Model training method, road structure identification method, and apparatus

Abstract

Provided in the present application are a model training method, a road structure identification method, and an apparatus, which are used for performing model training by combining manually labeled data with unlabeled data, thereby implementing model training in a distillation manner at a lower cost. The model training comprises: acquiring a training set, wherein the training set comprises first training data and second training data; and subsequently, inputting a sample from the training set into a neural network, so as to obtain a first output value and a second output value, wherein the neural network comprises a first network and a second network, the first output value is an output of the first network, the second output value is an output of the second network, and the first network is updated by using only the first training data. When the sample is a sample from the first training data, the first output value is used to update the first network, and the second output value is used to update the second network; and when the sample is a sample from the second training data, the first output value and the second output value are used to update the second network.

Images