INITIAL NEURAL NETWORK TRAINING METHOD AND APPARATUS, IMAGE RECOGNITION METHOD AND APPARATUS, DEVICE, AND MEDIUM

摘要

Disclosed in embodiments of the present disclosure are an initial neural network training method and apparatus, an image recognition method and apparatus, a device, and a storage medium. An initial neural network comprises multiple network layers and at least one enhancement layer, the enhancement layer being used for enhancing a received feature map and inputting the enhanced feature map into the next level of network layer of the enhancement layer. The method comprises: obtaining an original image sample; inputting the original image sample into an initial neural network to obtain a prediction resu obtaining an actual result corresponding to the original image sample; determining a loss value according to a loss function, the prediction result, and the actual result, and training parameters of the neural network on the basis of the loss value; when a training condition is satisfied, removing the enhancement layer from the current neural network to obtain a trained neural network. The occupancy rate of samples on memory space can be reduced, and the generalization performance of the neural network can be improved.