train and val on mot16

[iamZe]

Hello sir, nice work!

may I know what I need to do if I need to do training on mot16? Thanks.

[Newdxz]

hello，Did you run this program correctly，I encountered some problems in the evaluate, can I ask you something, Thanks

[guillembraso]

Dear @iamZe,

Thanks for your kind comment! MOT16 contains the same sequences as MOT17, but MOT17 has improved annotations. As for the public detections, MOT17's DPM detections are exactly the same ones given as MOT16 public detections. Hence, using MOT17 is preferable to using MOT16.

If you still want to train on MOT16, you'd have to download the annotations from this link, and extract them in DATA_PATH. Then you could essentially reuse the code in src/mot_neural_solver/data/seq_processing/MOTCha_loader.py used for the MOT17 dataset to load MOT16 annotations / detection files. Hope this was clear.

If you need further help please let me know!

[guillembraso]

Hi @Newdxz . What problem did you encounter? there was a typo in the configuration file path in scripts/evaluate.py. It was written configs/tracking_config.yaml instead of configs/tracking_cfg.yaml. But I have now fixed it. If you were facing a different problem, then please open a separate issue for it. Thanks!!

[iamZe]

@Newdxz sure, lets figure out problems together

[guillembraso]

Hi!
If I'm not mistaken, @Newdxz problems were already addressed in #22. Am I right?
@iamZe do you still need help with MOT16, or can I close this issue?

[iamZe]

yes please @gbraso