Hi ,

    RT, i found config/mot/qdtrack/qdtrack_faster-rcnn_r50_fpn_4e_dancetrack.py in mmtracking framework.

When OC_SORT can merge into mmtracking framework ?

Hello @noahcao,

Thanks for sharing the code. I have a question regard to speed of OC_Sort.
Im using YoloV5 as a object detector for both OC_Sort and ByteTracker tracking algorithm with the same setting. However, all the process for one frame (Detection and tracking) took almost ~14ms for ByteTracker and ~45ms for OC_Sort. I would like to ask you how can I speed up the OC_Sort (Based on your paper, OC_Sort should be faster than ByteTrack)?

Thanks!

@noahcao Hi! https://github.com/DanceTrack/DanceTrack/blob/18b60ea52ce03a7ed1710930663d810b461f460e/ByteTrack/exps/example/dancetrack/yolox_x.py#L25
Is it enough to train only 8 epochs on the DanceTrack dataset? Or is it a spelling mistake?Thanks for your reply!

Hello, thank you very much for your excellent work. I got TrackEval to evaluate mot_challenge. But TrackEval just have python files and samples for train sets, do not have files or samples for test sets. Would i get your python project for evaluating mot_challenge?

First of all, thank you for providing a valuable contribution to the Computer Vision community. It was interesting to read your paper and also thank you for providing your implementation. I have a few doubts to clarify,

1. From your paper in Table 7, you show that OOS strategy does not show much improvement for DanceTrack and a significant boost in HOTA for MOT17. Could you provide a reasoning for this? I am trying to understand the impact of the three strategies you provided (OOS, OCM, OCR).
2. I was able to visually understand the impacts of OCM and OCR by removing the components and evaluating on DanceTrack. I could not understand the same for OOS. I did not visually see any difference when using OOS or not. I understand intuitively (theoretically) its impact. I was wondering do you have example videos where you visually saw the impact of OOS.
3. What are your thoughts when we use a less powerful detector? I tried using mobilenet+ssd (tflite version) from TFHub and the performance was not as expected. It was also the case for SORT. What are your suggestions in cases where the detector is not very powerful?

when i run the demo given in README, i can not find the pretrained model "bytetrack_dance_model.pth.tar", and i used "ocsort_dancetrack.pth.tar" given in model zoo, python tools/demo_track.py --demo_type video -f exps/example/mot/yolox_dancetrack_test.py -c pretrained/bytetrack_dance_model.pth.tar --path videos/dance_demo.mp4 --fp16 --fuse --save_result --out_path demo_out.mp4, but the result has no any bounding boxes?

@noahcao Hi!
Setting self.data_num_workers=4, but it doesn't seem to work when running run_ocsort_dance.py (evaluate on dancetrack_val, large datasets), and it is very fast when running run_ocsort.py (evaluate on mot17_val_half, small datasets). Why is this? How can I fix this?

Dear author,
I saw the data in exps/permatrack_kitti_test contains frame_ids, object_ids, object positions, results of detection and the object_ids are already connected between frames.
Can you explain why there are object_ids here? Because I think the output of Permatrack-detector just contains object positions and accuracy of the detection and there no object_ids,
BRs,
cuonga1cvp

Hi,

I needed to run OC-SORT from MATLAB with a matrix of detections and get a matrix of tracks back. I wrote a simple wrapper function based on your code and it seems to be working. I needed to make a minor change to the init function of the OCSORT class and I changed the result writing portion as well. I seem to be getting the same results as your original script, so I think it is working fine. Would you like me to share the code with you?

Have you tried using your new Kalman filter in DeepSort?

all of datasets have a gt category of visability,

Is visibility essential? I'm curious to see if it's written by judging the number by seeing it with your own eyes, or if there is a formula.

exam of gt label
frame,id, x1, y1, w, h, -, -, *visibility
134, 1, 912, 484, 97, 109, 0, 7, 1
135, 1, 912, 484, 97, 109, 0, 7, 1
136, 1, 912, 484, 97, 109, 0, 7, 0.93878

are always the same number so it's not a problem, but the visibility may not be 1 depending on how visible it is, so I'm curious how to deal with it.
I am also wondering if this item is required. If anyone knows, please reply.

