This is C++ version of FairMOT working in Windows, requiring OpenCV and Eigen.
The tracking part has been ported from Python to C++.
The detection and feature extraction parts are still pending. The TensorRT version can refer to tensorRTIntegrate.
Currently, the converted TensorRT engine file can successully detect people, but the ReID feature extraction is different from the Python code issue. Suggestions are welcome.
How to Run
- extract people detection result in FairMOT by adding the following code in line 265 in "/src/lib/tracker/multitracker.py"
f = open('tmp_data/%03d.size'%self.frame_id, 'wt')
f.write('%d %d %d %d'%(dets.shape[0], dets.shape[1], id_feature.shape[0], id_feature.shape[1]))
f.close()
dets.tofile('tmp_data/%03d.det'%self.frame_id)
id_feature.tofile('tmp_data/%03d.feature'%self.frame_id)
- update the "list.txt" in this repository to locate the detection result and feature extraction result
TODO
add TensorRT for object detection and feature extraction
Acknowledgement
- Kalman Filter is borrowed from DeepSort, [deep_sort]https://github.com/apennisi/deep_sort
- [FairMOT]（https://github.com/ifzhang/FairMOT）