apennisi / fast_face_detector Goto Github PK

Thanks for the good job, I learn the ACF detector recently, and use to face detection.I try your FFD ,but the result boxes is erratic ,a example :

What is the format of TXT file in posGt folder?Does each image correspond to a TXT file?Or all images have only one TXT file, in the format "5002723411.jpg" x, Y, W h

Hi @apennisi,
Thanks for you great job.
Hope you can see this issue.

I have two question about this project.

1. the integral image, this is your new channel. 
    I found it is not a region cumsum of grey image, it is just the cumsum of the whole image. 
     I think this value depends on the face location in the whole image.
     It is NOT  location shift invariant. 
     Do you know why this channel let the system better?

2. I found you have the four pre-trained model.
    big_data/5x5, big_data/20x20, small_data/5x5, small_data/20x20, 
    5x5 is for 80x80 face size. 20x20 is for 100x100 face size. right?
    but what is the difference of big_data and small_data?

Thanks

When trying to build on Mac (High Sierra) using Clang, I get the following error: error: value of type
'int attribute((ext_vector_type(4)))' (vector of 4 'int' values) is
not contextually convertible to 'bool'
_m = MIN( RCPSQRT(_M2[y]), SET(1e10f) );

Any tips on how to get past this?

Thanks!

I tried to train a hand detector using your matlab code and saved the model into xml file using the savetoxml.m. When I run the c++ code using the xml file, I got the following error

terminate called after throwing an instance of 'boost::exception_detail::clone_impl<boost::exception_detail::error_info_injector<boost::property_tree::ptree_bad_data> >'
  what():  conversion of data to type "i" failed

It seems like the program cannot read parameter "type" of nms in the xml file. After a little checking, when I used savetoxml.m file, the parameter for nms became as below:

                <nms>                                                                                                                                                                                  
                        <type>max</type>
                        <overlap>0.650000</overlap>
                        <ovrdnm>min</ovrdnm>
                </nms>

However the parameter in your pretrained face_detector.xml, it is

                <nms>
                        <type>1</type>                                                                                                                                                                 
                        <overlap>0.650000</overlap>
                        <ovrdnm>1</ovrdnm>
                </nms>

Moreover, the classifer.cpp that parses all over the parameter of xml file, reads nms_type as int, not as string like "max".

    const int &nms_type = pt.get<int>("classifier.opts.nms.type");

I guessed that might be the problem why the c++ program cannot read my trained xml model, so I changed the line "max" into "1 as in your pretrained model but it is still not working.
Do you have any clues about this situation?

I use the savetoxml.m to get ACFcal.xml,but I meet the problem,

but other xml file is RUN OK,
the problem file is ,
acf.zip
how solve the problem?

Thanks in advance. 🤣

Hi , I run the code , but the result is not very good , have you tested , the result good or not?0.0

Have you tested on on FDDB so see how your improvement works compared to original?