visionml / pytracking

Visual tracking library based on PyTorch.
GNU General Public License v3.0
3.19k stars 603 forks source link

Hyper-parameter tuning #396

Open goutamyg opened 1 year ago

goutamyg commented 1 year ago

Hi!

In the tracker codes you have published, there is a section on tracker-related choices/hyperparameters https://github.com/visionml/pytracking/tree/master/pytracking/parameter

It has hyperparameter info related to training, advanced localization, IoUNet parameters, and some tracker-specific ones. I read the papers and couldn't find details on how these parameters were determined. Can you please suggest what is your approach to tune these hyperparameter values? Do you fine-tune them using the test-set/s itself?