search

a1600012888 / MUTR3D

Code for Paper, MUTR3D: A Multi-camera Tracking Framework via 3D-to-2D Queries. https://tsinghua-mars-lab.github.io/mutr3d/
MIT License
179 stars 19 forks source link

Can't reproduce the result. #14

Closed feifei-Liu closed 1 year ago

feifei-Liu commented 2 years ago

Hi,

Thank you very much for your excellent work. I tried to reproduce your results, but found that MOTA was always 0.

fa31ad082611c85dd914569c123e693

(1) When use your provided json files, it can achieve your results.

(2) When use your provided weights of r101, the MOTA is 0.

(3) When I retrain a model, the MOTA also is 0.

I checked the result json file, and it showed that the result was normal (not NAN), but the track id between the two frames did not correspond.

c2fa4af2d009a41efd2817d19572cdb

So I would like to ask how to solve this problem. Thank you very much!

a1600012888 commented 2 years ago

Hi, Can you share the scripts or config file you are using for (2) and (3)? Seems wired, the recall is non-trival but tracking related metrics is zero.

a1600012888 commented 1 year ago

Hi, I think this mismatch of tracking ids might be caused by this: https://github.com/a1600012888/MUTR3D/issues/15 Should already fixed. Thanks.

ghost commented 1 year ago

Hi,

Thank you very much for your excellent work. I tried to reproduce your results, but found that MOTA was always 0. fa31ad082611c85dd914569c123e693

(1) When use your provided json files, it can achieve your results.

(2) When use your provided weights of r101, the MOTA is 0.

(3) When I retrain a model, the MOTA also is 0.

I checked the result json file, and it showed that the result was normal (not NAN), but the track id between the two frames did not correspond. c2fa4af2d009a41efd2817d19572cdb So I would like to ask how to solve this problem. Thank you very much!

@feifei-Liu Hi, I met the same problem with you, how did you solve it later ?