search

longcw / MOTDT

Real-time Multiple People Tracking with Deeply Learned Candidate Selection and Person Re-identification
MIT License
533 stars 130 forks source link

Bugs to be fixed #12

Open civilpat opened 5 years ago

civilpat commented 5 years ago

Hi @longcw , thanks for the open source code. I have run the evaluation and got similar results that you have posted in the end. However, there are few bugs and solutions that I want to share with you.

(1) at line 124 within "eval_mot.py", it should be Evaluator.save_summary(summary, os.path.join(result_root, 'summary_{exp_name}.xlsx')), without "f" in front of 'summary_{exp_name}.xlsx'

(2) it worth noting that, python3 should be used instead of phython2 for your code. Otherwise, a bug listed below will show up. While there is no problem with python3 by using "from importlib import util" File "/usr/local/lib/python2.7/dist-packages/motmetrics/lap.py", line 149, in init_standard_solvers from importlib import util ImportError: cannot import name util

Thanks again for your work! @longcw

longcw commented 5 years ago

Thanks for your response. Actually, f-strings is another kind of str.format() that added in python3.6 https://cito.github.io/blog/f-strings/

So this repo requires python>=3.6 and I didn't test it with other python versions.

civilpat commented 5 years ago

@longcw Great! Thanks for sharing the info! Yep, no wonder that I used python 3.5. I bet if had 3.6, it would not be a problem for me. BTW, do you plan to open source the training code for MOTDT? It would be interesting if I can train weights using my datasets on your MOTDT! :)

longcw commented 5 years ago

Hi, thanks for your interests. I am working on some projects these days and have no plan to share the training code for now. But I will try it a few weeks later if I have enough time.

civilpat notifications@github.com 于2018年11月25日周日 下午12:22写道：

@longcw https://github.com/longcw Great! Thanks for sharing the info! Yep, no wonder that I used python 3.5. I bet if had 3.6, it would not be a problem for me. BTW, do you plan to open source the training code for MOTDT? It would be interesting if I can train weights using my datasets on your MOTDT! :)

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/longcw/MOTDT/issues/12#issuecomment-441414314, or mute the thread https://github.com/notifications/unsubscribe-auth/AF6UgB8KH2Jn2LJQkKP0qT_hZKn2diMSks5uyhr-gaJpZM4YxsEK .