samuelmurray / tracking-by-detection

Multiple object tracking (MOT) algorithm implemented in C++

samuelmurray

Hi! I'm a software developer located in Stockholm, Sweden.

tracking-by-detection's Issues

can this project apply on the mobile?

Running the demo

Hi,

I could set the path with the pkg-config for the dependencies and also did a make compile in the cpp/ folder.

I am not sure where the executable is built and how to run a demo with the project (is there a demo?). Or is this just a library and I have to link it to my project?

Thanks!

Is there any threshold I could adjust with my own data?

Hi there,
Thank you for your great contribution. I run the detectAndtrack code, the result is not bad. But there still exists obviously missing tracking ID between Neighboring frames. So Is there any threshold I could adjust with my own data.
Many thanks.

Compile errors with g++7

I get the following error when attempting to compile with g++7 or 6:

/usr/include/c++/6/cstdlib:75:25: fatal error: stdlib.h: No such file or directory
 #include_next <stdlib.h>

And this one when compiling with g++5

/usr/local/include/dlib/optimization/optimization_line_search.h:215:26: error: ‘__builtin_fpclassify’ is not a member of ‘std’
         if (temp == 0 || std::fpclassify(temp) == FP_SUBNORMAL)

Which version of g++ did you use to compile?

make this project more powerfull?

hi,professor:
your code very good! as we know, SORT is very fast, but its disadvantage is SwitchID too much, so , are you sitll work in MOT? can you make some improve on your project with some new papers method, follow tracking-by-detection paradigm,
thanks a lot!

Track demo produces weird values

I ran track demo with ./trackApp.out -s seqmap.txt -m det -f mot using the det.txt provided in the MOT16 data and some of the values in the output (e.g. MOT16-02) are off, like -nan values or pixel values bigger than the image size, as shown below.
216,84,-79658.6,-nan,0,-nan,1,-1,-1,-1,1

I am using dlib-19.8 with DLIB_ISO_CPP_ONLY set to ON, openCV-3.4

Please add some references to papers

I came here from the MOT challenge website interested in your tracker. Could you please add some references to publications in the readme so I can learn more about how it works?

cost = dlib::join_cols(cost, dlib::zeros_matrix<int>(cost.nc() - cost.nr(), 1));

when run demo, the bug is:
DLIB_ASSERT(a.nc() == b.nc() || a.size() == 0 || b.size() == 0,
"\tconst matrix_exp join_cols(const matrix_exp& a, const matrix_exp& b)"
<< "\n\tYou can only use join_cols() if both matrices have the same number of columns"
<< "\n\ta.nr(): " << a.nr()
<< "\n\tb.nr(): " << b.nr()
<< "\n\ta.nc(): " << a.nc()
<< "\n\tb.nc(): " << b.nc()
);
have you ever seen this bug?

Can we use without GPU / CUDA dependency ?

Can we use without GPU / CUDA dependency ?

how?

Python equivalent

Would you provide me the python Keras equivalent of this code. I really do not like C++ and caffe.

make compile success but make detect or make track wrong??

make compile [use_caffe=true] [force_rebuild=true]

(compile) CXX src/detector/SSDDetector.cpp
(compile) CXX src/detector/RandomDetector.cpp
(compile) CXX src/detector/Detector.cpp
(compile) CXX src/examples/DetectDemo.cpp
(compile) CXX src/examples/TrackDemo.cpp
(compile) CXX src/examples/DetectAndTrackDemo.cpp
(compile) CXX src/util/Detection.cpp
(compile) CXX src/util/BoundingBox.cpp
(compile) CXX src/util/DetectionFileParser.cpp
(compile) CXX src/util/Tracking.cpp
(compile) CXX src/ImageTracker.cpp
(compile) CXX src/tracker/RandomTracker.cpp
(compile) CXX src/tracker/Tracker.cpp
(compile) CXX src/tracker/predictor/Predictor.cpp
(compile) CXX src/tracker/predictor/kalman/KalmanPredictor.cpp
(compile) CXX src/tracker/predictor/particle/Particle.cpp
(compile) CXX src/tracker/predictor/particle/ParticleFilter.cpp
(compile) CXX src/tracker/predictor/particle/ParticlePredictor.cpp
(compile) CXX src/tracker/predictor/StationaryPredictor.cpp
(compile) CXX src/tracker/PAOT.cpp
src/tracker/PAOT.cpp: In static member function ‘static PAOT::Association PAOT::associateDetectionsToPredictors(const std::vector&, const std::vector<std::shared_ptr >&, double (*)(const BoundingBox&, const BoundingBox&), double)’:
src/tracker/PAOT.cpp:71:27: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (int i = 0; i < detections.size(); ++i)
^
src/tracker/PAOT.cpp:95:23: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
for (int d = 0; d < assignment.size(); ++d) {
^
src/tracker/PAOT.cpp:97:19: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
if (d < detections.size()) {
^
src/tracker/PAOT.cpp:100:31: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
if (assignment[d] < predictors.size()) {
^
(compile) CXX src/tracker/Affinity.cpp

but when i make detect:

make detect
(link) CXX detectApp.out
/usr/bin/ld: build/tracker/predictor/kalman/KalmanPredictor.o: undefined reference to symbol 'dgesvd_'
//usr/lib/liblapack.so.3: error adding symbols: DSO missing from command line
collect2: error: ld returned 1 exit status
makefile:84: recipe for target 'detectApp.out' failed
make: *** [detectApp.out] Error 1

What's the Sequence maps?

What's the Sequence maps?
thanks a lot!

detectAndTrack usage

I'm attempting to use the detectAndTrack demo, but I'm having issues with the configuration files.
For the models part, I've put the caffe ssd model under tracking-by-detection/models, and my model_config.txt file looks like this.

<model-type>
deploy.prototxt
VGG_VOC0712_SSD_300x300_iter_120000.caffemodel
<r_mean,g_mean,b_mean>

For the seqmap.txt file, I don't know what to do since I don't have any detections to point to, I thought the point of detectAndTrack was to have SSD do the detections.

What should the configuration look like?

if it helps my images are stored as data/myimages/images/xxxxx.jpg

Any help would be very much appreciated.

run trackApp.out error

Hi,
I use
./trackApp.out -s seqmap.txt -m det -f mot
then, there is an error :Segmentation fault (core dumped)
before that, it prints Sequence: images, that means the problem is on auto durationFrameCount = track(sequencePathString, modelType, detectionFormat, frameInterval, doOverwriteResult);
I do not what the problem is, can you help me?
Thanks a lot!

Compile demo error

Hi, there!
I have set the path with .pc file in config/ folder. After i make compile and make detect, i got the executable file detetcApp.out, but then i got an error when ./detectApp.out like the following:

./detectApp.out: symbol lookup error: /opt/intel/mkl/lib/intel64/libmkl_intel_lp64.so: undefined symbol: mkl_lapack_dgetrfnpi

which have been perplexing me for a long time. Actually, i cannot find the undefined symbol -- mkl_lapack_dgetrfnpi in my PC, and i am not sure whether it's the MKL error or some other code error or environment wrong.

My config is as follows:
Ubuntu 16.04
dlib -- 19.10.0
MKL -- Intel parallel_studio_xe_2017_update4
boost -- 1.67.0
caffe -- which is ok
cuda -- 8.0
opencv -- 3.2.0 with contrib

Thanks a lot!

CMakeLists.txt ?

The installation guide show that you are using cmake to maintain the project. But where is the CMakeLists.txt file ?