hfslyc / AdvSemiSeg

Adversarial Learning for Semi-supervised Semantic Segmentation, BMVC 2018
504 stars 129 forks source link

all losses are NaN #61

Open aoaoaoaoa opened 3 years ago

aoaoaoaoa commented 3 years ago

why all my losses in experiment are NaN? I used the same code, parameters and dataset. like this "iter = 19998/20000, loss_seg = nan, loss_adv_p = nan, loss_D = nan, loss_semi = nan, loss_semi_adv = nan semi ratio: 1.0000"