*log(cte.*sqrt(s2/s1)) = 2*s1*s2.*log(cte.*(s2/s1))

delta = beta.*beta - 4*alpha.*gamma;

% delta = beta.*beta - alpha.*gamma;

if(delta < 0)

delta = -delta;

end

if(alpha ~= 0)

x1 = (-beta + sqrt(delta))./(2*alpha);
x2 = (-beta - sqrt(delta))./(2*alpha);
%
% x1 = (-beta + sqrt(delta))./(alpha);
% x2 = (-beta - sqrt(delta))./(alpha);

else

x1 = -gamma/beta;
T = (m1 + m2)/2 + (sqrt(s1)/(m1 - m2))*log(cte);
x2 = x1;

end

[zhu ?]

Hi Sebastien
i think i find a bug:
if(alpha != 0.0)
{

x1 = (-beta + sqrtdelta)/(alpha);

x2 = (-beta - sqrtdelta)/(alpha);

}
else
{

x1 = -beta/gamma;

x2 = x1;

}
i think when alpha=0, x1=x2=-gamma/(2*beta) !

[zhu ?]

Hi Sebastien
i think i find a bug:
if(alpha != 0.0)
{

x1 = (-beta + sqrtdelta)/(alpha);

x2 = (-beta - sqrtdelta)/(alpha);

}
else
{

x1 = -beta/gamma;

x2 = x1;

}
i think when alpha=0, x1=x2=-gamma/(2*beta) !

[Sebastien PARIS]

Please read the referenced paper ....

[zhu ?]

Hi Sebastien
Can you tell me how to calculate th_opt in fast_haar_adaboost_binary_model_cascade.c .Specially, i didn't undetstand the code like this :
alpha = var1c - var2c;

beta = (m1c*var2c - m2c*var1c);

gamma = m2c*m2c*var1c - m1c*m1c*var2c + 2*var1c*var2c*log(ctep*(var2c/var1c));

delta = beta*beta - alpha*gamma;

if(delta < 0)
{

delta = -delta;

}

sqrtdelta = sqrt(delta);

std1c = 1.0/sqrt(var1c); /* 1.0/sqrt(2.0*var1c); */

std2c = 1.0/sqrt(var2c); /* 1.0/sqrt(2.0*var2c); */

if(alpha != 0.0)
{

x1 = (-beta + sqrtdelta)/(alpha);

x2 = (-beta - sqrtdelta)/(alpha);

}
else
{

x1 = -beta/gamma;

x2 = x1;

}

if(m1c > m2c)
{

Err1 = 0.5*(p2*(1.0 - erf((x1 - m2c)*std2c)) + p1*(1.0 + erf((x1 - m1c)*std1c)));

Err2 = 0.5*(p2*(1.0 - erf((x2 - m2c)*std2c)) + p1*(1.0 + erf((x2 - m1c)*std1c)));

}
else
{

Err1 = 0.5*(p1*(1.0 - erf((x1 - m1c)*std1c)) + p2*(1.0 + erf((x1 - m2c)*std2c)));

Err2 = 0.5*(p1*(1.0 - erf((x2 - m1c)*std1c)) + p2*(1.0 + erf((x2 - m2c)*std2c)));

}

if(Err1 < Err2)

{
if(Err1 < Errormin)
{

Errormin = Err1;

featuresIdx_opt = f;

th_opt = x1;

a_opt = -1.0;

}

}
else
{
if(Err2 < Errormin)
{

Errormin = Err2;

featuresIdx_opt = f;

th_opt = x2;

a_opt = 1.0;
}

}

}

[zhu ?]

Hi Sebastien,
I don not know BLAS fucntions, i can't find it in matlab help documents.

[Sebastien PARIS]

BLASCALL is an alias for calling a BLAS function....

[zhu ?]

Hi Sebastien,
Can you tell me what "BLASCALL" mean in fast_haar_adaboost_binary_model_cascade?

[Louis]

Hi Sebastien,
I have XP, dual core , and visual studio 2010. It does compile well, but when i run it on some models that i generate using chlbp_gentleboost_binary_model_cascade.c, it says "Matlab has encountered internal problem and needs to be closed". I tried to trace the eval_chlbp.c file and "i think" there are some errors there, i am not sure because my knowledge in C is close to none.

eval_chlbp.c is supposed to extract only the models features instead of extracting all the features as in chlbp.c; hence, it should be significantly faster correct?

