Question about loss_ins,loss_cate and overall loss
sparshgarg23 opened this issue · comments
I am currently training SOLO on colab(with one GPU!),and I noticed that the logs gives me information about loss_cate,loss_ins and overall loss.
I am not sure what is loss_cate here refers to Is it the combined loss (Dice loss+Focal loss)
Also currently when training on 1 GPU in colab,the estimated time for training is 2 days,is there any way to speed up the training?
I read the paper,cate here refers to focal loss and ins is the instance segmentation loss,However,I would still like to know how to speed up the training process.Increasing the batch size from 4 to 6 leads to cuda out of memory error.