latest/trackers/

[giscus[bot]]

latest/trackers/

A set of easy-to-use utilities that will come in handy in any computer vision project.

https://supervision.roboflow.com/latest/trackers/

[FrancisDrakeII]

in "update_with_detections(detections)" example, "tracker = sv.ByteTrack()" "detections = byte_tracker.update_with_detections(detections)" should be "detections = tracker.update_with_detections(detections)"

[SkalskiP]

Thanks for letting us know! Fixed!

[s-madhavan001]

how to track in each person wait time in using supervision ?
i can detect in every single people with track id. what will do next. for track each id's idel-time(standing/waiting time)
in using supervision or ultralytics(yolov8)

[SkalskiP]

Hi @s-madhavan001 👋🏻 All you need is here: https://github.com/roboflow/supervision/tree/develop/examples/time_in_zone

[SkalskiP]

Not really sure what you mean. It is already tracking and showing IDs.

[s-madhavan001]

have any Tutorial videos ?

[SkalskiP]

Working on it ;) It should be out within the next 2 days.

[SkalskiP]

Hi @s-madhavan001, the video was released yesterday: https://youtu.be/hAWpsIuem10. I hope you'll find it helpful. I'm sorry it took a bit longer. It's a complicated topic. :/

[Shahid-TheMatrixMaker]

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
chex 0.1.86 requires numpy>=1.24.1, but you have numpy 1.23.0 which is incompatible.
pandas-stubs 2.0.3.230814 requires numpy>=1.25.0; python_version >= "3.9", but you have numpy 1.23.0 which is incompatible.
tensorflow 2.15.0 requires numpy<2.0.0,>=1.23.5, but you have numpy 1.23.0 which is incompatible.

i found this error while installing super-gradients

[aberg2323]

I am currently testing out the bytetrack code with my yolov8 object detection model. I am annotating my video with tracers. The problem I am encountering is that sometimes the tracers are working perfect, but then it seems the model "loses" the object for a little bit, and the tracer restarts. I have already tried changing some of the parameters of bytetrack and am seeing marginally improved results, but nothing that completely fixes the problem. Is this something that is solvable by changing the parameters of bytetrack or something that would be fixed by me training my object detection model again with more data?

[rolson24]

Hi @aberg2323!

Usually when ByteTrack is not performing as expected it is due to the object detection model have less than ideal performance. The tracker relies on objects being detected in nearly every frame, and also relies heavily on the confidence score from the detection model to determine if the detection is a valid object. I would try training your object detection model more and once it seems to have good performance on the task you want it to perform, then you can adjust the parameters of ByteTrack.

[aberg2323]

That is what I figured! Thank you for the help

[AshishreddyM26]

hey @SkalskiP

I am doing an analysis with object tracking, so far bot-sort performed well than Bytetrack, but I want to go with Bytetrack, as I aimed to work with videos at 30FPS.

Firstly, performed the detection and tracking using supervision. Got the results, but as I want to do lot of analysis in different parts of the frame, to do this I need to pass the detections separately by leaving out some of the labels, to achieve this, I tried the following way.

I performed detection and tracking separately (meaning

i. Initially, I performed detection on the video and stored the results in a variable 'custom_results'
ii. Then the detections of each frame (custom_results[i]) were given to the tracker manually, as expected the results were same by the both approaches.)

Is there a way that I can pass the labels of required area of interest to the tracker? (As I referred the documentation of supervision, I saw that the detections were taken as tensors.)

Can you give any leads on approaching this.

Thank you,
Ashish

[AshishreddyM26]

Thanks,
Figured it out, thanks for the 'supervision'

Ashish

[AdelShousha]

I'm tracking cars using an rtsp link, after a while a get errors such as:

[rtsp @ 0x5cf0efc9cf80] RTP: PT=60: bad cseq 1f2e expected=df9f
[hevc @ 0x5cf0ef87c200] The cu_qp_delta -31 is outside the valid range [-26, 25].
                            # cu_qp_delta 125
Frame resolution: 2560x1920

0: 480x640 (no detections), 4.1ms
Speed: 1.6ms preprocess, 4.1ms inference, 0.5ms postprocess per image at shape (1, 3, 480, 640)
[hevc @ 0x5cf0efb23140] Could not find ref with POC 30

[rtsp @ 0x5d018ba71180] Illegal temporal ID in RTP/HEVC packet

[hevc @ 0x5d018b974ec0] CABAC_MAX_BIN : 7

I'm using cv2.VideoCapture(RTSP) to capture the rtsp stream and I have these lines to check the continuity of the stream

while True:
    ret, frame = cap.read()
    if not ret:
        print("No frame read")
        break

so after a while from using the rtsp stream it breaks

I know this error might not be related to tracking exclusively but I'm not able to understand why is this problem happening is it something with the rtsp stream itself or my processing or the use of cv2.VideoCapture() or what exactly, can someone help?

[AdelShousha]

yes the stream is stable and I'm able to watch ot using VLC without any problems

[SkalskiP]

Have you tried using InferencePipeline? https://inference.roboflow.com/using_inference/inference_pipeline/

[AdelShousha]

Thank you for suggesting inference!, I have tried it and the previous error is resovled but I have a new problem. My code works normally for a while then the terminal yields "Killed" stopping the process, when running "sudo dmesg | grep -i 'killed'" I get :
[695416.057171] Out of memory: Killed process 1711614 (pt_main_thread) total-vm:23405672kB, anon-rss:6538112kB, file-rss:68640kB, shmem-rss:12160kB, UID:1000 pgtables:19644kB oom_score_adj:0

I'm using a custom inference logic like the one provided here :https://inference.roboflow.com/using_inference/inference_pipeline/#how-to-provide-a-custom-inference-logic-to-inferencepipeline
and I'm using a custom sink that uses two other yolo models on the video frame coming from my original yolo making it a total of three models running at the same time.

I don't know if the way I'm using three models is causing the problem or if something else causing this problem. Can someone help?

[SkalskiP]

@AdelShousha, let me ask someone from the Inference team.

Btw, could you be a bit more specific here? My code works normally for a while How long does it take for the pipeline to crash? How much RAM do you have on that machine?

[grzegorz-roboflow]

Hi @AdelShousha , this is Grzegorz here, I'm part of inference team.

It seems OOM kicks in and kills your script. Since inference-pipeline is buffering frames it is quite possible that your RAM is maxed out, can you try export VIDEO_SOURCE_BUFFER_SIZE=2 when running the script and see if the script completes without abnormal termination?