[Sebastien PARIS]

Louis : which system you have ? CPU, OS, Compiler ?.
Regards,

[Louis]

Hi Sebastien,
I am using the function eval_chlbp which you have in the toolbox but it is giving me memory error saying that i have to close matlab :( ... I understand that you havent used it throughout the demo, but i need it since i am implementing a chlbp detector.

Do you have an updated version of or i am doing something wrong.

Many thanks

[zhu ?]

thank you ,i have download it in ieee

[Sebastien PARIS]

As referenced in the mex-file :

Reference : M-T Pham and T-J Cham, "Fast trainning and selection of Haar features using statistics in boosting-based face detection",
--------- in Proc. 11th IEEE ICCV'07

[zhu ?]

dear Sebastien
i can't understand you FastAdaboosting algorithm, because i can't understand "Haar_matG" funciton.
please tell me what your purpose for sparse,and what references you have used.

[Sebastien PARIS]

Hi,
Of course, you have different results especially with N = 4. You want to define th_opt = (a[i-1] + a[i])/2 and me by th_opt = (a[i] + a[i+1])/2, when N is large (and it's the case usually), it won't change a lot of for the value of th_opt. In your example, of couse, It changed with N = 4. You have 2 choices for defining the interval. You prefered the second choice, but in practice, with N> 1000, it won't change a lot....

[zhu ?]

dear Sébastien
i think is different，i have compared them with vc++6.0 and they have different result.

#include "stdafx.h"
void qsindex (double *a, int *index , int lo, int hi)
{

int i=lo, j=hi , ind;

double x=a[(lo+hi)/2] , h;

do
{
while (a[i]<x) i++;
while (a[j]>x) j--;
if (i<=j)
{
h = a[i];
a[i] = a[j];
a[j] = h;
ind = index[i];
index[i] = index[j];
index[j] = ind;
i++;
j--;
}
}
while (i<=j);

if (lo<j) qsindex(a , index , lo , j);
if (i<hi) qsindex(a , index , i , hi);
}

/*select best th_opt and a_opt,by Sebastien Paris*/
/*void main(int argc, char* argv[])
{

double a[4]={4,3,2,1};
int index[4]={0,1,2,3};
int y[4]={1,-1,1,1};
int N=4;
int N1;
N1=N-1;
double Tplus=0,Tminus=0;
double Errormin=2;
double w[4]={0.15,0.35,0.25,0.25};
int ind,ytemp;
double wtemp;
////////////////////////////before sort/////////////////////////////
for(int i=0;i<4;i++)
{
printf("a[%d] ",i);

}
printf("\n");
for( i=0;i<4;i++)
printf(" %g ",a[i]);
printf("\n");

for( i=0;i<4;i++)
printf(" %d ",index[i]);
printf("\n");
for( i=0;i<4;i++)
printf(" %d ",y[i]);
printf("\n");
for( i=0;i<4;i++)
printf(" %g ",w[i]);
printf("\n");


int lo=0,hi=N1;
qsindex (a, index , lo, hi);

///////////////////////////////after sort

for( i=0;i<4;i++)
{
printf("a[%d] ",i);

}
printf("\n");
for( i=0;i<4;i++)
printf(" %g ",a[i]);
printf("\n");

for( i=0;i<4;i++)
printf(" %d ",index[i]);
printf("\n");
for( i=0;i<4;i++)
printf(" %d ",y[index[i]]);
printf("\n");
for( i=0;i<4;i++)
printf(" %g ",w[index[i]]);
printf("\n");

for(i = 0 ; i < N ; i++)
{

if(y[i] == 1)
{

Tplus += w[i];

}
else
{

Tminus += w[i];

}
}

printf("Tplus=%g,Tminus=%g\n",Tplus,Tminus);
printf("*********************\n");

double Splus=0,Sminus=0;
double Errplus,Errminus;
double th_opt,a_opt;
//select best th_opt and a_opt,by Sebastien Paris
for(i = 0 ; i < N ; i++)

{

ind = index[i];

ytemp = y[ind];

wtemp = w[ind];

Errplus = Splus + (Tminus - Sminus);

Errminus = Sminus + (Tplus - Splus);

printf("Errplus=%g,Errminus=%g\n",Errplus,Errminus);

if(Errplus < Errormin)
{

Errormin = Errplus;

if(i<N1)
{

th_opt = (a[i] + a[i+1])/2;

}
else
{

th_opt = a[i];

