Retraining a neural network

Retraining a neural network 

The "Gambit" eliminates false workings. 

The quality of any neural network is determined by the material (dataset) on which it is trained. We conduct regular courses on data collection methodology; you can sign up by emailing Virtual@goal.ru.

You can understand that your neural network is triggering on something else besides what it was trained for, when you launch it into operation. For example, instead of a pistol, the neural network triggers on a mobile phone. But what to do to get rid of this?

With the help of the Gambit program, you can retrain any neural network in any of the ONNX formats. To do this, simply transfer the false positives to a different category.

The "SegmentMaster" module for cleaning the neural network from false earnings costs 250 euros.