I 'm confused about finding OOS code for virtual trajectory generation, can you offer some help?

Are you using the bytetrack detector directly to produce the test results? The test results for sensory validation are not quite correct.Have you changed any parameter values?

Hi, your work is wonderful! I'm trying to use it in my project. But in my project, I need to track 7 kinds of viechles in traffic carema videos. But it seems that the update method of OCSort class doesn't care about the class. So is there any way that I can get the tracking result with classes or how I can change your code to manage that. Thanks!

The code is not available now. Look forward to access the code as soon as possible.
Thanks to the author.

Hi, thanks for the interesting points in your paper and the code you provided.

Could you help me to find where is the linear interpolation post-processing code described in Table 10 of the paper located?

Thanks for the amazing work again!

After replacing the SORT kalman filter in ocsort.py with the JDE kalman filter, I got higher HOTA and faster speed, which may indicates that ocsort with SORT settings can be improved.

So, do you plan to provide a version of ocsort with BYTE?

作者您好，很感谢您开源这么优异的跟踪器。在阅读您的代码时，我注意到您在处理angle_diff_cost时，将其与scores相乘。这个操作我不太理解，如果您方便的话，请您给讲解一下~
另外，您在处理IoU距离时，将其resize到[0,1]。diff_angle值域是[-0.5,0.5]，由于乘系数vdc_weight 0.2，所以值域变为[-0.1, 0.1]。那么，最终两个框之间的距离在区间[-0.1, 1.1]，请问我这样理解对吗？期待您的回复，感谢！

when I tried [2. Docker build]
docker build -t ocsort:latest .
The following problem occurred.

Processing triggers for sgml-base (1.29.1) ...

WARNING: apt does not have a stable CLI interface. Use with caution in scripts.

Removing intermediate container 67ee1859253d
---> 686356aae519
Step 6/11 : RUN git clone https://github.com/noahcao/OC_SORT && cd OC_SORT && git checkout 3434c5e8bc6a5ae8ad530528ba8d9a431967f237 && mkdir -p YOLOX_outputs/yolox_x_mix_det/track_vis && sed -i 's/torch>=1.7/torch==1.9.1+cu111/g' requirements.txt && sed -i 's/torchvision==0.10.0/torchvision==0.10.1+cu111/g' requirements.txt && sed -i "s/'cuda'/0/g" tools/demo_track.py && pip3 install pip --upgrade && pip3 install -r requirements.txt -f https://download.pytorch.org/whl/torch_stable.html && python3 setup.py develop && pip3 install cython && pip3 install 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI' && pip3 install cython_bbox gdown && ldconfig && pip cache purge
---> Running in 66120b5d54d0

Cloning into 'OC_SORT'...
fatal: reference is not a tree: 3434c5e8bc6a5ae8ad530528ba8d9a431967f237

The command '/bin/sh -c git clone https://github.com/noahcao/OC_SORT && cd OC_SORT && git checkout 3434c5e8bc6a5ae8ad530528ba8d9a431967f237 && mkdir -p YOLOX_outputs/yolox_x_mix_det/track_vis && sed -i 's/torch>=1.7/torch==1.9.1+cu111/g' requirements.txt && sed -i 's/torchvision==0.10.0/torchvision==0.10.1+cu111/g' requirements.txt && sed -i "s/'cuda'/0/g" tools/demo_track.py && pip3 install pip --upgrade && pip3 install -r requirements.txt -f https://download.pytorch.org/whl/torch_stable.html && python3 setup.py develop && pip3 install cython && pip3 install 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI' && pip3 install cython_bbox gdown && ldconfig && pip cache purge' returned a non-zero code: 128

If you have encountered a similar problem or know a solution, please comment.