}

a_opt = 1.0;

}

if(Errminus < Errormin)
{

Errormin = Errminus;

if(i<N1)
{

th_opt = (a[i] + a[i+1])/2;

}
else
{

th_opt = a[i];

}



a_opt = -1.0;

}


if(ytemp == 1)
{

Splus += wtemp;

}
else

{

Sminus += wtemp;

}
printf("when i=%d,a_opt=%g,th_opt=%g,Errormin=%g\n",i,a_opt,th_opt,Errormin);
printf("*********************\n");
}
printf("************finally*********\n");
printf("best a_opt=%g,best th_opt=%g,Errormin=%g\n",a_opt,th_opt,Errormin);
}
*/
/*select best th_opt in my opinion*/
void main(int argc, char* argv[])
{

double a[4]={4,3,2,1};
int index[4]={0,1,2,3};
int y[4]={1,-1,1,1};
int N=4;
int N1;
N1=N-1;
double Tplus=0,Tminus=0;
double Errormin=2;
double w[4]={0.15,0.35,0.25,0.25};
int ind,ytemp;
double wtemp;
///////////////////////////////before sort
printf("before sort:\n");
for(int i=0;i<4;i++)
{
printf("a[%d] ",i);

}
printf("\n");
for( i=0;i<4;i++)
printf(" %g ",a[i]);
printf("\n");

for( i=0;i<4;i++)
printf(" %d ",index[i]);
printf("\n");
for( i=0;i<4;i++)
printf(" %d ",y[i]);
printf("\n");
for( i=0;i<4;i++)
printf(" %g ",w[i]);
printf("\n");

int lo=0,hi=3;
qsindex (a, index , lo, hi);
///////////////////////////////after sort
printf("after sort:\n");
for( i=0;i<4;i++)
{
printf("a[%d] ",i);

}
printf("\n");
for( i=0;i<4;i++)
printf(" %g ",a[i]);
printf("\n");

for( i=0;i<4;i++)
printf(" %d ",index[i]);
printf("\n");
for( i=0;i<4;i++)
printf(" %d ",y[index[i]]);
printf("\n");
for( i=0;i<4;i++)
printf(" %g ",w[index[i]]);
printf("\n");

for(i = 0 ; i < N ; i++)
{

if(y[i] == 1)
{

Tplus += w[i];

}
else
{

Tminus += w[i];

}
}
printf("Tplus=%g,Tminus=%g\n",Tplus,Tminus);
printf("*********************\n");
double Splus=0,Sminus=0;
double Errplus,Errminus;
double th_opt,a_opt;
//select best th_opt and a_opt,in my opinion

for(i = 0 ; i < N ; i++)
{

ind = index[i];

ytemp = y[ind];

wtemp = w[ind];

Errplus = Splus + (Tminus - Sminus);

Errminus = Sminus + (Tplus - Splus);
printf("Errplus=%g,Errminus=%g\n",Errplus,Errminus);

if(Errplus < Errormin)
{

Errormin = Errplus;
if(i==0)
th_opt = a[i];

else
{

th_opt = (a[i-1] + a[i])/2;

}

a_opt = 1.0;

}

if(Errminus < Errormin)
{

Errormin = Errminus;

if(i==0)
th_opt = a[i];


else
{

th_opt = (a[i-1] + a[i])/2;

}


a_opt = -1.0;

}


if(ytemp == 1)
{

Splus += wtemp;

}
else

{

Sminus += wtemp;

}
printf("when i=%d,a_opt=%g,th_opt=%g,Errormin=%g\n",i,a_opt,th_opt,Errormin);
printf("*********************\n");
}
printf("*********finally********\n");
printf("best a_opt=%g,best th_opt=%g,Errormin=%g\n",a_opt,th_opt,Errormin);

}

[Sebastien PARIS]

I think it does not change really results .... just shit de threshold.

