I am training a detectnet model, and wanted to get a bit deeper understanding behind the loss score. My loss score at epoch 119 is 0.00040; is there any intuition you can share with the ideal loss score for a well generalized model for detectnet. I am asking, because every model on TLT describes loss differently. Should, I run for higher number of epochs like 300 or 120 is fine. My dataset currently contains around 1365 images.
Epoch 119/120: loss: 0.00040