Hi,

I've been trying to use dti() in interpolation.py with the results I get from demo_track.py, but I think this function has a problem - on line 78 of interpolation.py, I believe n_frame = tracklet.shape[0] will always return 1. Therefore I don't think the following code is ever run. And when I have tried to change this, e.g. n_frame = int(tracklet[:,0]) , I run into different problems, e.g. I think the score threshold doesn't work because in demo_track.py the score is always written as 1.0 -
results.append(f"{frame_id},{tid},{tlwh[0]:.2f},{tlwh[1]:.2f},{tlwh[2]:.2f},{tlwh[3]:.2f},1.0,-1,-1,-1\n")) .

If you could tell me how to usethis code with the text files output by demo_track I would be very grateful.

Thanks

Hi, really nice work here. Excited to potentially test this software out. I apologize for posting this as an issue but I wanted to get your opinion.

I am currently looking for a tracking algorithm to track bacterial cells. I know I will have to figure out the cell divisions but do you think OC-SORT could track bacterial cells? I provded a movie below as an example. The DanceTrack dataset is the closest thing I've found that is relevent to cells growing (uniform appearance and nonlinear motion) and OC-SORT was able to get SOTA results.

I've read through a fair amount of your code but it will still take a fair amount of work for me to adapt this to tracking cells (as I am currently learning about the kalman filter) so I just wanted to get your opinion before I spend more time on it.

Edit: I have a model that can segment the cells so I will not need YOLO to detect the cells. I am primarily interested in just tracking the cells.

Thanks in advance.

I want to try and run OC-SORT with a few different detectors (YoloV5, F-RCNN, CNN) just to experiment and see how well the detectors perform with OC-SORT.

@noahcao Hi,Thanks for open-sourcing such a great work.
When I run the gp_interpolation.py file, the compiler warns that these places have been modified,

After a comparison with Bytetrack, I saw the powerful effect of OC-SORT. I would like to ask when the c++ deploy version of OC-SORT will be updated. I am really looking forward to it!Thanks a lot~

https://github.com/CaptainEven/MCMOT-ByteTrack

Thank you for your great work.
when I evaluate the dancetrack verification, I get an import package error, what is the reason?

When I run test on MOT17, the error occurs below.

cannot import name 'MOTDataset'

And I search in the whole project, and I have not found the definition of "MOTDataset".

How can I solve it.

Many thanks.

When I run python setup.py develop, a problem occurs.as:
g++: error: /media/peng/Data/Myproject/Chapter2-Track/code/run/OC_SORT-master/build/temp.linux-x86_64-3.6/media/peng/Data/Myproject/Chapter2-Track/code/run/OC_SORT-master/yolox/layers/csrc/vision.o: No such file or directory
g++: error: /media/peng/Data/Myproject/Chapter2-Track/code/run/OC_SORT-master/build/temp.linux-x86_64-3.6/media/peng/Data/Myproject/Chapter2-Track/code/run/OC_SORT-master/yolox/layers/csrc/cocoeval/cocoeval.o: No such file or directory
error: command 'g++' failed with exit status 1

I used gcc 5.4 version.
I haven't solved this problem.

Thank you very much for your excellent work, can you please provide the detection data for the KITTI tracking dataset！

Hey, great work!
Would a Deepstream tracker implementation be possible?

Dear author,
I had some problem when I ran OC-SORT on KITTI dataset. Maybe my steps were wrong. Could you please guide me how to run OC_SORT on KITTI dataset?
Thank you so much.
BRs,
cuonga1cvp.

could you provide tracking demo of MOT17 videos? If possible, could you provide the private detections. Thank you.

I've following the instruction and running the training code.
But got this error:

here is my running command:

python tools/train.py -f exps/drive/yolox_m_drive.py \
-d 1 \
-b 4 \
-c pretrained/yolox_m.pth

Could you please helping me fix it?