Sébastien

[zhu ?]

thank you ,Sebastien Paris!
I have resolved this problem. when i read the code in "haar_adaboost_binary_model_cascade.c",i the think the best th_opt should be selected like this:
for(i = 0 ; i < N ; i++)
{

ind = index[i];

ytemp = y[ind];

wtemp = w[ind];

Errplus = Splus + (Tminus - Sminus);

Errminus = Sminus + (Tplus - Splus);
printf("Errplus=%g,Errminus=%g\n",Errplus,Errminus);

if(Errplus < Errormin)
{

Errormin = Errplus;
if(i==0)
th_opt = a[i];

else
{

th_opt = (a[i-1] + a[i])/2;

}

a_opt = 1.0;

}

if(Errminus < Errormin)
{

Errormin = Errminus;

if(i==0)
th_opt = a[i];


else
{

th_opt = (a[i-1] + a[i])/2;

}


a_opt = -1.0;

}


if(ytemp == 1)
{

Splus += wtemp;

}
else

{

Sminus += wtemp;

}

[Sebastien PARIS]

If your'are not able to recompile vcapg2, please download compiled mex file at : http://www.mathworks.com/matlabcentral/fileexchange/2939-vcapg2

as mentioned in description

[zhu ?]

Hi！
I want to use demo_detector_haar in matlab 2009a(7.8.0),but it does't work. matlab remind you:
>> demo_detector_haar
??? Undefined function or method 'vcapg2' for input arguments of type 'double'.

Error in ==> demo_detector_haar at 33
aa = vcapg2(0,2);

so I recompile all code，I do like this:
>> mex -setup
Please choose your compiler for building external interface (MEX) files:

Would you like mex to locate installed compilers [y]/n? y

Select a compiler:
[1] Lcc-win32 C 2.4.1 in D:\PROGRA~1\MATLAB\R2009a\sys\lcc
[2] Microsoft Visual C++ 6.0 in D:\Program Files\Microsoft Visual Studio

[0] None

Compiler: 2

Please verify your choices:

Compiler: Microsoft Visual C++ 6.0
Location: D:\Program Files\Microsoft Visual Studio

Are these correct [y]/n? y

Trying to update options file: C:\Documents and Settings\Administrator\Application Data\MathWorks\MATLAB\R2009a\mexopts.bat
From template: D:\PROGRA~1\MATLAB\R2009a\bin\win32\mexopts\msvc60opts.bat

Done . . .

**************************************************************************
Warning: The MATLAB C and Fortran API has changed to support MATLAB
variables with more than 2^32-1 elements. In the near future
you will be required to update your code to utilize the new
API. You can find more information about this at:
http://www.mathworks.com/support/solutions/data/1-5C27B9.html?solution=1-5C27B9
Building with the -largeArrayDims option enables the new API.
**************************************************************************
>> mexme_fdt
compiling chlbp
compiling chlbp_adaboost_binary_model_cascade
compiling chlbp_adaboost_binary_predict_cascade
compiling chlbp_gentleboost_binary_model_cascade
compiling chlbp_gentleboost_binary_predict_cascade
compiling detector_haar
compiling detector_mblbp
compiling eval_chlbp
compiling eval_haar
compiling eval_haar_subwindow
compiling eval_mblbp
compiling haar
compiling haar_ada_weaklearner
compiling haar_adaboost_binary_model_cascade
compiling haar_adaboost_binary_predict_cascade
compiling haar_featlist
compiling haar_gentle_weaklearner
compiling haar_gentleboost_binary_model_cascade
compiling haar_gentleboost_binary_predict_cascade
compiling haar_scale
compiling imresize
compiling int8tosparse
compiling mblbp
compiling mblbp_ada_weaklearner
compiling mblbp_adaboost_binary_model_cascade
compiling mblbp_adaboost_binary_predict_cascade
compiling mblbp_featlist
compiling mblbp_gentle_weaklearner
compiling mblbp_gentleboost_binary_model_cascade
compiling mblbp_gentleboost_binary_predict_cascade
compiling rgb2gray
compiling fast_haar_ada_weaklearner
compiling fast_haar_adaboost_binary_model_cascade
I find it does not compile vcap2.cpp！why？
In my PC I have installed DirectShow SDK，and I can find dxtrans.h,what should I do next?

Please help me,thank you very much!

[adam akbar]

Wow amazing program, finally i can run it...
i have question, how to increase fps??

[adam akbar]

yes i have.
the vcapg2.cpp is not compile yet.
can you show me how to compile it??
my matlab wont compile c++ file
or can you give me result from compiling vcapg2 file?

[Sebastien PARIS]

Do you have a functional webcam installed ?.

[adam akbar]

i got some error when run demo_detector_harr

Error in ==> demo_detector_haar at 33
aa = vcapg2(0,2);

how to solve this?

[Sebastien PARIS]

I don't undestand your question. Please run the four demo files...

