Does adversarial training require very few epochs?

wasidennis / AdaptSegNet

Learning to Adapt Structured Output Space for Semantic Segmentation, CVPR 2018 (spotlight)

850 stars 203 forks source link

Does adversarial training require very few epochs? #110

Open soans1994 opened 3 weeks ago

soans1994 commented 3 weeks ago

Dear authors,

The iterations you have used in the code is 150000 and the gta 5 dataset has 24966 samples. So with the batch size of 1 it will be 150000/ 24966 = 6 epochs. Is this enough to learn the source dataset behavior. Is it more dependent on the adversarial learning? I am confused about the number of iterations.

Thank you