Closed sparrow2015 closed 2 years ago
Hi @cooper87 ,
We did not train for a fixed number of epochs. Instead, we trained it until the loss stagnated after decreasing the learning rate twice. In practical terms, with our training set of > 70K cubes, the loss function required ~30 epochs to get small enough and another 50 or 60 epochs to get close to the best value.
thanks very much
hi, I have train it for learning, but after I train 100 epochs, the loss just decrease a bit. how many epochs when you training?