search

open-mmlab / mmtracking

OpenMMLab Video Perception Toolbox. It supports Video Object Detection (VID), Multiple Object Tracking (MOT), Single Object Tracking (SOT), Video Instance Segmentation (VIS) with a unified framework.
https://mmtracking.readthedocs.io/en/latest/
Apache License 2.0
3.55k stars 595 forks source link

I can't get a good performance by ByteTrack #445

Open 376498485 opened 2 years ago

376498485 commented 2 years ago

I use the following command:

python demo/demo_mot_vis.py \
    configs/mot/bytetrack/bytetrack_yolox_x_crowdhuman_mot17-private-half.py \
    --input demo/demo.mp4 \
    --output mot.mp4 \

And the output video 'mot.mp4' is as follows image How can I do to get a good performance by using ByteTrack.

JingweiZhang12 commented 2 years ago

Did you change any hyper-parameters? Without any changes, I have just tested the demo code and got a good tracking performance by using the pretrained ByteTrack model.

376498485 commented 2 years ago

I didn't change anything. I just set up the environment and then run the demo.

------------------ 原始邮件 ------------------ 发件人: "Jingwei @.>; 发送时间: 2022年3月2日(星期三) 下午2:58 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [open-mmlab/mmtracking] I can't get a good performance by ByteTrack (Issue #445)

Did you change some hyper-parameters? Without any changes, I have just tested the demo code and got a good tracking performance by using the pretrained ByteTrack model.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

JingweiZhang12 commented 2 years ago

You can download the latest code and run the demo again.

376498485 commented 2 years ago

OK, thanks for your reply. I will try again.

------------------ 原始邮件 ------------------ 发件人: "Jingwei @.>; 发送时间: 2022年3月2日(星期三) 下午3:02 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [open-mmlab/mmtracking] I can't get a good performance by ByteTrack (Issue #445)

You can download the latest code and run the demo again.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

376498485 commented 2 years ago

I have download the latest code, run the demo again without any hyper-parameters changing and get the following result: image I will appreciate it if you can help me to solve the following questions:

  1. There are not only pedestrians, but also cars. So how can I do to only detect pedestrians.
  2. It didn't perform as expected, there are many undetected pedestrians. So Is it related to the version of mmcv or mmdet?
JingweiZhang12 commented 2 years ago

You should load the pretrained checkpoint (downloaded before) of ByteTrack. Here is an example of a command: python demo/demo_mot_vis.py configs/mot/bytetrack/bytetrack_yolox_x_crowdhuman_mot17-private-half.py --input demo/demo.mp4 --output mot.mp4 --checkpoint checkpoints/MOT/bytetrack_yolox_x_crowdhuman_mot17-private-half_20211218_205500-1985c9f0.pth

376498485 commented 2 years ago

Thank you! It works well!

------------------ 原始邮件 ------------------ 发件人: "Jingwei @.>; 发送时间: 2022年3月3日(星期四) 上午9:36 收件人: @.>; 抄送: @.>; "State @.>; 主题: Re: [open-mmlab/mmtracking] I can't get a good performance by ByteTrack (Issue #445)

You should load the pretrained checkpoint (downloaded before) of ByteTrack. Here is an example of command: python demo/demo_mot_vis.py \ configs/mot/bytetrack/bytetrack_yolox_x_crowdhuman_mot17-private-half.py \ --input demo/demo.mp4 \ --output mot.mp4 \ --checkpoint checkpoints/MOT/bytetrack_yolox_x_crowdhuman_mot17-private-half_20211218_205500-1985c9f0.pth

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you modified the open/close state.Message ID: @.***>

AndrewGuo0930 commented 2 years ago

Excuse me. Where can I download the checkpoint?

376498485 commented 2 years ago

https://download.openmmlab.com/mmtracking/mot/bytetrack/bytetrack_yolox_x/bytetrack_yolox_x_crowdhuman_mot17-private-half_20211218_205500-1985c9f0.pth