Hi, I have a segmentation model to segment two datasets, Camvid and Cityscapes. I do not know how long should I train the model. So, I set the model to be trained for 400 epochs and when the training was finished by looking at learning curves I found that the loss is still decreasing and the Dice is improving. So, can I load the trained model and let it to be trained for longer epochs, instead of training it from scratch?
I think so, it’s even one of the essences of saving models if I’m not mistaken.
If your 400 epochs only take a few minutes (I’m not familiar with model segmentation), you can start again without any problem. But when you have to start training models for days, it becomes more than necessary.
Ok, thank you for your answer.