the EAO results is not so correct

STVIR / pysot

SenseTime Research platform for single object tracking, implementing algorithms like SiamRPN and SiamMask.

Apache License 2.0

4.44k stars 1.11k forks source link

the EAO results is not so correct #515

Open haishibei opened 3 years ago

haishibei commented 3 years ago

"NOTE we not use gmm to generate low, high, peak value" ------in pysot/toolkit/evaluation/eao_benchmark.py line 19.

The result evaluated by the tookit is not so correct, because the values of "low"、“high "、"peak" is fixed. It should be auto computed by KDE(GMM), because different testing result is corresponding to differente values of "low"、“high " and "peak". I will appreciate it if someone can update the compute code.

Leong1230 commented 1 year ago

Hi @haishibei

Do you know how to determine the values of "low", "high" and "peak" for a specified dataset? Could you please provide a demo about this? And how are these values in eao_benchmark.py line 19 determined?

Thank you very much！