About tracking performance on Posetrack dataset.

[nguyenquivinhquang]

Thanks for your great work, I have been diligently working on implementing your code and conducting validations, specifically with regard to tracking performance on the Posetrack dataset using the evaluation code provided in evaluate_PHALP.py,

During my evaluation process, I encountered some unexpected results. The IDF1, MOTA, and HOTA scores I obtained were lower than the values reported in your paper. However, it's worth noting that the ID switch metric showed improvement. Here are the results I obtained:

          IDF1    IDP    IDR   Rcll   Prcn   GT   MT  PT  ML    FP   FN IDs   FM     MOTA  MOTP IDt IDa IDm

OVERALL  71.9%  61.8%  86.1%  88.7%  63.7% 1374 1112 122 140 22682 5053 326  146    37.5% 0.342 290 174 146
ID switches   326
MOTA          0.37510299521211443
AssA          0.6308350892268777
DetA          0.4474141820195844
HOTA          0.5191260712874806

To ensure the accuracy of my evaluation and to better understand the discrepancy in results, I kindly request that you share the ground truth data used for validation or provide detailed steps for reproducing and evaluating the results as presented in your paper. Your assistance in this matter would be greatly appreciated.

Thank you very much for your time and consideration.

[brjathu]

could are you evaluating this on PHLAP (HMR 1.0) or 4D humans (HMR 2.0)?