search

riips / AIC24_Track1_YACHIYO_RIIPS

MIT License
10 stars 1 forks source link

Errors when replicating your results #1

Open iflab-felipe opened 3 months ago

iflab-felipe commented 3 months ago

Hi, first of all, thank you for your great work !

I was trying to replicate the results on Scene 1. I followed your setup steps and I sampled the dataset so I got only 1003 images (1:5 sampling rate of 5000 total frames, the rest was eliminated because I wanted to quickly reproduce the results). I got an error (please see terminal.txt), where I concluded that there are not GlobalOfflineIds assigned during the SCPT tracking. This later produces an error because feature_stack (line 361, mcpt.py) is None. Also I noticed that there are a lot of OfflineID assigned inside the .json of the cameras (also enclosed), which makes me think that the scpt is not doing well the reid process. Could you please confirm me if this is normal, and what could be going wrong here?

Thank you so much.

terminal.txt camera001_tracking_results.json fixed_camera001_tracking_results.json representative_nodes_scene1.json

riips commented 3 months ago

Our team has already tested several cases (contains our original data) and this kind of error did not occur. We think mcpt process is cause of error. This feature_stack (line 361, mcpt.py) is made by node of hierarchical clustering. If feature_stack is None, clusters may not be made by mcpt process.

Based on this idea, please check the following two points.

  1. Change the parameter keypoint_th=1 to keypoint_th=2. →We assume that all tracklets are filtered as noise when using keypoint_th=1.
  2. Check the key named "GlobalOfflineID" in the file whole_tracking_results.json. →If mcpt is done correctly, you will see several numbers.

We will investigate the cause of the error and modify the program later. Thank you for your attention to this issue.