Validation set ratio

[Yashay]

Hi,

Hope you are well.

In your DOAM article on page 7 you said:

"We further select the best performance model to calculate the AP of
each category to observe the performance improvement in different
categories. "

Does this mean after every epoch you evaluate on the testset and select the best performing model?
Or do you complete training and then evaluate on the testset once?

I ask this because I want to find a stopping criteria for training, usually monitoring validation-set loss does this.
Because using the testset as a validation-set we can overfit the testset.

Kind Regards
Yashay

[LoveIsAGame]

Can you share with me the download address of opixray dataset？thanks！！