Training in Progress

What happens while your model trains, and how to read its metrics

As the model begins training, EyePop.ai’s training system first augments the dataset. This automatically applies transformations such as noise, rotation, blur, color, and position transformation to simulate a larger dataset. After augmentation, the model begins training in earnest, and the console shows a progress bar while it runs.

Model metrics

Two measures describe how a detection model is doing, and they trade off against each other: Precision and Recall.

Precision answers the question: did the model find only what I wanted it to find and nothing else? We want to know the model isn’t predicting false positives.

Recall answers the question: did the model find the object in each scene regardless of whether extra objects were identified? We want to know the model is picking up all instances of the object, even if there are extra boxes.

You can expect a training duration of approximately 10-20 minutes for 500 images, depending on the complexity. Training runs server-side and continues if you navigate away; come back once it has finished running.