search

abnerwang / py-Vital

Vital tracker implemented using PyTorch
37 stars 9 forks source link

Modify the loss function #4

Open wysot opened 5 years ago

wysot commented 5 years ago

Excuse me, if I want to modify the loss function, should the loss function of the training phase and the loss function of the tracking phase be changed? Or just change the loss function of the training phase? Thank you

abnerwang commented 5 years ago

I think all the loss functions should be changed.

Excuse me, if I want to modify the loss function, should the loss function of the training phase and the loss function of the tracking phase be changed? Or just change the loss function of the training phase? Thank you

I think all the loss functions should be changed.

wysot commented 5 years ago

Thank you very much.

wysot commented 5 years ago

Thank you very much for your work. the result on OTB2015 is 0.6708, but the result on the paper is 0.68. Can your result reach 0.68? Thank you in advance.

abnerwang commented 5 years ago

Thank you very much for your work. the result on OTB2015 is 0.6708, but the result on the paper is 0.68. Can your result reach 0.68? Thank you in advance.

You need to adjust the hyperparameters according to your needs.

wysot commented 5 years ago

The article has repeatedly mentioned that they only use G in the training phase, and only use D in the testing phase. However, the code did not find the use of G during the training phase. Instead, it used G in the test phase. How do you correspond to this part?

abnerwang commented 5 years ago

The article has repeatedly mentioned that they only use G in the training phase, and only use D in the testing phase. However, the code did not find the use of G during the training phase. Instead, it used G in the test phase. How do you correspond to this part?

You didn't understand the paper and didn't understand the code. You can read it carefully several times.