Skip to content

Domain adaptation


When a neural network model has been trained on a set of images representing the realizations of the input data distribution, it performs well on test data from the same distribution. However, in a large number of applications, the domain from the training data and the domain from the production data differ for many reasons (too little data in the training database, modification of the process, acquisition conditions, …). It is therefore important to develop training techniques capable of reducing this phenomenon.

Leave a Reply