[adam akbar]

i already do this step

>> mex -setup
Please choose your compiler for building external interface (MEX) files:

Would you like mex to locate installed compilers [y]/n? y

Select a compiler:
[1] Lcc-win32 C 2.4.1 in C:\PROGRA~1\MATLAB\R2009b\sys\lcc

[0] None

Compiler: 1

Please verify your choices:

Compiler: Lcc-win32 C 2.4.1
Location: C:\PROGRA~1\MATLAB\R2009b\sys\lcc

Are these correct [y]/n? y

Trying to update options file: C:\Documents and Settings\Adam\Application Data\MathWorks\MATLAB\R2009b\mexopts.bat
From template: C:\PROGRA~1\MATLAB\R2009b\bin\win32\mexopts\lccopts.bat

Done . . .

**************************************************************************
Warning: The MATLAB C and Fortran API has changed to support MATLAB
variables with more than 2^32-1 elements. In the near future
you will be required to update your code to utilize the new
API. You can find more information about this at:
http://www.mathworks.com/support/solutions/data/1-5C27B9.html?solution=1-5C27B9
Building with the -largeArrayDims option enables the new API.
**************************************************************************

>> mexme_fdt
compiling chlbp
compiling chlbp_adaboost_binary_model_cascade
compiling chlbp_adaboost_binary_predict_cascade
compiling chlbp_gentleboost_binary_model_cascade
compiling chlbp_gentleboost_binary_predict_cascade
compiling detector_haar
compiling detector_mblbp
compiling eval_chlbp
compiling eval_haar
compiling eval_haar_subwindow
compiling eval_mblbp
compiling haar
compiling haar_ada_weaklearner
compiling haar_adaboost_binary_model_cascade
compiling haar_adaboost_binary_predict_cascade
compiling haar_featlist
compiling haar_gentle_weaklearner
compiling haar_gentleboost_binary_model_cascade
compiling haar_gentleboost_binary_predict_cascade
compiling haar_scale
compiling imresize
compiling int8tosparse
compiling mblbp
compiling mblbp_ada_weaklearner
compiling mblbp_adaboost_binary_model_cascade
compiling mblbp_adaboost_binary_predict_cascade
compiling mblbp_featlist
compiling mblbp_gentle_weaklearner
compiling mblbp_gentleboost_binary_model_cascade
compiling mblbp_gentleboost_binary_predict_cascade
compiling rgb2gray
compiling fast_haar_ada_weaklearner
compiling fast_haar_adaboost_binary_model_cascade

but fdtool is not appear in toolbox please help.
thank you

[Qiu dw]

Thank you for sharing code.

[Kai Tang]

To Bluebird,

you can also test the LBP in matlab from MVG.
http://www.ee.oulu.fi/mvg/page/lbp_matlab

By the way, I have done the eye, even gaze detection based on paris' excellent work. Thanks a lot!!!

[Sebastien PARIS]

Please configure your compiler by "mex -setup"

[Tobin]

I've already tried, I get an error: Error using mexme_fdt>getmeopts at 208, getmeopts [Bruno]:Cannot open comopts.bat file

[Sebastien PARIS]

Simply run "mexme_fdt;" witout any options structure...

[Tobin]

Hi,
I'd like to disscus fd toolbox. I've got matlab 2010.
mexme_fdt.m ends with error: Could not find specified options file
'mexopts_intel10.bat'. Even if i have it in matlab path. Could anyone help me please?
Thanks, Tobias

[Samuel]

Thank you for this software. The haar functions in .c are fast and very useful.

[Sebastien PARIS]

Hello,

Probably to many values to output ..... You can modify chlbp.c in order to output theses outputs.....

[Louis]

Awesome toolbox. I have a question regarding the CHLBP, does it happen that you have the one that has the LBP values output (i.e. not the histogram bins output).

What i am looking for is you give an image and get the the LBP(R,N) for the pixels; it should a step before CHLBP which calculates the distibution of the LBP pixels

Thanks

[HMN]

 

[Sebastien PARIS]

Hello,

You installed the toolbox in 'C:\Program Files\MATLAB\R2009a\toolbox\fdtool' ? strange location, anyway ..
Which compiler didyou use ?. did you run mexme_fdt from the right dit ?i.e. 'C:\Program Files\MATLAB\R2009a\toolbox\fdtool' ?

like :

cd 'C:\Program Files\MATLAB\R2009a\toolbox\fdtool'
mexme_fdt

[Iqbal Nouyed]

Hi Sebastien,

I am having trouble in compiling the toolbox. I have tried the already given instruction as follows:

path('C:\Program Files\MATLAB\R2009a\toolbox\fdtool', path);
mex -setup

but after running the mexme_fdt i'm getting the following error:

compiling chlbp

C:\PROGRA~1\MATLAB\R2009A\BIN\MEX.PL: Error: 'chlbp.c' not found.

