Closed carbonox-infernox closed 6 years ago
I think I understand this now. The batch size is the number of images per batch, so if there are 400k images then there would be 20k batches. I think the model is using a reduced number of images per epoch, though.
@carbonox-infernox correct, batch size is number of images in the batch :)
In Neptune.yaml, I have left the default parameters alone. One portion reads:
Training schedule
epochs_nr: 100 batch_size_train: 20 batch_size_inference: 20
However, I am currently training the model and am on epoch 0, batch 275. My understanding is that this should only go up to 20 batches (or maybe 40, if it's doing train and inference in one go) per epoch. What's going on here?