Thanks for sharing your work. I modified your run_ocsort_public script to run on the detection files from MOT20-train. Your code ran fine, but the tracking results seem very poor (combined HOTA = 10.7, combined MOTA = 3.4). In general I would like to use your code with a MOT-style det.txt file, but I am concerned that it is not currently working right.

In this zip file, I attached the modified script "ocsort_custom.py", pictures with the settings I used for OC-SORT and the evaluation results using the official eval code. I placed the folder "MOT20-train" under datasets and wrote to a file named "output."

I ran this from the root folder...
python ocsort_custom.py --raw_results_path datasets/MOT20-train --hp --dataset MOT20-train --out_path output

Archive.zip

Thank you for the contribution and the great work! I'm confused about some variables of the traker, can you explain some terms in the code please?

I want to implement on Mobile(Android/iOS), Deploying with NCNN is possible?
Thanks for sharing amazing MOT implementation.

When I tried python3 tools/demo_track.py video -f exps/example/mot/yolox_s_mix_det.py --trt --save_result,
The error is occured.

Traceback (most recent call last):
File "tools/demo_track.py", line 280, in
exp = get_exp(args.exp_file, args.name)
File "/workspace/OC_SORT/yolox/exp/build.py", line 47, in get_exp
assert (
AssertionError: plz provide exp file or exp name.

Forcibly entering exp name did not solve the problem.
ex)args.exp_file ='~~~~.pth'

If anyone knows a solution, please share.

fps

I got errors like these:

my environment like this:
pytorch==1.7.1 python==3.8 cuda==11.0

and i run this:
conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=11.0 -c pytorch

Hi, thanks for your great work. When trying to run your code, such as python run_ocsort_dance.py, it will report that the package cannot be imported. It seems that you forgot to upload the content under yolox/data/datasets. If possible, please upload this part of the code, Best Wishes!

@noahcao Hi!
when I run :
python3 tools/run_ocsort.py -f exps/example/mot/yolox_x_ablation.py -c pretrained/bytetrack_ablation.pth.tar -b 1 -d 1 --fp16 --fuse --expn MOT17

the error is :

File "tools/run_ocsort.py", line 32, in compare_dataframes
accs.append(mmp.utils.compare_to_groundtruth(gts[k], tsacc, 'iou', distth=0.5,vflag = vflag))
TypeError: compare_to_groundtruth() got an unexpected keyword argument 'vflag'

I successfully ran the code last week. But i failed today.
I strcutly follow the readme, and installed libglib2.0-dev.
I want to test the video you provided.

python3 tools/demo_track.py --demo_type video -f exps/example/mot/yolox_dancetrack_test.py -c pretrained/bytetrack_dance_model.pth.tar --path videos/dance_demo.mp4 --fp16 --fuse --save_result --out_path demo_out.mp4

And i got an error like this:

I also got this error last week, and change the path to solve it, and then i got other errors.
I do not want to modify this code.
Why is it happen?
What should i do to sovle it?

Thanks for the amazing work!

But I find something confusing when using oc-sort. At the beginning of tracking (like frame 2), the input variables of function speed_direction_batch() is detections and previous_obs.

But the previous_obs is [-1, -1, -1, -1, -1], which I don't think is normal.

Is it an issue, or I should just ignore it?

I have this problem while importing lab. Any idea why this happen? And how to fix this? Thank you

pip install lap
Collecting lap
Using cached lap-0.4.0.tar.gz (1.5 MB)
Preparing metadata (setup.py) ... done
Building wheels for collected packages: lap
Building wheel for lap (setup.py) ... error
error: subprocess-exited-with-error

