Training Dashboard

The Training Dashboard gives an overview of the entire training process. On this screen we can monitor the training in real time, as well as evaluate the history of previous training and edit various parameters.

Indicators

In these tables are the indicators of the selected training. The date that was performed, duration, and final results of the training. We also have a button bar with actions such as Downloadand* Publication * of the model, visualization of parameters and visualization of test images.

Final training results

Sample expansion test image - Data Augmentation

End of training test image

Graphics

The graphs give an idea of the training progression, either in real time or in history.

LossandVal Loss indicate the error measured in the training, and are common indicators for all training. They must be descending curves that approach 0. The lower the Loss, the less the neural network is missing.

There are also graphs of neural network efficacy, and for these, the bigger the better.

• mAPforObjectDetection datasets. It is generally considered good above 0.6.
• AccuracyforClassification datasets. It is considered good above 90%.

Training History

It is also possible to check the previous trainings and thus have insights on whether the actions that were taken in the construction of the dataset and parameterization of the training had positive or negative effects.

Where should I go now?

• Dataset
• Training