??? Error using ==> mex at 218
Unable to complete successfully.

Error in ==> mexme_fdt at 148
eval(['mex ' str])

Could you please help me?

Thanks.

[Sebastien PARIS]

mex -setup

[rajendra gharase]

hii.. i am weak in matlab.
can u tel me..
how to do the mex set up?
thanks...

[Kai Tang]

 

[Kai Tang]

To jethro tan,
Paris has done an excellent work on face detection. Maybe he is busy for other things, but he has figured out this question as
[model cascade]=multi_exit_asymetric_boosting(Xpos, negatives_path, options)

[jethro tan]

I trained my own model file "model.mat" by your file "fast_haar_adaboost_binary_model_cascade.c". But a cascade matrix must be created properly for detection module,such as [3 4 2 1;0 0 0 0]. How do I construct a cascade matrix properly?

[Mohammad Ghahramani]

It is a very useful set of codes but would you please change the imresize to sth else.

I added the path to fdtool folder but MATLAB mixes up with two imresize functions at a time and returns an empty matrix if you just use imresize in a code. The same error reported here: http://www.mathworks.co.uk/matlabcentral/newsreader/view_thread/265938

[Fumin Shen]

I am sorry I made such a stupid mistake!
Thanks for your time!

[Sebastien PARIS]

did you did a mex -setup before ?

[Fumin Shen]

This is an excellent work!!
But there's a problem when i run "mexme_fdt.m" to compile *.c files.
??? Error using ==> mexme_fdt>getmexopts at 208
getmexopts [Bruno]: cannot open comopts.bat file

And i failed to find the file "comopts.bat" or 'mexopts.bat' in the files downloaded.
Could any one tell me why?

[Sebastien PARIS]

Add fdtool dir in your path and run "mexme_fdt.m" to compile *.c files

[Nessrine]

i 'm not soo good in matlab implementation but i would like to test this code. can any one help me specialy i have no idea to use .c file with matlab for me this is the first matter.
thank an advance

[Sebastien PARIS]

Thanks you Soeren for these indications and comments.
Please if you have better models than those already included, I'll include them in a future release.

Sébastien

[Soeren Sproessig]

to build vcagp2.cpp (used in demo_detector_haar as camera acquisition provider) there are some other tricks
1. create a dummy dxtrans.h which seems to be missing in the SDK:
c:\Program Files\Microsoft SDKs\Windows\v6.0A\Include\dxtrans.h
#define __IDxtCompositor_INTERFACE_DEFINED__
#define __IDxtAlphaSetter_INTERFACE_DEFINED__
#define __IDxtJpeg_INTERFACE_DEFINED__
#define __IDxtKey_INTERFACE_DEFINED__

see: sample from http://www.riseoftheants.com/mmx/faq.htm#mjpegsample

2. some changes in code to be able to compile with VC08 and the current SDK in Win64:

82d81
<
629,630c628
< // pbmi->hInst = (HINSTANCE)GetWindowLong(GetFocus(),GWL_HINSTANCE);
< pbmi->hInst = (HINSTANCE)GetWindowLongPtr(GetFocus(), GWLP_HINSTANCE);
---
> pbmi->hInst = (HINSTANCE)GetWindowLong(GetFocus(),GWL_HINSTANCE);
958c956
< wc.lpfnWndProc =(WNDPROC) WndProc;
---
> wc.lpfnWndProc = WndProc;

[Soeren Sproessig]

in addition to my prior post:
to build vcapg2.cpp you need to have a current Microsoft DirectX SDK installed...

[Soeren Sproessig]

excellent work!

to use it with matlab R2009b x64 on win7pro64 I had to change mexme_fdt.m a little bit:

44c44,46
< lib = ['"' , fullfile(matlabroot , sprintf('extern\\lib\\%s\\microsoft\\libmwblas.lib',computer('arch'))) , '"'];
---
>
> lib = ['"' , fullfile(matlabroot , 'extern\lib\win32\microsoft\libmwblas.lib') , '"'];
>
135c137

[Soeren Sproessig]

 

[Bodla Rakesh]