× python setup.py bdist_wheel did not run successfully.
│ exit code: 1
╰─> [26 lines of output]
Partial import of lap during the build process.
Generating cython files
running bdist_wheel
running build
running config_cc
INFO: unifing config_cc, config, build_clib, build_ext, build commands --compiler options
running config_fc
INFO: unifing config_fc, config, build_clib, build_ext, build commands --fcompiler options
running build_src
INFO: build_src
INFO: building extension "lap.lapjv" sources
INFO: building data_files sources
INFO: build_src: building npy-pkg config files
running build_py
creating build
creating build\lib.win-amd64-3.10
creating build\lib.win-amd64-3.10\lap
copying lap\lapmod.py -> build\lib.win-amd64-3.10\lap
copying lap_init.py -> build\lib.win-amd64-3.10\lap
running build_ext
INFO: No module named 'numpy.distutils._msvccompiler' in numpy.distutils; trying from distutils
INFO: customize MSVCCompiler
INFO: customize MSVCCompiler using build_ext
INFO: CCompilerOpt.cc_test_flags[1029] : testing flags (/O2)
error: Microsoft Visual C++ 14.0 or greater is required. Get it with "Microsoft C++ Build Tools": https://visualstudio.microsoft.com/visual-cpp-build-tools/
INFO: CCompilerOpt.cache_flush[825] : write cache to path -> C:\Users\cuong\AppData\Local\Temp\pip-install-pmxzp02k\lap_7292aac9a6284f5f81a842387df85a51\build\temp.win-amd64-3.10\Release\ccompiler_opt_cache_ext.py
[end of output]

note: This error originates from a subprocess, and is likely not a problem with pip.
ERROR: Failed building wheel for lap
Running setup.py clean for lap
Failed to build lap
Installing collected packages: lap
Running setup.py install for lap ... error
error: subprocess-exited-with-error

× Running setup.py install for lap did not run successfully.
│ exit code: 1
╰─> [26 lines of output]
Partial import of lap during the build process.
Generating cython files
running install
running build
running config_cc
INFO: unifing config_cc, config, build_clib, build_ext, build commands --compiler options
running config_fc
INFO: unifing config_fc, config, build_clib, build_ext, build commands --fcompiler options
running build_src
INFO: build_src
INFO: building extension "lap.lapjv" sources
INFO: building data_files sources
INFO: build_src: building npy-pkg config files
running build_py
creating build
creating build\lib.win-amd64-3.10
creating build\lib.win-amd64-3.10\lap
copying lap\lapmod.py -> build\lib.win-amd64-3.10\lap
copying lap_init.py -> build\lib.win-amd64-3.10\lap
running build_ext
INFO: No module named 'numpy.distutils._msvccompiler' in numpy.distutils; trying from distutils
INFO: customize MSVCCompiler
INFO: customize MSVCCompiler using build_ext
INFO: CCompilerOpt.cc_test_flags[1029] : testing flags (/O2)
error: Microsoft Visual C++ 14.0 or greater is required. Get it with "Microsoft C++ Build Tools": https://visualstudio.microsoft.com/visual-cpp-build-tools/
INFO: CCompilerOpt.cache_flush[825] : write cache to path -> C:\Users\cuong\AppData\Local\Temp\pip-install-pmxzp02k\lap_7292aac9a6284f5f81a842387df85a51\build\temp.win-amd64-3.10\Release\ccompiler_opt_cache_ext.py
[end of output]

note: This error originates from a subprocess, and is likely not a problem with pip.
error: legacy-install-failure

× Encountered error while trying to install package.
╰─> lap

note: This is an issue with the package mentioned above, not pip.
hint: See above for output from the failure.

Hello,
I ran the evaluation code "python tools/run_ocsort_dance.py -f exps/example/mot/yolox_dancetrack_val.py -c pretrained/bytetrack_dance_model.pth.tar -b 1 -d 1 --fp16 --fuse --expn /output", and got the dancetrack0004.txt like this:

"1,3.0,772.2,309.5,261.9,767.8,-1,-1,-1,-1
1,2.0,969.3,414.5,299.7,579.2,-1,-1,-1,-1"
what are the means of these numbers, and how can i use them?