In this document, we will explore the most important graphs seen on tensorboard and how to judge if a model is good or not
When monitoring our training using tensorboard we care mainly for three criterias:
- The training loss
- The evaluation loss
- The evaluation mAP
This function labeled as loss_1 represents the loss according to your training data.
It is decreasing function, which means that its value shoud be as small as possible.
These functions labeled as loss represent the loss according to your test data.
The total loss function summarizes everything. It is also a decreasing function.
Caution: If your training loss has much lower values than your evaluation loss that means that your model is overfitting
These functions labeled as DetectionBoxes_Precision represent the mean average precision or (mAP). It is a value between 0 and 1. The closer it is to 1, the better.
The graph DetectionBoxes_Precision/mAP, summarizes the mAP measure.