some question for loss

[lvwuyunlifan]

hi, @jeong-tae :
I follow your model to train. The number of pre-apn trainings is 2000, and it seems strange to see the loss of the cls and the loss of the rank, as shown.

Also accuracy has been very low, no more than 1. What is the reason for this?

[jeong-tae]

this is because APN predict wrong location. I didn’t find how to train APN correctly yet.
There are some missing details...and I think nobody knows about that. If you can find pretrained APN or code/detail param, let me know.

[lvwuyunlifan]

Oh, thank you. If I know the code/details about APN I will tell you.