bochinski / iou-tracker Goto Github PK

hi nice work!
I want to reproduce the mot17 track results. the result of mot17 FRCNN in README.txt is the average result of all FRCNN? because there are many FRCNN gt and det.

Question 1: How this the algorithm behave in a scenario with two tracks of two persons, where the detector fails to detect one person in one frame but manage to detect the person in the next frame. Will it recover the track or simple assign it a new trackid?

Question 2: Any plans to make an online version of this algorithm (like SORT) ?

Hi, everyone! I want to run this code and I create a vitrual environment. But I can't install the lapsolver with the pip. Have any of you been in this situation? Thanks a lot.

Hi,
First of all, great work. Thanks for making it available to the public.

It seems that MOT17 datasets have changed since the code for load_mot() in util.py was written. Now the sequences from detectors FRCNN and SDP only provide det.txt files with 7 columns (no information for classes) and load_mot() fails here.

I fixed it by calling load_mot() in mot17.py with flag with_classes=False. Just wanted to let you know to check that the fix is ok and to suggest to update these demos accordingly, if you reckon appropiate.

Hello,

First of all, thank you for posting the V-IOU code :)
I would like to reproduce the results that you've obtained on CVPR19 MOT Data but I can't seem to find where to download it from.

The links https://motchallenge.net/data/CVPR_2019_Tracking_Challenge/ and https://motchallenge.net/data/CVPR_2019_Detection_Challenge/ are not working.

Do you happen to have the data? If yes, could you please drop me a link to download it?

Thanks a lot,
Nicolae

thanks

Hi,
Can u provide the detections for vis-drone dataset?

Hi,

Have you added a script to visualize the results ?

we need install KCF tracker before run VIOU?

Hi,
I am unable to produce the results for visdrone-mot-val dataset based upon v-iou as reported in the repository. Procedures , I am following, are:

1. Run the "./visdrone-mot.sh" command and it produces separate csv(equivalent to number of input sequences) files based upon v-iou tracker. I dont know where to give these cvs files as input for the producing the mentioned results. Can any one guide please? Visdrone toolkit has no option to give these files as input. DETRAC toolkit is not available to download.

2. Also, I downloaded the Visdrone-mot-toolkit for the evaluation. I placed the v-iou tracker files in the ./tracker folder of the toolkit. I run the "run_tracker.m" file from the matlab to produce the results but following error is produced every time. Can anyone able to resolve this issue in producing the repository results for visdrone dataset. I shall be thankful.
   Kind Regards,

tracking the sequence 1/6...

error while calling the python tracking module:

Error using viou_tracker>track_viou_matlab_wrapper (line 220)
Python Error: ValueError: cannot reshape array of size 138720 into shape (7,newaxis)

Error in runTrackerAllClass (line 32)
[resClass, runTime] = run_tracker(sequence, curdetections);

Error in evalMOT (line 29)
runTrackerAllClass(isSeqDisplay, isNMS, detPath, resPath, seqPath, evalClassSet, threSet, nmsThre, trackerName);
Error using runTrackerAllClass (line 44)
error in running the tracker!

Error in evalMOT (line 29)
runTrackerAllClass(isSeqDisplay, isNMS, detPath, resPath, seqPath, evalClassSet, threSet, nmsThre,
trackerName);

hi, thanks for your work! Could you please tell me how to get the test results like pr-mota on UA-DETRAC? Now the website of UA-DETRAC is could not to submit results.

Hello,

I was trying to reproduce your results for the IOU-tracker on DETRAC-Train using detector Mask-RCNN. Although I have used the provided detection results and a new clone of your tracker, I keep getting results with MOTA of about 29.4 as opposed to the reported value of 34.36.
Tracker iou-tracker-master + Detector frcnn by DETRAC-MOT Evaluation:
PR-Rcll PR-Prcn PR-FAR|PR-MT PR-PT PR-ML|PR-FP PR-FN PR-IDs PR-FM| PR-MOTA PR-MOTP PR-MOTAL
34.8 41.0 0.36|28.8% 11.7% 26.7%|30139.2 187059.4 1006.6 995.2| 29.4 38.4 29.6
Would you mind looking at my attached logs below and see if I did anything wrong? Thank you very much for your help!
log.txt

Hi, what is the format of the input to the tracker?
is it:
frame bbleft bbtop bbwidth bbheight confidence
?

I am not using any of the standard MOT detections.
thanks.

I'm new to tracker. No knowledge of the data set used in the evaluation.
Why do you need to use the training data set when evaluating?
and test data not have gt.txt, how using test data to eval tracker?

Hello,

Is the v-iou implementation available somewhere ? (based on paper Extending IOU Based Multi-Object Tracking by Visual Information)

I'm doing something close to V-IOU idea in a node.js implementation: https://github.com/tdurand/node-moving-things-tracker but in a much less "academic" way.

There is not much details about the "prediction" method you use for the visual tracking.. On my project I'm using velocity / acceleration vectors but do not change the bounding box width / height

From you paper, I'm gonna try this idea of searching "back-wards through the last ttl frames for each new track" , this is something I'm not doing currently

Thanks !

Thibault

Hi,
Thanks to write this repo for objects tracking using both IoU and V-IoU. Can you please solve following queries:

1. I am unable to get the visdrone detections file to test the mot tracking.
2. Also, which arguments to use on command line to generate the mot-visdrone results?
   Thanks in advance for your kind response.
   Kind Regards,

please help , where i need to modify for the custom data loading and how to visualize the results ??

TypeError: integer argument expected, got float

detections path to csv file containing the detections or numpy array containing them.
Loads detections stored in a mot-challenge like formatted CSV or numpy array containing them.

numpy array（fieldNames = ['frame','id','x','y','w','h','score']）, What does ‘id’ mean？

Thank you for your reply

Hi!
I would like to use DETRAC-toolkit, but it's no longer available (on https://detrac-db.rit.albany.edu/download).
Could you please tell me where to download it? Or other alternatives how to test your tracker on DETRAC dataset.
Thank you a lot

I quite often loose some tracks because of missing detections in a single or two frames.

I added a "minimum age" criteria to the finishing track and save the tracks for one or two frames.

This will propably just work with high framerates of about 20fps and more.

adding a track['age'] = 0 right before the updated_tracks.append(track) in line 45 for matched tracks and changing line 52 if track['max_confidence'] >= sigma_h and (track['frames']) >= t_min: to

if track['age'] < 2:
  track['bboxes'].append(track['bboxes'][-1])
  track['age'] += 1
  updated_tracks.append(track)
  track['max_confidence'] >= sigma_h and (track['frames']) >= (t_min + track['age']):
...

will do it.

Somebody could prove that? Am I missing something?