hi,i use my own dataset train the model with RFRNet_Smaller_Hole. my dataset has 150k train imgs and i train 180k iters with 2gpu, batch size = 2*6 =12 in the first trianing phase, i also finetune it with 60k iters.
whatever train imgs or val imgs, the result all like this above. the result is the same as random weights.
the g_loss can converge to 0.015 from 1.xx in the begining. can you give some suggestions, thx in advance~
hi,i use my own dataset train the model with RFRNet_Smaller_Hole. my dataset has 150k train imgs and i train 180k iters with 2gpu, batch size = 2*6 =12 in the first trianing phase, i also finetune it with 60k iters. whatever train imgs or val imgs, the result all like this above. the result is the same as random weights. the g_loss can converge to 0.015 from 1.xx in the begining. can you give some suggestions, thx in advance~